# Full Text: An Active Inference Ontology for Decentralized Science: from Situated Sensemaking to the Epistemic Commons

> Extracted from `2022_ActiveInferenceOntology.pdf`

> 8 figures extracted to `images/`

---

## Page 1

An Active Inference Ontology for Decentralized Science:
from Situated Sensemaking to the Epistemic Commons
Authors (ORCID):
Daniel Friedman 1,2 (0000-0001-6232-9096)
Shaun Applegate-Swanson 1,2 (0000-0001-9474-6038)
Jessica Angeli Balbuena1,2 (0000-0003-2741-4476)
Arhan Choudhury 1 (0000-0001-6897-2947)
RJ Cordes 1,2 (0000-0002-9913-7159)
Shady El Damaty 3 (0000-0002-2318-4477)
Avel Guénin—Carlut 1,4 (0000-0001-8239-7264)
V. Bleu Knight 1,2 (0000-0002-9894-1989)
Ivan Metelkin 1 (0000-0003-2899-6690)
Siddhant Shrivastava 1,2,5 (0000-0002-9688-4730)
Amit Kumar Singh 1 (0000-0002-8870-9676)
Jakub Smékal 1,3 (0000-0003-4989-4968)
Caleb Tuttle 3 (0000-0002-8469-2221)
Alexander Vyatkin 1 (0000-0003-1306-4620)
Author Affiliations:
1. Active Inference Institute (activeinference.org/)
2. Complexity Weekend (complexityweekend.com/)
3. OpSci (opsci.io/)
4. Kairos Research (kairos-research.org/)
5. Kernel (kernel.community/)
Email address for correspondence & collaboration: ActiveInference@gmail.com
Abstract
In this work, we examine science from the vantage points of blockchain technology
and its connection to decentralized science (DeSci). We consider science as a collective
process using Active Inference, an integrative framework that models the cognitive
processes of perception, planning, and action selection in terms of Bayesian probabilities
and updating. We present the Active Entity Ontology for Science (AEOS, available at
coda.io/@active-inference-institute/active-entity-ontology-for-science-aeos) as a
composable and versionable system for modeling various science systems, using the
Active Inference entity partitioning. Such DeSci systems are considered from the
perspective of BOLTS (Business, Operations, Legal, Technical, Social). Further steps for
developing and utilizing AEOS in the context of scientific ecosystems are provided.
1

## Page 2

Table of Contents
Abstract
1
Table of Contents
2
Driving Questions
3
Introduction
4
Science and the Epistemic Commons
4
Centralized Science (CeSci) and Decentralized Science (DeSci)
7
CeSci: centralizing institutions against situated sensemaking
8
DeSci: governing epistemic commons
10
Implementing Decentralized Science in the Web3 era
13
Web3 for DeSci
13
Web3
13
Blockchain
14
Smart Contract
14
Token (fungible and non-fungible)
15
DeSci Organization: DAOs, Blockchain, and Markets
15
DeSci Vision and Open Questions
17
DeSci Token Engineering
18
DeSci Research
19
DeSci Publishing and Review
20
DeSci Funding and Finances
21
DeSci Data and Knowledge Markets
24
Teamwork Modeling with Systems Engineering
26
BOLTS of the CeSci to DeSci Integration
28
Business
29
Operations
29
Legal
30
Technical
30
Social
30
BOLTS and Reliability for Science Systems
31
Active Inference, Systems Engineering, and Science
33
What is Active Inference?
33
Active Inference and Science
34
Active Entity Ontology for Science (AEOS)
36
AEOS Entity partitioning
38
AEOS Entity Classes and Types
38
AEOS Policy Selection, Areas of Concern, and Roles
40
Discussion
44
Works cited
46
2

## Page 3

Driving Questions
Centralized Science (CeSci): Historical science operations, community, and practices
●
What are successful, effective, and productive scientific practices for communities and individuals?
●
What frameworks and tools can help form and support communities of practicing researchers to
promote long-term collaboration and impact?
●
How are preferences and expectations aligned and communicated in ecosystems of scientific
collaboration?
●
How do we avoid perverse incentives in systems of scientific inquiry?
●
How do we ensure effective and transparent resource allocation and funding as research becomes
more interdisciplinary, interorganizational, and international?
●
How can scientific careers be started and nurtured, increasing the accessibility and vitality of
science as a community and body of knowledge?
●
What are fair and effective means of elevating voices to scientific thought leadership roles?
●
What are good ways to identify strategic priorities for scientific funding, intellectual capital
allocation, and focused development?
●
How to deal with new opportunities and challenges related to increases in remote work?
●
How to enable efficient data and code access?
Decentralized Science (DeSci): The future of coordination and Knowledge Engineering
●
How can decentralized teams and organizations coordinate research using effective distributed
mechanisms?
●
What scientific outputs are to be expected of Decentralized Science (DeSci)?
●
How will DeSci produce traditional research products such as papers, as well as other outcomes,
such as products, services, platforms, tools, and protocols?
●
What artifacts and design patterns will stimulate the development of integrated DeSci systems?
●
How can Distributed Autonomous Organizations (DAOs) scaffold and catalyze research, grants,
education, and community development?
●
What methods for value capture and financialization improve the process of scientific discovery,
and which methods result in poor outcomes that are at odds with open science community values?
●
What improves the reliability of outcomes and accountability of funds in emergent collaborations?
●
What are the relations between the trajectories of scientific governance, and
social/ecological/political change and collapse?
●
How can scientific careers become more accessible to researchers who are non-PhD holders, and
those with other commitments such as family, participation in industry, etc?
●
How can we use blockchain and related technologies to cultivate cultures of fair work attribution,
through records of contribution to knowledge artifacts and proof-based protocols?
●
How can we maintain a favorable signal-to-noise ratio for DeSci work output?
●
Can the Active Inference framework and ontology be used to understand complex sociotechnical
systems such as epistemic communities?
●
How are epistemic commons created and maintained in an era of Human-Computer collaboration?
3

## Page 4

Introduction
Science is a cumulative, collective endeavor that augments a single human's
sensemaking ability with tools, practices, and processes we use together to observe,
measure, and understand the world around us and to agree on a shared reality. Even
before the invention of modern scientific instruments such as microscopes and particle
accelerators, humans relied on their senses to observe the environment and on their
brains to make inferences about these observations. This is still true today. Any study of
how knowledge is accumulated must consider how the brain works when making
predictions based on observations and incomplete data. Additionally, as the information
required for scientific research extends far beyond the cognitive capacity of a single
unaided human, tools for scientific organization, knowledge management, and
communication become increasingly necessary to consider as well.
Science and the Epistemic Commons
A recent shift in science is the move towards greater recognition of the complex
nature of systems such as brains or societies [1,2]. For example, cognitive science has
largely moved away from the ambition to discover universal laws of behavior to focus more
on the causal structure of the brain and its patterns of activity in different scenarios [3].
Utilitarian scientific models today strive to explain natural phenomena by piecewise
integration of generative models expected to match the causal structure of the target
system at the relevant scale of behavior [4,5]. The development and validation of
mechanistic models is based on multiple cognitive strategies, for example involving
computational modeling, empirical observation, statistical inference, and domain-specific
theoretical considerations. Despite this modeling shift, the social structure of science has
not yet adapted to embrace integrative practices. While the scientific community has
long-standing practices which they optimize for, many of these practices have lost their
validity, efficacy, or legitimacy [6,7]. This results in suboptimal integration of complex
transdisciplinary knowledge by favoring the partitioning of research along the lines of
traditional domain-specific methodologies, organizational processes, thematics, and
vocabulary.
Indeed, the modern era has seen the emergence of an unprecedented hegemony
of administrative institutions over all other forms of human social organization. Political
anthropologist James C. Scott explains the emergence of top-down structures for complex
problem solving as a default reflex to compress complexity. Scott claims that the capacity
of bureaucratic states and administrators to understand and manipulate the world relies on
a systematic reduction of the world onto standardized, legible tokens (teachable facts),
4

## Page 5

such as accounting, and measurement units or legal identities based on a permanent
patronym [8]. The need for this "view from above" has placed pressure on research
institutions to adapt both their objectives and operating activities to meet the needs of
administrators. This "technoscientific" logic has left a complex and pervasive legacy in the
organization of contemporary research, including in the way we currently understand and
practice scientific research. In worst-case settings (which are unfortunately common),
research quality is evaluated on superficial measures with no principled consideration of
deeper relevance, leading to a raging competition between researchers on the number of
words written, sheer number of publications of any kind, or the ability to use technical
language [9–12], without considering the positive epistemic impact of the work. Entire
research programs can be funded because of insider networks, or correlation to a hype
cycle, while more thoughtful and critical (and ultimately constructive) research may easily
go unnoticed. In other words, the key metrics for tracking global scientific progress have
become increasingly divorced from their ability to be replicated or have meaningful impact
[13,14], and more correlated with social and cultural signaling within the unrepresentative
and often biased scientific community. When the professional scientific community is only
a small unrepresentative fraction of society, there is the possibility of detachment or
isolation from broader goals. Modern scientific careers can be exclusive and select for
certain types of people, leading to less representative population practicing science
professionally with key perspectives and backgrounds missing from the table
Given the incentive misalignment and power imbalances between administrators
and practitioners, the scarcity of resources, competition over resources both between
researchers and research institutions, and the inflexible nature of the bureaucracies which
connect and govern extant research institutions, there are few mechanisms available for
improving traditional research institutions aside from increasing funding. Luckily, human
prosociality has taken many forms throughout our evolutionary history from which we can
draw inspiration. We could develop tools and protocols which treat the outcomes of
research as contributions to an epistemic commons, a web of informational or
knowledge-oriented systems which would allow individuals with disparate motivations, skill
sets, and beliefs to collaborate at the intersection of their interests and tackle specific
questions or systems with the most contextually relevant tools, in the best interest of
cumulative, collective scientific knowledge. In other words, we argue that relaxing the
constraints of legibility associated with bureaucracy and its “view from above” could help
develop the "view from the ground" (i.e., pragmatic understanding by practitioners) as well
as the "view from within" (i.e., scientific knowledge, in the classical sense of an objective
description of system organization and activity).
Human prosociality has contributed to the persistence of our species. The emergent
structure of social organization is variable but can develop towards hierarchical, top-down
5

## Page 6

control which are transactional and mutually beneficial [15,16]. However, hierarchical
control can lead to the sequestration of wealth and power over time, and a centralized
motivation that does not always act in broader society's best interest or satisfy all relevant
perspectives. Hierarchical structures develop cognitive biases over time that limit the
ability to parse multiple, conflicting streams of information that do not fit into the standard
decision making templates of those on top. Decentralized structures may be more capable
of acting on divergent information streams, however decentralization alone is not a
mechanism that provides intelligent coordination for effective collective action [17,18].
Decentralized Autonomous Organizations (DAOs) are an emerging form of coordination
between humans that enables new kinds of cyber-physical communities mediated by
immutable, cryptography-protected code [19,20] and other artifacts. DAOs allow for
programmable digital social structures, embedded mechanisms for consensus,
intermediation of trust-requiring actions, and independent operation of workstreams with
multiple leaders instead of a top-down hierarchy. Coordinating such decentralized
communities of practice, or more generally governing "complex commons", is novel,
difficult, and high-stakes work, often without the scaffolds and norms that traditional offline
governance benefits from [21]. For example, systems that are decentralized in principle
may still be functionally centralized in terms of power or asset distribution [22].
We propose that epistemic drives and norms can grow organically from
communities of practice, defined by open questions, shared value, interoperable
processes, methodological synergies, personal affinities, diverse contributions,
serendipitous encounters, and Ontologies, Narratives, Formal documents, and Tools
(ONFT, [23]). In this paper, we model the similarities and differences between
Decentralized Science (DeSci) and Centralized Science (CeSci) through the Active
Inference framework, understood here as a framework addressing how living systems
create an understanding of their environment [24] and act based on this understanding.
The Active Inference formalism affords an integrated account of the multiscale dynamics
shaping research, from the constitution of epistemic affordances shaping scientific
practices [25] to the coevolution of epistemic beliefs and communities [26], as a process of
uncertainty reduction (technically described by the descent of a gradient of variational free
energy [27,28]). This model implies that cognitive understanding derives from adaptive
control in agentive engagement rather than explicit representation [29–31], and that the
activity of epistemic communities is shaped by the sociocultural constraints defining their
structure rather than individual states of mind [32]. Based on these considerations, we will
articulate a conceptual framing and Active Entity Ontology for Science (AEOS).
Before we dive deep into head-on contrastive definitions of CeSci and DeSci, a
primer on decentralized systems is crucial. We would like the first definition not to be
simply an anti-definition of its counterpart, as we aim to incentivize effective research and
6

## Page 7

set up collaborative relations between emergent DeSci and incumbent CeSci, not simply
to ignite controversy. Both CeSci and DeSci will exist in collaboration in the future, in the
sense that scientific systems will vary in their extent of centralization. Therefore, even
though here we will use the terminology of decentralization, a host of related terms may
also be relevant [22,33–35]. One interesting perspective to apply here is that of “Self
Certifying Systems” [36]. These self-certifying systems are those in which an entity
validates its own existence in the system, and the incentive mechanism of the system is
intrinsic to it. For instance, a dataset in a database system such as InterPlanetary File
System (IPFS) [37,38] validates its own legitimacy by hashing the contents using a
cryptographic hash function. In these dynamics, the root of trust lies within the network
entity itself, and not necessarily through legitimation by sources outside of the system.
Centralized Science (CeSci) and Decentralized Science (DeSci)
Decentralized Science (DeSci) is a relatively new term, introduced by Web3
collectives to describe the use of recent digital tools for funding, training, planning,
coordination, execution, dissemination and archival of scientific activities and assets by
digitally connected communities. DeSci is an emerging area with multiple, potentially even
incompatible, forming senses, so all explorations here should be considered preliminary
and partial. The term DeSci suggests, by contrast, the existence of Centralized Science
(CeSci), which would stand for the continued status quo organization of science as a
highly bureaucratic activity managed by select academic and private institutions. Some
motifs or patterns that are found in CeSci and DeSci can be found in Table 1 and Figures
6-8.
DeSci can be defined as the use of Web3 technologies to introduce epistemic
markets through the deployment of open source and financial tools. In this view, DeSci
represents the introduction of commodity markets for scientific assets and services where
such markets were previously impossible (e.g. financialization through tokenization of
intellectual property, scientific platform governance, peer review or curation services, or
access privileges to data or infrastructure). In another perspective, DeSci is seen as a set
of mechanisms for bottom-up individual sensemaking. From this perspective, DeSci is the
capability of individual agents or communities of practice to make sense of the world
autonomously, by defining their own questions, language and methodology. These
disparate views represent the distinction (without preference or judgment) between DeSci
as a set of emerging Web3 tools and DeSci as a research ecology, respectively.
Tool use, including markets and organizational technologies like DAOs, are
fundamental to the ecological view of science [39]. However, proponents of DeSci do not
conceive of this distinction as a purely technological one, for example simply describing
7

## Page 8

the fact that DeSci organizes with the Internet while CeSci utilizes paper as well as digital
media is an insufficient distinction. For example DAOs don't replace off-chain or offline
human organization, they augment it with on-chain guarantees and some automatic
processes for human organization (i.e. voting is still a human process, but once the
parameters are set the voting itself may be automated and free of human corruption). The
broader perspective on DeSci is that it aims to reform the organization of scientific activity
itself, which is expected and preferred to translate into an increased ability for Science to
fulfill its mandate and align with social values [40]. We fully agree that this expectation is
warranted, at least in principle, as the socio-epistemic dynamics we call Science clearly
derive, at some level, from the concrete organization producing it. We will therefore
attempt to characterize the distinction between CeSci and DeSci in terms of their
respective organizational and structural aspects.
Both CeSci and DeSci aim at knowledge discovery, cultivation, and curation, but
they differ in their means to this end in terms of their incentives, structure, and norms, or
more generally, in terms of their research ecology. In CeSci, the language, methods for
discovery, and strategic priorities are imposed from the top down by a core group of
decision makers acting out the mandate of government policy positions. Core groups of
decision makers also receive external input and constraints, for example in the case of
peer review. In DeSci, methods and norms emerge from the bottom-up interactions of a
web of loosely connected communities of practices with diverse coordination and
prioritization mechanisms. Therefore, CeSci and DeSci diverge significantly in terms of
their sensemaking; one imposes specific meaning, understanding, and other cognitive
constraints to scientific activity so as to keep it legible to the "view from above", the other
cultivates the "view from the ground" by allowing those same cognitive objects to unfold
spontaneously throughout the course of the scientific activity.
Though there is fundamental overlap, and perhaps more similarity than not between
CeSci and DeSci at this incipient stage, those terms stand for a fundamental divergence in
the social and epistemic dynamics shaping the organization of research. Here we explore
some of those dynamics by considering key differences between DeSci and CeSci.
CeSci: Reducing Sense-Making Complexity Through Centralizing Institutions
CeSci is characterized by the presence of centralizing agents (such as government
institutions, philanthropic foundations, private businesses, or universities) having the
power and willingness to establish and enforce rules, goals, ontologies, and other informal
constraints within the wider scientific ecosystem. Its purpose can be understood as
pragmatic rather than purely epistemic, in that CeSci institutions are funded and
motivated, internally and externally, to build epistemic value (i.e. knowledge) that can
8

## Page 9

assist other institutional actors and stakeholders understand and manipulate reality in
terms of their objectives, for better or worse. An example of this is the 19th Century
development of statistics, which was explicitly intended to help the developing modern
States to perceive and control their population through the collection and analysis of
demographic data, and effectively shaped what it meant to "think like a State" [32]. Due to
extrinsic mandates, CeSci orients the sensemaking activity of its participant researchers
toward specifical technical and pragmatic purposes.
Although the goal of scientific research has largely extended beyond war and
administration, the modern logic of centralization and technical management is largely
continued in contemporary research. The managerial role over the epistemic commons
was granted to modern academic institutions, and complemented their role of supporting
technological development through research on the property of chemicals, mechanical
systems, and the like. These institutions have a core mandate of assisting outside
institutions in their pragmatic endeavors. Their overarching goals are to accumulate
epistemic value in that direction. Influence over research institutions by external
organizations can be direct or indirect. In the first case, it may mean cadres of
administrators exercising direct control over what research objectives should be, what
funding agencies can be or should be approached, and recruiting based on compliance
with external mandates. Indirect influence can be exerted, for example, by external
organizations offering funding based on use of specific language or alignment with
pragmatic research outcomes.
Accordingly, the production of knowledge artifacts, such as journal articles, is
currently enacted according to a hierarchical structure of funding agencies, scientists,
institutions, and publishers. There are in-groups and out-groups, as well as unspoken
rules to obtain funding for and publish results of scientific literature. Moreover, there are
agendas driving all agencies that fund scientific research, which are not necessarily
transparent. Many basic research questions are framed in a way that adheres to these
agendas. For example, researchers trying to characterize a mechanism by which a brain
protein folds will need to emphasize its supposed role in curing Alzheimer's disease, when
in fact any actionable outcome may be unclear or distant from the research [41]. This
situation provides a strong incentive for researchers to leverage the information advantage
they have over administrators to grab their attention by making incredibly grandiose,
possibly dishonest claims on the meaning of their research. Administrators simply lack
time or incentive to check on such claims, and active researchers lack the power to modify
the incentive structure and would risk their job if they refused to follow them.
One aspect of the difficulty in communicating between administrators and active
researchers is related to a discrepancy between the relevant time scales and perspectives
9

## Page 10

for their respective activities. Bureaucratic funding agencies (private or governmental)
have specific, long-standing research directives that cannot be modified easily, both
because of long-term commitments and because of the limited time officials have to
understand, evaluate, and integrate feedback. For example, an academic administrator
will tend to lack domain expertise in the details of protein folding, and to lack motivation to
discuss how category theory or information geometry could lead to major scientific
breakthroughs. Administrators are motivated to discuss how the research they manage
contributes to the competitiveness and prestige of the institution they answer to, and why it
is therefore worthy of more funding. They may encourage their employees working on
protein folding to claim in bad faith that their research will solve cancer, to publish a lot of
papers into highly competitive (although not necessarily domain-relevant) journals, or to
collaborate with companies for patents. If, for example, experts on protein folding want to
express the benefits of future conceptual advancements, the importance may not be
conveyed among fellow researchers and administrators due to a lack of common
language, and therefore an inability to make decisions in a framework of mutual
understanding.
Because of this fundamental asymmetry in power, information, and incentives, the
centralizing hierarchical structure of science interferes with the rapid adaptation and
evolution of scientific ideas and the exploration required for this. The need for researchers
to make their ideas and objectives legible from the detached perspective of administrators
conflicts with their ability to develop a context-relevant ontology on the ground and to focus
on actually understanding the target system in front of them, or aiming to understand
phenomena beyond the scope of expected research narratives or outcomes. Centralizing
institutions need to build largely artificial semantic and incentive silos, both as a way to
compete against rival institutions and to measure and control the activity of researchers.
Centralizing institutions organize science around outcomes which are legible and
manageable, generating a specific type of situated and transdisciplinary sensemaking with
certain strengths and weakness.
DeSci: Governing and Utilizing Epistemic Commons
In contrast to the centralizing institutional view presented above, research can be
organized as a decentralized federation of individuals and communities of practice, each
striving to address specific issues in a contextually-relevant way. Such communities, by
definition, lack a direct dependence on any centralizing institutions, or at the very least
such institutions lack the ability to step in and manage their activity. That these
communities do not need to adapt their language or objectives to the expectations of
administrators, does not make their activity intrinsically scientific in nature. However,
decentralization offers specific opportunities to scientific communities by allowing them to
10

## Page 11

focus on developing pragmatic solutions to concrete problems and/or purely advancing the
epistemic understanding of the natural world they are focused on, free of centralizing
institutional bias. As decentralized communities of practice lack the coercive power which
defines centralizing institutions, their continued existence relies entirely on the voluntary
participation of their members, and therefore necessitates an institutional structure
facilitating and incentivizing participation. Consequently, and also due to the lack of
semantic silos (see below), decentralized scientific communities will have a strong
incentive to cooperate with each other, to make relevant issues discoverable to wider
sections of the population, to allow participation purely out of voluntary interest, and to
develop unifying languages easing their integration.
The opportunity for knowledge appropriation (via legal enclosure [42] through
intellectual property institutions, as well as siloing via disciplinary boundaries) relates the
development of scientific knowledge to a broader, well-studied category of collective action
problems: the governance of common goods [43]. Common goods (commons) are defined
as resources that are non-excludable (one cannot be forbidden from accessing them). The
potential for misuse or pollution of the commons consequently necessitates governance at
the scale of communities of stakeholders, and regulation needs to follow certain conditions
that were initially formalized in economic sociology [11,12] and recently generalized in
evolutionary theory [44–46].
Roughly, the implementation of governance systems for commons entails:
1. the unambiguous definition of a problem to solve;
2. the construction of a community of stakeholders willing to help solve this problem;
3. the institution of a decision system allowing participation on an equal footing by all
members of the community;
4. agreement on a set of rules (which are interactive and mutable with validity checks
over time), which fairly reward active cooperation within the community;
5. the iteration of (1)-(4) at higher scales where the community as a whole is
confronted with a situation requiring governance.
Thus far, the distinction we have made between Centralized and Decentralized
Science is largely organizational. For example, early European academia emerged from
the clerical legitimation of scholarly "guilds," professional associations similar to
contemporary trade unions or cartels, which effectively restricted who could practice a
certain craft or access a specific kind of knowledge within a certain area (generally a city).
Thus, patterns of sensemaking were immune from any outside oversight, and were
imposed by the community on each and every one of its individual members. Centralizing
institutions can constrain community members with respect to important aspects of their
11

## Page 12

intellectual autonomy, and enacted autonomous agency (see Thinking like a State [32] for
a discussion of institutional autonomy). In contrast, decentralization facilitates broad
access to the epistemic commons, for example as in the case of the printing press. When
we highlight the differences between CeSci and DeSci, we aim to frame a discussion on
how specific kinds of organization can facilitate differences in the integration of information
about the world over multiple dynamical scales, and what it means for the broader
scientific community.
Today, digital technologies provide clear opportunities for proponents of DeSci.
They facilitate the widespread diffusion of information with a relatively low entry bar, and
therefore facilitate the voluntary cooperation of multiple communities of practice engaged
in situated sensemaking. This encourages the decentralized adoption of standardized
methodologies and languages, transparency over code and data management, and
willingness to produce information of clear value to outsiders. We see the success of such
a decentralized yet unifying project in Wikipedia, which is beyond any doubt the most
important example of a largely standardized, accessible, and transparent knowledge
system, and is based entirely on voluntary and decentralized cooperation. At the moment,
there is however no appropriate, analogous system to use in the organization of scientific
activity, as an alternative (or complement) to the currently dominant role of centralizing
institutions.
12

## Page 13

Implementing Decentralized Science in the Web3 Era
Contemporary digital tools have the potential to establish a distributed and
transparent scientific community with open participation, where every participant has root
level access to all knowledge and the rules for its production. Online communities have
seized the opportunity to develop and utilize such digital tools for funding, training,
planning, coordinating, executing, disseminating, and archiving scientific assets and
activities, therefore establishing the blueprint of a Web3 DeSci. There is no defining
characteristic of the whole set of Web3 DeSci tools. Nonetheless, some features
characterize a larger subset of these tools and are worth highlighting.
First, Web3 tools allow for more interactions (including economic transactions)
between economically- and geographically-diverse groups. Second, Web3 tools can
increase the transparency of products, activities, and interactions that are themselves
digital, or that can be represented digitally. Third, Web3 tools allow for the introduction of
novel markets that, in some pockets, might increase decentralization but in others, might
increase centralization. Instead of using general terms to elaborate on each of these
aspects, we hope to demonstrate their reality and significance as we overview more
specific features of and developments toward a Web3 DeSci implementation.
Web3 Terminology
Before heading into some details about the implementation of DeSci (which may
just be one iteration or genre of the infinite game of Science), the reader should be
equipped with familiarity around some basic Web3 terms. Note: the following terms are
just current technological means to building a desired framework, and are presented here
as introductory, not as comprehensive reviews or final answers.
●
Web3
●
Blockchain
●
Smart contract
●
Token (fungible and nonfungible)
Web3
First, "Web3" is an imprecise term referring to an ethos that values digital
self-sovereignty and, more important for our purposes, recent developments in
peer-to-peer protocols that service this ethos. Part of the Web3 ethos is to value open
source software and the personal control of one's data and assets [47]. These values have
13

## Page 14

influenced and continue to influence the design and adoption of these protocols. Indeed,
the term "Web 3.0" was introduced by Gavin Wood [48], a key figure in the development of
the Ethereum and Polkadot protocols, who proposed Web3 as a "post-Snowden" Web, a
decentralized Web similar to the Web1 of the early internet and in contrast to Web2 where
a small number of centralized agents control people's data, access to websites, and the
web applications that many use every day [49]. Web3 protocols are diverse, but the ones
at the heart of most functional systems are so-called "blockchain" or distributed ledger
technology (DLT) protocols that support complex smart contracts.
Blockchain
Recent developments in peer-to-peer (i.e. decentralized, distributed) protocols,
particularly blockchain protocols such as ones that facilitate Distributed Autonomous
Organizations (DAOs [20,21,50]), are the biggest enablers of a Web3 DeSci
implementation. Without going into detail, a blockchain protocol enables a network of
self-interested, competing computer nodes to agree on the state and immutable history of
a shared ledger. Anyone with an internet connection and enough funds can append the
ledger or verify its integrity, though it is still possible to implement access control with
smart contracts. Blockchains are a natural fit for building currencies, which historically
emerged as an extension to bookkeeping by institutional actors [51], and currency building
in fact constitutes their most widely known application today. However, the principles of
data integrity and access apply just as well to running any record or application that must
be tamperproof. Blockchains can be used to establish ownership of digital assets (e.g.
tokens, datasets, or intellectual property [52]), mediate between pseudonymous actors
(e.g. during peer review), establish decentralized identities used for reputation (e.g. when
applying for grants), prove an actor has done something (e.g. contributed work, stored a
dataset, or run a compute service), and more. For DeSci, blockchains enable actors to
keep track of scientific activity in permissionless systems, and to build systems of
exchange and accounting around any kind of digital information – which includes much of
the products and tools of modern science.
Smart Contract
Blockchains can establish digital records via programs called smart contracts. A
smart contract (in the context of blockchains) is a computer program whose code is
publicly available yet cannot be changed by anyone, is executed on computers across the
blockchain network, and can be invoked (i.e., started) by anyone [53,54]. When the
blockchain's record is updated by a smart contract, the update is verified by the distributed
nodes on the blockchain network. Despite the presence of "contract" in the term, a smart
contract does not refer to a legal contract, though it is still possible to implement certain
14

## Page 15

legal agreements in smart contracts, such as the triggering of payments in the case of
predetermined on-chain activity. Not all blockchain protocols support the complex smart
contracts that enable the features listed in the previous paragraph.
Token (Fungible and Non-fungible)
One kind of smart contract is called a token. A token is a digital asset. It can be
bought, sold, or transferred on the blockchain. The implementation of a token in code is
little more than a record of which addresses (i.e., users, loosely speaking) own how many
units of the token. When units of a token are exchanged, the record is updated, not unlike
what happens in a bank's database when two of its customers use the bank to transfer
money to each other. Tokens come in various forms, broadly separable into fungible and
non-fungible (though see [55] for a more complete discussion). A token of the latter form is
often referred to as an NFT (Non-Fungible Token), and may be transferable or
non-transferable. The salient difference between these types of token is in the name; there
are many interchangeable units of any given fungible token, and there is only one unit of
any given non-fungible token. Because it is easy to use tokens to record ownership, they
have been used by DAOs to designate voting rights (similar to shares of stock in a
corporation) or provide proof of participation [56,57]. Tokens have many other use cases
(many of which haven't been invented yet) and have even enabled the emergence of a
discipline known as token engineering [36], which studies the use of tokens to incentivize
certain behaviors within a market.
Decentralized Science (DeSci)
The blockchain technology affords certain advantages over alternative systems in
terms of transparency, decentralization, and tamperless regulation. However, its
development was quite explicitly motivated by the implementation of market mechanisms
at the exclusion of other forms of social organization, and therefore focused heavily on the
implementation of monetary transactions. Multiplicative investments in a monetary
economy supports the emergence of an extremely polarized power law distribution in
wealth, and can (even in the absence of strategic interaction or information asymmetry)
lead to the control of a large proportion of the economy by a few individuals [26]. In reality,
strategic interaction and information asymmetry are ubiquitous, and the power entailed by
the wealth of a few aristocrats gives them the capability to build a population of
dependents and influence the rules of the game to their advantage. A complete
disinvestment of public powers or otherwise democratic institutions from economic
redistribution and regulation would therefore expectedly lead to the development of a
15

## Page 16

feudal-like economy, where a few wealthy individuals use their economic leverage to yield
near absolute political power over their fiefdom.
The fundamental vulnerability of marketization-oriented Web3 technologies is
deeply aggravated by preexisting asymmetries in wealth and power. Any token that can be
traded freely (while not pegged to a stable asset) is vulnerable to price instability entailed
by speculative investment, which would make it essentially worthless as a currency (i.e., a
means of holding and exchanging value). On the other hand, any token that is not coupled
in some way to the mainstream economy cannot be used to pay for basic goods such as
food or shelter, and therefore lacks any economic value. Historical currencies were
instituted by States through 3 main leverages: by creating demand through taxation, by
backing their value to some external value holder (generally rare resources such as
precious metals or recognition of debt by institutional actors), and by forcefully crushing
the non-monetary economy and the communal institutions regulating it [51]. None of these
leverages are accessible to decentralizing collectives, at least not without significant State
/ capital backing and an open betrayal of their explicit mandate (see [58] for a general
discussion of value generation and regulation in cryptocurrencies). Although the problem
of monetary value is specific to currencies, marketization entails the ability to create
economic value, or to find a buyer for what you’re trying to sell. For example, if no one is
willing to pay you (in fiat currency, crypto, or some other value) the equivalent of a living
wage for cleaning data, you are not going to make a living cleaning data.
Despite the aforementioned shortcomings, marketization can still play a substantial
role in scientific decentralization. Markets offer an opportunity for large scale economic
integration between peers, and therefore for the basis for an "exodus" from the grasp of
centralizing institutions [59]. For example, tokens could provide an explicit and unfalsifiable
account of traditionally less-visible or rewarded work. However, blockchain-powered
marketization alone cannot provide a sufficient solution for a successful decentralization of
scientific activity. The blockchain only ensures decentralization in the limited sense that it
makes breaking the rules that were written onto the blockchain extremely difficult, without
consideration of the human layer outside of the blockchain. Building DeSci entails a
broader shift toward positive liberty to do science in a decentralized way, i.e. the
empowerment of individuals and collectives to organize scientific activity around
decentralizing values and institutions. This process entails that we pay attention to the
actual dynamics of decentralizing systems, and ensure that individuals and collectives can
effectively enact their formal freedom of collaborative sensemaking.
DAOs, whether enforced by blockchain technology, or more classical means, are
organizations defined by an explicit mandate and specific rules for participation and
decision, and are usually open to anyone willing to accept these constraints. In the extent
16

## Page 17

that these rules are reasonably fair, and that individual participants get a say in the
decision system and overall organization, DAOs constitute a natural fit for the
management of commons.
Of course, none of this excludes the value problem discussed above: science
mandates that scientists be paid a living wage while they focus on their work, or otherwise
to have enough time and resources to be able to conduct scientific work. In consequence,
a radical shift to DeSci would likely necessitate broader social reform, such as the
institution of a right to work in the research domain of one's choice, or perhaps a solution
such as Universal Basic Income. Some cryptocurrencies, like the Ğ1 currency [60], try to
build such change from the ground up by distributing a predefined rate of monetary
creation equally between members, and explicitly grounding the networks to social webs
of trust and exchange. But implementing such changes, especially at a global scale and
within a reasonable time frame, would be entirely beyond the ability of contemporary
DeSci communities. In the remainder of the document, we will therefore focus on steps
that DeSci can practically take right now, many of which are related to tokenization and
marketization mechanisms.
DeSci Vision and Open Questions
In our society today, a relatively small number of individuals have full-time careers
in scientific research. Those with family commitments, a significant role in industry, or an
aversion to academic social systems are largely absent from the process of creating and
curating widely-distributed knowledge artifacts using the scientific method.
Can we support and develop communities that train researchers from many diverse
backgrounds and provide opportunities for scientific collaboration, funding, and publication
outside of the narrow constraints of CeSci career trajectories? How can we motivate and
reward productive scientific engagement in this system, where there are less rigid barriers
for who is allowed to conduct research? Can we benefit from neurodiversity and
large-scale harmonization of different perspectives to reach epistemic consensus states
beyond the reach of the narrow confines of what constitutes a "traditional scientific mind,"
or collection of such minds, in our society today? Decentralized communities of practice
focused on research are already self-organizing, and will begin to answer these questions
over the coming decade.
17

## Page 18

DeSci Token Engineering
Blockchains and tokens can be used to create incentive mechanisms to promote
desired behavior and punish undesired behavior, and these incentive mechanisms can be
used in DeSci to encourage valuable scientific activity and the creation and maintenance
of epistemic commons. Blockchain and other cryptographic technologies offer tools to
observe, design, and control complex system behavior through the means of token
engineering [61]. The practice of developing systems and mechanisms that utilize tokens
is referred to as "token engineering" [20]. There are many ways to use tokens to facilitate
and incentivize behavior. In Filecoin, for example, a storage node is rewarded tokens for
storing data only if it submits successful cryptographic proofs demonstrating it has stored a
unique copy of the data over a predetermined period of time; the node's collateral is
reduced if it fails such proofs. Token-engineered DeSci systems can be created for
research-specific use cases, for example [62]. Token engineering can also be used
specifically to reward researchers, attribute impact fairly, and establish a value flow for a
scientific ecosystem, a topic which we turn to now.
Were we to model entire communities as a collection of interacting agents which
transact, perform work, and make economic decisions (e.g., investment, staking), this
simulation could provide insight into the effectiveness of the overarching incentive
structure, the resilience of the system to various risks, and the capacity of the organization
for interoperability. We leave this as an opportunity for future study.
Token system engineering might ask questions like:
●
How are stakeholders rewarded for meaningful participation?
●
What is the value flow? Where does value start/enter? Where does value go/exit?
●
What is the value distribution, internal and external to the system of interest?
●
How does the system shape patterns of attention?
●
What perverse incentives exist by tokenizing (i.e. tying price of token to speculative
trading)?
●
How do constraints on action influence system behavior, such as: not allowing the
selling of tokens, not allowing the transfer of tokens, making different token classes
transfer together, etc?
One example of token engineering used for research value allocation is based on
the Ocean Data Marketplace [63], in which a community-funded treasury disburses funds
in the form of fungible tokens to research teams based on criteria defined by the
community. At this point, the research team has full autonomy and is responsible for
efficiently distributing the funds across different researchers. Researchers can spend
tokens in a decentralized knowledge market to gain access to datasets, papers, or any
other scientific work published by other researchers. Tokens spent in this marketplace
18

## Page 19

would provide compensation to those who have contributed to the DeSci ecosystem.
Every transaction made in this decentralized knowledge market is facilitated through smart
contracts, which collect transaction fees that are fed back to the community treasury.
At this point, to better grasp the opportunities that Web3 protocols have for
publishing and reviewing in science, we might benefit from a brief look at the protocols
themselves. Affordances for increasing transparency and accessibility are primarily in
decentralized storage networks, such as IPFS and Arweave [64]. IPFS, the InterPlanetary
File System, indexes files with content addressing based on cryptographic hashes. In
practice, this allows two things. First, a file can be stored on any IPFS node in the world
and can still be accessed by anyone who has the file's address. Second, an address can
never correspond to more than one file. These properties mean that any file uploaded to
IPFS is available to anyone with an internet connection (this increases transparency and
accessibility), and that a file cannot be changed without its address also changing (this
increases transparency). Arweave increases transparency and accessibility by providing
permanent and open storage. It does this with a blockchain-inspired data structure called a
blockweave and a proof for the network's consensus mechanism called Proof of Access in
which, to receive a reward, a miner must prove it is storing some randomly determined
previous block of data. There are other storage protocols used to store papers and related
data, but these examples sufficiently illustrate the common benefits of these systems.
The discussion on science token engineering leads to a natural question: what are
the limits of decentralized science? Alternative value flows designed by the practice of
token engineering assume new laws are enforced by digital contracts deployed on a
transparent blockchain. As such, their ability to enforce particular actions outside the
digital world become very limited. In its current form, DeSci is constrained by the
affordances available to digital entities; hence the scope of scientific outputs is also
constrained to research that has digital elements.
DeSci Research
Currently, most active DeSci projects primarily focus on building the infrastructure
necessary to allow the dynamics described above, while only a fraction use their
resources on research (as of 2022 a non-exhaustive list of the DAOs and groups
mentioned is available here, building on [65]). Examples of research-focused DAOs
include VitaDAO and Molecule which, at the time of writing, have funded active projects in
longevity research. Other active groups include the Governance Research Institute,
Governauts, and Metagov DAO, researching governance, smart contracts, and other
aspects of coordination in the Web3 space, all of which are closely related to the
development of the DeSci infrastructure. As such, there is no clear distinction between
19

## Page 20

infrastructure research and development at this time. Furthermore, many DeSci DAOs
have a clear vision of the type of research they want to conduct, but are currently in the
process of getting funding or curating an initial community of active contributors. In DAOs
as in other communities of practice, groups organize around specific problems or
questions. Groups define rules for their participation, and coordinate with each other to
define and solve meta-problems - such as the development of standard tools or a common
language or unifying framework which facilitates progress within each community.
There are many open questions relating to creating and sustaining impactful
communities of practice in a Web3 DeSci setting. Technological changes to the process of
science in DeSci may be exciting and impactful, but cannot circumvent the fundamental
challenge of carrying out high-quality research. Even with hypothetically-successful
systems in place for proposing projects, knowledge management, financial support, and
decentralized publishing, individual researchers must still be motivated, committed, and
engaged to produce useful work. Creating and sustaining impactful DeSci communities is
challenged by the scale of information available for review or meta-analysis projects, the
size of the design space of possible experiments, the multiple individual ways of working,
and the difficulty in producing quality research. Without long-term mentorship, training, and
support for scientists at all career stages, it is difficult to imagine that merely increasing the
extent or type of system decentralization would result in a successful scientific ecosystem.
DeSci Publishing and Review
DeSci seeks to use Web3 tools to improve the transparency and incentives within
academic publishing and to improve the accessibility of published products. Efforts in
DeSci around publishing are, in most cases, responding to the occasionally-misaligned
incentives within current academic publishing, such as the lack of incentive for reviewers
to provide unbiased and quality reviews, the incentive for journals to publish
revenue-generating papers (even at the expense of rejecting epistemically-profitable, but
less monetarily-profitable, research), the disincentive to share negative results, etc. [66].
With respect to transparency and incentives, Web3 DeSci projects are experimenting with
using tokens (i.e. cryptocurrencies), markets, reputation metrics, and forum-like platforms.
Increasing accessibility involves using decentralized storage networks to store published
products (e.g. papers, reviews, datasets, code). This section hints at the direction of a
DeSci approach to publishing and peer review with a discussion of current DeSci projects
dealing with academic publishing and how blockchain-based technologies could potentially
fit into the picture.
20

## Page 21

Since 2017, a number of authors have proposed blockchain-based systems for
peer review [39,67–72]. Various works such as [39,71], [72] and [70] propose the use of
blockchains to make the peer review process transparent and incentivize good behavior
through reputation points recorded on the blockchain, though they differ in details, with
[39,71] and [70] focusing on the entire pipeline even including publication. Below, we
discuss several of these systems.
Avital [68] focuses on the motivation for a blockchain-based peer review market,
where authors passively or actively seek reviewers, and all participants interact through
smart contracts on blockchains. Ants-Review builds on this idea of a blockchain-based
peer review market [67] and outlines an implementation that uses Ethereum [73] and the
InterPlanetary File System (IPFS) [37,38]. In Ants-Review [67], an issuer (i.e. author)
creates an AntReview for a paper. The AntReview specifies the paper's issuers, file
address (on IPFS), requirements, deadline, and approver. The approver is responsible for
approving and paying reviewers (using the funds of the AntReview) for their reviews. A
reviewer submits a review by uploading it to IPFS and sending the review's IPFS address
to the AntReview for the approver's assessment. Anyone can add funds to an AntReview.
These funds can be distributed only by the approver and can be sent only to reviewers. If
the funds aren’t spent before the review's deadline, those who contributed funds are
refunded. Note that, in AntsReview, all parties are pseudonymous, associated only with
their Ethereum addresses, and that the use of IPFS and Ethereum means everything is
open access.
With Reddit-like features for curating, discussing, and incentivizing research,
ResearchHub [70] is an even further departure from the current peer review and
publishing system. Its users can upload, comment on, summarize, upvote, and downvote
papers and posts. Users are rewarded RSC, an Ethereum-based token, for these
activities. Papers and posts are aggregated within "Hubs" which function like journals or
subreddits. Allowing ideas to circulate in the open, ResearchHub shuns the traditional peer
review model of waiting until a paper is mostly or entirely finished before having a small
number of reviewers comment on it. ResearchHub relies on the community and the
incentives created by the RSC token to create a system in which the better a post or paper
is, the more it is seen and its author(s) rewarded. ResearchHub also plans to create a
feature allowing researchers to raise funds from others in RSC. The platform differs from
other projects in the DeSci space in that it does not currently rely much on peer-to-peer,
Web3 protocols, such as Ethereum and IPFS.
21

## Page 22

DeSci Funding and Finances
A more traditional method of funding a DeSci research project is through grant
applications. In this case, the process starts with a research investigator submitting a
proposal for a particular research endeavor, which is reviewed by the individuals or
funding agencies who allocate the awards based on their own criteria determining merit. In
the case of DeSci and DAO-based organizations, funding can be provided to the DAO
within their internally-defined structures and further distributed within the DAO in order to
fulfill the initiative of the grant. Alternatively, stakeholders or investors may provide funding
due to their initially genuine research interest or other incentives, and ultimately become
detached from the original research motive, thus allowing the DAO to distribute resources
to fulfill emergent agendas.
One alternative approach for funding DeSci is the creation and maintenance of
markets for projects (bounties) or the outcomes of projects such as data or intellectual
property. When an organization/individual is interested in conducting research based on
interests, a systematically-managed decentralized exchange for task performance can
play a pivotal role. The marketplace could have research proposals by organizations,
individuals, and researchers as well. A base level requirement for any market-supported
project would be to produce some monetary or other form of value investors agree on. Key
to supporting a synergetic researcher-investor dynamic would be finding an alignment
across multiple aspects of the project such as focus, constraints, cost, and timelines. The
machine learning platform Kaggle implements methods similar to this, where only
researchers that perform best according to the project-defined metrics are rewarded.
Once a project has begun, it can benefit from mutual researcher-investor direction
and redirection. Regulation of such direction is imperative; DeSci could build on top of
existing blockchain infrastructure with a few modifications. For instance, a structure known
as a DAICO (DAO+ICO [74]) proposes to lock all the proceeds of an ICO (Initial Coin
Offering) into a decentralized autonomous organization (DAO) smart contract and put the
governance over that DAO in the hands of the investors. This locking needs to be
consensual from both ends and needs to be recalibrated as projects change. A roadmap
with the specific, measurable, achievable, relevant and time-bound targets can be agreed
upon; the investor gets the required output and the researchers get a well-defined source
of funding. Other than direct funding, people interested in the project can involve
themselves actively with participation, or more passively by forming a support network  or
keeping a watchful eye on the project. An example of this passive or implicit contribution
type could be building a community with social outlets such as newsletters and citation
value networks, or using online measures of stigmergic attention [75]. These communities
can contribute collectively in terms of contributions to reduce the reliance on initial major
investors and even foster the project in the case that investors pull out, hence
22

## Page 23

decentralizing the investor influence. One particular project in this regard, Science Fund
[76] (Figure 1), allows donors to contribute to science funding pools opened for a specific
topic with an amount of their choosing. In return, donors receive a NFT receipt (a "Science
Funding Token", or SFT).
Figure 1. Dynamic research funding model of Science Fund [76].
These mechanics allow for the following affordances:
●
Curated Funding Pools: Each donation is added to a chosen funding pool.
Researchers can apply through an open protocol with their track record from
already-published groundwork. As new results emerge from the funded work, the
NFT’s metadata is updated.
●
Impact Tracing: Donors have a much more significant participation and visibility
into the projects they helped fund. Their SFTs are a personal key to unlock a
continuously-evolving experience data trail about their donation’s impact.
●
Future Participation: The NFT receipts become the key to future interactions with
the scientists they supported, with evolving forms of participatory governance in the
field(s) they contributed to.
●
Funding allocation: When researchers apply to one of the topic-specific funding
pools, they are not coerced to reveal all their invaluable insight upfront. They simply
need to share their most influential scientific work (any already-published scientific
artifact they created, such as a publication, preprint, or dataset) to a given pool.
Selected scientists' works then form the basis of a growing evidence trail that is
matched to the donor's minted NFT. The NFT receipts become the key to future
interactions among scientists and donors, and open a universe of endless new
possibilities of mutual exchange and collaboration.
23

![page23_img1.png](images/page23_img1.png)

## Page 24

DeSci Data and Knowledge Markets
Web3 tools enable markets for data storage and compute services that could be
used by DeSci advocates to develop a "knowledge market” or decentralized data
commons [77]. Agents in this potential market exchange data and other services such as
data analysis and model training. Note that, because of their digital nature, such markets
welcome opportunities to integrate human and machine agents, allowing for a significant
increase in automation. While the base layer here is a data marketplace, it is plausible that
future developments could emerge layered on top of this, due to the incentives and
informational integrity provided by the market. This section covers the base layer of a data
marketplace and how it encourages certain beneficial developments. In particular, data
marketplaces have the potential to facilitate data archival, interoperable data standards,
data discoverability, and increased automation, and they provide an easier way for industry
to build on and fund work in academia.
The base layer of a DeSci knowledge market is first a data market. Multiple
blockchain-based data marketplaces have emerged [78–80] since the Ethereum mainnet
launch in 2015 [73]. Blockchains provide a number of benefits over traditional data
markets, principally the ability to create datatokens, a construct pioneered by Ocean [78].
A data token is, just like any other blockchain token, a digital asset. However, unlike
regular tokens, datatokens can be used to access data services. Consuming a data
service involves either being sent the associated dataset or having the data provider run a
compute job on the dataset. This datatoken construct is significant because it allows data
services to be purchased on decentralized (blockchain-based) markets. Anyone can
publish a dataset as a type of token, and anyone with sufficient funds can purchase one
such token to consume the data service. The datatoken construct is even more valuable
within the context of another mechanism developed by Ocean, namely, one that
incentivizes the curation of quality data: Users can lock-up funds to signal support for a
particular token (called staking) and earn a portion of the transaction fees associated with
purchases of the token. Rational actors will seek out and stake on high quality datasets to
receive a higher portion of the total fees, thereby curating quality data.
While this base layer is only a curated data market, it indirectly incentivizes the
development of additional components relevant to a DeSci knowledge market, such as an
increase in archival of scientific data. With a marketplace, data collectors have the
opportunity to be financially rewarded and otherwise recognized for simply storing and
selling their data. With the transparency of blockchain-based markets, it would be easy to
keep track of how many times a researcher's or lab's datasets have been requested,
allowing for the emergence of reputational metrics associated with data. It has been
24

## Page 25

observed that currently, "there is little incentive to invest time in archiving or repackaging
data sets" because "investing time in a project beyond its usefulness for publication is
counterproductive, given the high expectations for producing research publications" [81].
Filling this incentive gap, a data marketplace could provide another avenue of being
rewarded and recognized for the valuable contribution of collecting quality data, and could
therefore lead to increased data archival and availability.
Interoperable data standards are incentivized by a data market. The benefits of
interoperable data standards can be seen in, for example, the neuroscience community,
with its adoption of BIDS, "a standard for organizing and describing MRI datasets" [82].
This standard allows neuroscientists to more effectively collaborate within and across labs.
It has reduced the amount of time they spend reorganizing datasets, leaving them more
time for downstream scientific work such as data analysis. In a data marketplace, where
data suppliers are in competition and where the data users prefer certain formats over
others, the data suppliers can be relied upon to provide data in those preferred formats,
given sufficient competition. Even if a dataset is originally published in an esoteric data
structure, someone can profit by purchasing it, organizing it into a known format, and
reselling it. In a market where data purchasers may buy data from varied sources to
improve their models, they are likely to seek out datasets that are compatible with each
other. That is, they will seek out data structured in interoperable formats. The result is a
data marketplace in which suppliers are actively organizing data to suit the needs of
purchasers, converging on interoperable data formats.
Another potential emergent benefit of the data marketplace is increased data
discoverability. Currently, finding scientific data is cumbersome [83]. Researchers often
rely on social connections and fragmented data repositories. A data marketplace helps to
address this issue through the competition between data suppliers who can gain from
increasing the discoverability of their data. The more discoverable a dataset is, the more
likely it is to be purchased. Data suppliers can increase discoverability by, for example,
adding relevant tags and other metadata. Tags can help data search engines sort
datasets. Metadata might even include a cryptographic signature from the researcher(s)
who collected the data, attesting to the source and quality of the data. Additional
organization and discoverability of datasets might come in the form of a knowledge graph,
in which datasets can be related to each other, to researchers, to papers, to code that has
been used to analyze it, etc. Note again that the data marketplace can be read and
operated on by both humans and machines; much of the sorting of datasets in the
marketplace might be done by automatons created by data suppliers. In the end, the
suppliers within a data marketplace have a strong incentive to make their datasets easy to
find, so they will likely adopt whatever methods best accomplish this.
25

## Page 26

Moreover, a data marketplace where data are curated, highly available, structured
in known and interoperable formats, meaningfully-tagged, and discoverable, provides
good conditions for an increase in automated scientific discovery and, more specifically,
better conditions for meeting the Nobel Turing Challenge introduced by Hiroaki Kitano [84]
and further refined by The Alan Turing Institute [85]. This challenge involves creating "AI
systems capable of making Nobel-quality scientific discoveries highly autonomously at a
level comparable, and possibly superior, to the best human scientists by 2050" [85].
Smaller challenges within this grand challenge include ensuring data are shared, linked,
and machine-readable. The data marketplace described above could help address these
and possibly other challenges, such as simply financially-incentivizing researchers to
make their valuable datasets available to all scientists.
Finally, a data marketplace affords profitable interactions between academic labs
and organizations in industry. Labs can sell data to companies that have AI expertise but
may  not have the expertise or equipment needed to collect training data. This not only
allows companies to more easily find and use scientific data, but by buying such data,
companies would also be funding the scientists who collected it. This opens up a market in
which data with both epistemic and pragmatic value can be easily priced and discovered.
It also makes it easier for companies and the economy at large to communicate their
practical data needs to researchers; this could enhance the current system in which a
study's perceived pragmatic value is only a function of how well its value is communicated
through a grant proposal. A data marketplace could lead to a greater flow of information
among academic and non-academic (e.g. citizen, government, non-profit, profit-oriented)
organizations, with respect to the value of various kinds of data, allowing easier and
potentially better funding of data with pragmatic value (though these benefits of funding
may not extend to data which has only epistemic value). A well-designed data marketplace
enables fair price discovery of digital knowledge assets and encourages productive
applications of meticulously-collected, high-quality scientific datasets outside of basic
research, boosting research applicability and downstream innovation.
Teamwork Modeling with Systems Engineering
To decrease the complexity and uncertainty that individuals face collaborating within
large-scale DeSci projects, it is useful to introduce the systems engineering approach
[23,86]. For better communication and team attention management, it is important to have
consensus on the team objectives, project lifecycle management, practices in use, and
metrics for tracking the performance and relevant behavior of all team members. Targets
and goals of the projects are set by the needs of external stakeholders and individuals
fulfilling specific project roles. Governance for decisions, assignments, and resources are
based on unequivocal agreement according to the team members’ affiliation and authority
26

## Page 27

to manage each other's labor. We consider such a group of individuals as an organization
or organizational unit in the case of specific tasks to resolve.
Akin to the use of systems engineering in the building of complex infrastructure
(e.g. bridges, planes), we propose that people involved in DeSci projects document the
roles involved in these team dynamics, as well as their concerns and preferences. This
allows DeSci activity to be structured according to an established lifecycle with needed
practices [87,88]. Here, we are not concerned with the domain-specific practices of a
project lifecycle, but with general practices that are relevant for DeSci development, such
as social modeling, economics, governance, and development in a broad sense. During
any DeSci project, we can associate a human or other entities as actors who are fulfilling
the roles that need to be performed. These entities fulfilling project roles engage in
practices that entail completing tasks or cases. All the relevant objects and open cases
must be resolved during the project for it to be a success (for instance, knowledge artifact
production or fundraising). These cases need to be documented with reference to the
roles, processes, practices, and actions required for successfully achieving the target goal.
This case/issue tracker functionality is an attention management tool for the people
involved in the project. This process allows for productive, auditable, and coherent work on
projects.
Like CeSci, DeSci faces challenges and failure modes across multiple domains,
some of which were briefly described above. At this incipient and formative stage, what
may provide utility for DeSci would be a useful unifying framework to allow for system
conceptualization, design, and implementation. In the following sections, we describe a
possible starting point for the integration of DeSci and CeSci systems, through the
combined application of Active Inference and systems engineering [23].
27

## Page 28

Active Inference, Systems Engineering, and Science
In this section we justify the use of Active Inference in systems modeling, introduce
an entity description for science modeling consistent with Active Inference and Systems
Engineering models, and revisit patterns in CeSci and DeSci.
What is Active Inference?
Active Inference is a framework for the integrated modeling of entity perception,
cognition, action, and impact in the environment. Active Inference draws upon the
formalism of the free energy principle, which provides a single statistical imperative
underpinning the assembly of organic and inorganic matter: the minimization of
informational free energy [89–91]. This principle applies to systems at every scale. The
process of free energy minimization has been positioned as the driver of autopoietic
processes in living systems ranging from cells to cities [92,93]. Its core meaning is that it
allows a goal-oriented, agentic account of self-organization across scales [90]. The
minimisation of free energy at the system's scale entails the minimisation of variational
free energy for each of its constituting particles, which can be interpreted as the
optimisation of a Bayesian model describing the particle's structure (more specifically, the
difference between model Complexity and Accuracy) [94]. Because of this
correspondence, we can claim that entities aggregate in virtue of reducing their uncertainty
about future sensory and active states [95,96].
The related framework of Active Inference leverages this formalism to model one or
multiple interacting entities as involved in continual perception and action angled towards
reducing the expected uncertainty about their perceptual flow and action selection [97,98].
They can do so either by altering internal states so as to infer the causes of sensory
states, or by altering external states so as to cause expected states. Both processes
participate in a common control architecture based on predictive processing, which can be
interpreted as a hierarchy of "algorithms" matching input to output over nested time scales
[99]. In this hierarchical model, previously-acquired beliefs about controlling behavior in
different contexts are internally encoded, and result in adaptive arbitration between
epistemic and pragmatic value choices (i.e. exploration vs exploitation, [100]). Thus, the
framework of Active Inference allows us to model the observations, beliefs, and
predictions (i.e. expected states) of agents as statistical dependencies, or Bayesian
probability distributions.
28

## Page 29

Much of the focus in the Active Inference literature has been on organisms,
especially in the context of human behavioral neuroscience [101]. However, Active
Inference has also been applied to various other informational and cyberphysical settings
such as human communication [102], remote teams [23], human conflict [103], trust
interactions with robotics [104], and social media discourse [26]. Here we apply Active
Inference and Systems Engineering to the case of modeling centralized and decentralized
science.
Active Inference and Science
Evidence of collective intelligence, understood as the capability of human societies
to solve problems, surrounds us. In most cases, it is instantiated in cultural artifacts or
cognitive gadgets [105] that follow from a process of cumulative cultural evolution dating
from the branching of the Homo genus [106,107]. We are surrounded by living fossils of
this cumulative process, from smartphones to hospitals, from net fishing to general
agreements about the recognition of private property. Our capability to understand and
navigate the world does not simply follow from information in our minds, but it is (to some
extent) entailed by the very structure of the ecological niche and the social constraints that
shape our behavior. This means that entities are, to some extent, autonomous or
displaying agency regarding their own cognition [32]. Entities engaged in Science as a
cognitive endeavor are no exception.
Like other instances of collective intelligence involving stigmergy (collective
behavior mediated through modification of a shared niche) [108,109], Science is a
cumulative endeavor that relies on previously-acquired beliefs and established
communities of practice. Scientists and other cognitive entities perceive stigmergic cues
from the niche (including epistemic resources, as well as other social entities), update their
cognitive models (learning), and act to modify the niche in the form of research outputs
(Figure 2). As scientists follow new leads or choose to ignore them, communities coalesce
and disperse, and scientific theories are created, developed, and maintained or forgotten.
Frequently, scientific endeavors are undertaken with multiple, potentially competing, goals
in mind. For example, scientists may work with an aim to achieve some or all of the
following outcomes: earn a paycheck, improve their reputation, improve the reputation of
their company or institution, conduct empirical research without bias, solve a particular
problem, or discover something novel, simple, or beautiful.
29

## Page 30

Figure 2. Two active entities interacting via their shared epistemic niche. For more
detail on the entity perception-cognition-action loop in Active Inference, and the Markov
Blanket formalism comprising the partition shown here, see [90].
Active Inference is relevant for model-based understanding of scientific systems
because it offers a composable multiscale ontology and an integrative framework for entity
and system behavior [93,110]. Thus the usage of Active Inference can help us understand
the entanglement of multiple processes occurring at nested scales of behavior during
scientific research. In Active Inference, actions across scales are modeled as a systematic
attempt by cognitive entities (i.e. scientists and institutions) to reduce uncertainty about
aligning their future perceptions with their preferences, so as to ensure their persistence of
their defining organization. For example, the development of administrative systems and
agriculture was described within Active Inference formalism as a product of City-States’
attempt to understand and control their reality [32], analogously to the process of
morphogenesis as a process of cells manifesting a body plan which reduces their
uncertainty about their cellular niche [95].
Importantly, the Active Inference account is not limited to conservative
(homeostatic) behavior, but also to more complex phenomena such as memory,
anticipation, abductive logic, and autopoiesis. For example, Active Inference has been
used to model the systematic alteration of one’s niche [99,111,112], as well as the creation
of new tools and their incorporation into one’s cognitive identity [95]. Therefore, it very
much provides grip on innovation, by contextualizing subjective systems of meaning in
30

![page30_img1.png](images/page30_img1.png)

## Page 31

relationship to the concrete (biological and sociocultural) organizations which underlie
cognition [113].
In the context of action selection related to research (e.g. experimental design,
decisions about scientific communication, acquiring funding), Active Inference provides a
division between two kinds of motivation in research: epistemic value (informational gain
related to e.g. new research findings or knowledge) and pragmatic value (reward
associated with increases in wealth, longevity, or status). Any innovation in institutional or
cognitive processes will affect the broader system both epistemically, by defining the social
norms which entail scientific activity, its cognitive underpinnings, and the kind of cues and
actions scientific activity involves; as well as pragmatically, by reinforcing the specific
behaviors which help the relevant (institutional or human) agent to reduce (subjective)
uncertainty about future sensorimotor states. When viewed through this lens, the scientific
process balances action selection based upon the both epistemic and pragmatic value,
and this process plays out across multiple scales of nested organization. For example, at
different stages of their day/career, the individual human researcher prioritizes more
exploratory/learning activities, while at other times prioritizes more exploitative/productive
activities. At the scale of an individual, institution, or even scientific field, the epistemic
activities are reflected by basic research, while the pragmatic activities can be considered
as translational or application-oriented research, or engineering.
As metascience and sociology of science typically focus on the motivations and
systems of incentive on the scale of individuals and institutions, we consider the multiscale
nature of Active Inference to provide a relevant angle for future research. This work
reflects early steps towards leveraging this framework into a more detailed understanding
of how scientific entities interact, and how contemporary changes associated with the
challenges and opportunities of DeSci could improve on the status quo.
Active Entity Ontology for Science (AEOS)
Here we describe the attributes and structural basis of a descriptive entity-oriented
ontology for Science, consistent with the principles of Active Inference and Systems
Engineering previously described. This project has the working title and acronym Active
Entity Ontology for Science (AEOS). Previously, among many other use cases, ontologies
have been used for schematic mapping of cryptographic systems and knowledge
ecosystems [114]. Additionally, approaches for complex systems modeling based upon
entities and their affordances have precedence in economics [115] and other areas. We
build on this work to create a useful and versionable resource which stands to be
developed and extended by researchers and practitioners in the coming years. This work
is meant to be a starting point for the construction of active entity-oriented ontologies of
31

## Page 32

science. See coda.io/@active-inference-institute/active-entity-ontology-for-science-aeos
for an interactive site that provides an updated version of AEOS.
Table 1 below provides example entity interaction motifs for CeSci and DeSci
ecosystem models. These motifs are presented as initial drafts on the specification and
visualization of subsystems/modules that are important for larger networks of interacting
entities.
Area of Concern
Example CeSci Motif
Example DeSci Motif
Funding
Team Publishes Knowledge Artifact (Grant),
then Requests Funding From Funding
Agency, which Scientifically Assess the
Grant.
Funding Agency Provides Opportunity for
Researcher, Provides Constraints for
Researcher, and Sends Fungible Asset to
the Team
Team Publishes Knowledge Artifact
(Grant), then Requests Funding From a
DAO which Scientifically Assesses the
Grant
The DAO Engages in governance
regarding the Grant and Sends Fungible
Asset to the Team
Communication
Academic University Communicates
Asynchronously To Lab via Computer
System
Human Communicates Asynchronously
or Synchronously To Human via the
Computer System
Cryptographic
Relationship
Research investigator uses Cryptographic
Identity on Computer System to Provides
Funding To Team and Provides Updates to
Funder like Research Agency (Government)
Research investigator Team transfers
Intellectual Property as Non-Fungible
Asset to a DAO Cryptographic Identity
that is a Smart Contract .
Scientific
Review
A Research investigator Sends Fungible
Asset to Journal/Publisher to Publish
Knowledge Artifact like a Paper. The
Journal/Publisher Requests Scientific
Review from other Research investigators
After Publishing a preprint Paper, the
Research investigator receives a
Fungible Asset as a function of the use of
the Knowledge Artifact
Research &
Analysis
Research investigators in Lab Team
Creates, Scientifically Assesses, then
Publishes Knowledge Artifact
Research investigators in DAO
Community of Practice Creates,
Scientifically Assesses, then Publishes
Knowledge Artifact
Publishing
After Scientific Review, a Research
investigator Publishes Knowledge Artifact at
Journal/Publisher.
A Research investigator Publishes
Knowledge Artifact like a Paper on a
DAO-produced knowledge market.
32

## Page 33

Table 1. Some example motifs of Research and Funding in CeSci (Centralized Science)
and DeSci (Decentralized Science). See the AEOS site for an interactive version of this
Table, where all entities and affordances can be unpacked and explored. Capitalized
words are AEOS terms.
AEOS Entity partitioning
In accordance with the theory of Active Inference, any AEOS entity consists of 3
kinds of states that are partitioned off from the entity’s environment: Internal states (states
inside the entity that are hidden from the outside, for example those involved in cognition),
Sense states (incoming statistical dependencies), and Active states (outgoing statistical
dependencies). Together these three states (Internal, Active, and Sense) constitute the
"particular" or autonomous states of the entity [94]. One of the advantages of Active
Inference as an integrative framework is this: used instrumentally, it can model various
phenomena and ensembles across scales from the quantum [116] to the social and
planetary [117,118], including heterogeneous types of entities which enact the scientific
ecosystems of today and tomorrow.
Using the Markov blanket partitioning of variables on a Bayesian graph: Entities
consist of specified Internal states, reflecting the (observer's model of) the Entity's
generative model of perception, cognition, and action. External states are those that are
inferred or acted on by Entities. Internal states are partitioned from external states via a
Markov blanket, which is further divided into nodes with incoming statistical dependencies
(Sense states) and outgoing statistical dependencies (Action states) [90,93,94]. Active
states are reached via an internal process of policy selection, which incorporates
information about what affordances (capacities for action) exist and the preferences that
the entities would like to reduce uncertainty about the realization of (for example a
homeostatic preference for comfortable temperatures).
Entity models of this type have advantages in that they can be compared across
entity types and environmental surroundings [111], and enable variational Bayesian
inference and message passing algorithms [119]. In this case of online DeSci, a type of
digital message passage algorithm is occurring on a bipartite graph composed of active
and informational entities in the AEOS. This may be represented as a factor graph (a
mathematical formalism that enables computation on large Bayesian models), which
presents developers with an interesting avenue for future research and application
[119–121].
33

## Page 34

AEOS Entity Classes and Types
The fundamental object or kind of thing in the AEOS is the Active Entity, which in
online settings is almost always interacting with Informational Entities [23]. Drawing upon
the Active Inference framing of a system, this entity is a system of interest that is
partitioned from its niche surroundings based upon its capacity for persistence, thus
making it a "thing" [122,123].
In AEOS, there are two Classes of Entity (Figure 3): Active epistemic entities (which
select action policies and enact affordances), and Informational entities (which contain
information and are acted upon). There are multiple Entity Types within each Class. In the
digital/online setting, Active Entities are always intermediated by Informational entities
(e.g. People can talk online through modification of a shared epistemic niche reflected by a
computer system). Informational entities are always intermediated by Active entities (e.g.
two databases can only communicate with each other through a computer system). The
distinction between "Adaptive" and "Mere" Active Inference entities (e.g. those that engage
sentiently and adaptively in the world) in introduced in [124].
Figure 3. Entity Classes in Active Entity Ontology for Science (AEOS). Active
Entities (blue boxes) are any kind of organization with agency at any scale, for example
humans, teams, organizations, institutions, and DAOs. Informational Entities (orange
circle) is any type of computational data or epistemic resource, for example a dataset,
grant, article, or fungible/non-fungible asset.
34

![page34_img1.png](images/page34_img1.png)

## Page 35

Both classes of entities (Active and Informational) inherit the same type of Active
Inference partitioning scheme described above. See the AEOS site for an updated list of
the diverse types of entities currently modeled. Some examples of Active entities in AEOS
include Human, Computer System, Organization, Team, DAO, Academic University, and
Community. Of special note are the Organizational entities, composed of groups of
humans, computer systems, or other organizations (e.g. a university made up of
departments, made up of labs, made up of people), along with their informational
resources. As with any other Active Entity, Organizational entities engage in outgoing
actions from a set of discrete or continuous options (affordances), and also can be the
recipient/target of action from other entities.
Entities of the Informational class are those which are being modeled for their
informational content, and are solely acted upon. Some types of Informational entities may
also display agentic qualities when used by Organizational and Computer System entities,
for example a software program. The Informational entities are used as epistemic
resources in the niche of other Agent-class entities. Some examples of Informational
entities in AEOS include Knowledge Artifact, Fungible Asset, Non-Fungible Asset,
Dataset, Metadata, Paper, Intellectual Property, Code, Grant, and Blockchain.
AEOS Policy Selection, Areas of Concern, and Roles
Active entities engage in policy selection based upon their internal generative
model (Figure 4). The behavioral policies that active entities enact are selected from a set
of agent-specific possible action sequences for that entity: their affordances. In AEOS, the
affordances are determined by the type of entity, its roles, and its context – for example a
human entity with administrator role may have the affordance to modify a digital file in a
certain situation. The specific affordances that an entity engages in, in a specific setting,
reflect the role (formal/assigned or informal) that the entity has (Figure 5). This role-based
approach towards organization draws on Systems Engineering [23], and allows the
flexibility to both describe current systems as well as design new systems. For example if
a human is providing grant funding to a team, that human entity has the role "Funding
provider". With respect to the affordance of providing funding to a research team, this role
could also have been fulfilled by a funding agency or a DAO.
35

## Page 36

Figure 4. Active entity action selection. Using the Active Inference partitioning
[101,123], the Active Entity at top left can be modeled as having incoming Sense States
and outgoing Action States which interact in an affordance-based fashion with the niche.
The internal generative model of the entity, here modeled as a partially observable Markov
Decision Process, describes the process of action selection given an incoming stream of
observations.
Figure 5. Entity Roles and Role Based Access Control (RBAC). Each Entity has
Affordances (capacities for action) and can have Roles in the organizational context
(assigned performances to be done) In the online context, all Active entities are
36

![page36_img1.png](images/page36_img1.png)

![page36_img2.png](images/page36_img2.png)

## Page 37

intermediated by computer systems, and implicitly informational entities (shared epistemic
niche).
Below in Figure 6-8 are several initial graphical examples of applications of the
AEOS. The natural language sentences at the bottom of each image reflect CeSci and
DeSci motifs to be found within each Area of Concern (Communication, Scientific Review,
Funding). Words in blue reflect AEOS entities and active relationships (affordances) that
link the entities. These examples show the capacity of AEOS to model various key motifs
and patterns found in both CeSci and DeSci systems. One possible advantage of a
system such as AEOS is that scientific motifs can be described in natural language, have
an interactive visual representation, and also provide an avenue for formal modeling.
​
Figure 6. Example motifs of Funding in CeSci (left) and DeSci (right). Blue boxes
are Active Entities in AEOS, orange circles are Informational Entities. In this example, the
initial timestep t=1 is noted on the top of each side, and the following timestep t=2 is
below.
37

![page37_img1.png](images/page37_img1.png)

## Page 38

Figure 7. Example motifs of Communication in CeSci (left) and DeSci (right).
Figure 8. Example motifs of Scientific Review in CeSci (left) and DeSci (right).
38

![page38_img1.png](images/page38_img1.png)

![page38_img2.png](images/page38_img2.png)

## Page 39

BOLTS of the CeSci to DeSci Integration
The difficulties of successful implementation of novel systems, even where they are
simple or viable to be implemented in parallel rather than as replacements to extant
systems, lend themselves to underestimation. Successful, ubiquitous systems often feel
obvious in terms of their value proposition and their reasons for adoption. For example, the
product "bar code" would appear to be ubiquitous and obvious in terms of its benefits, but
despite its conception in the mid-20th Century, by 2004 only around 80 to 90% of the
United States’ top 500 companies had adopted them [125]. In terms of its roughly 50-year
climb to this broad adoption, it took a quarter of a Century just to find adoption within a
single market segment after a small collection of large institutions began courting potential
standards [125,126]. The difficulties of implementation, adoption, and integration of novel
systems relevant to our purposes can be summarized through a qualitative use of the
Active Inference Conflict Model, which helps to frame conflict across business, operations,
legal, technical, and social (BOLTS) surfaces, especially where it relates to information
differentials [103]. Below, the bar code is used as an example that characterizes how the
BOLTS surfaces below apply to complex patterns of adoption and operation:
●
Business. Integration with and replacement of extant receipt and payment
systems, few of which would have been compatible with any bar code standard,
represented a significant difficulty. Additionally there are the financial and
non-financial costs of barcode adoption that an organization would have to
consider.
●
Operations. Integration with and replacement of extant inventory systems, few of
which would have been compatible with any bar code standard, represented a
significant difficulty.
●
Legal. For some products, technical standards and instantiations of product IDs
would have to map to regulatory compliance systems, contracts, and be available
for use in advanced records systems.
●
Technical. While the idea of the bar code was conceived in the mid-20th Century, it
took decades to confirm widely adopted standards. Adoption at any given point in
the climb to broad adoption meant the risk of implementing a pattern that might
become obsolete.
●
Social. People didn’t necessarily trust machines, and were concerned with the
potential for pricing errors at the time-of-scan. Companies were also concerned with
how bar-codes would affect the shelf-appeal and artistic design of products.
This “bar code” example helps to illustrate how the implementation, adoption, and
integration of new systems is roughly proportionate with the abstract and real integration
conflicts with extant systems. While modern, shared technological infrastructure has
39

## Page 40

allowed for more expedient implementation of new systems, conflicts across BOLTS
surfaces still exist and may be greatly exacerbated when systems are intended to be used
by institutions and organizations, which have their own formally-defined BOLTS protocols,
in addition to individuals where BOLTS protocols are likely to be encoded in norms, habits,
and narratives [103].
While DeSci may be implemented in parallel to CeSci, the overwhelming majority of
research funding is channeled through large organizations, foundations, and especially
government agencies. Each of these and their respective channels are beholden to a
complex, complicated, and interconnected web of BOLTS standards, norms, and controls,
which are a struggle to navigate. While it is tempting to make do with what funds might be
available, a bridge between CeSci and DeSci is not just about access to funds at scale but
also about access to tangible resources, influence on institutional funding and agendas,
and ease of collaboration within and among organizations and individuals. In addition,
solving conflicts relevant to a DeSci-CeSci bridge may lead to greater accountability,
reliability, and impact of DeSci information products, and therefore improve the ability for
potential investors to trust in the new marketplace of ideas, in terms of epistemic
commons. Areas of tension and interoperation between DeSci and CeSci are summarized
through the lens of BOLTS surfaces below:
Business
While government research agencies and non-government research organizations
are not operating under the return on investment model found in for-profit businesses, they
still consider the dollar cost of impact, opportunity costs, and likelihood of impact.
Agencies have to consider how to structure "portfolios" of funded projects and initiatives
based on their current mission, the missions of their stakeholders, their sources of funding,
and the track-records of the organizations and people they fund. In order to make DeSci
compatible with or even superior to CeSci projects and organizations in this context, DeSci
systems would have to enable a high level of situational awareness for funders in terms of
both progress and potential for impact, while offering advantages in terms of cost and
comparability. This may not be immediately achieved – like most new services and
systems, DeSci may start off with inefficient processes, and will become cheaper and
more efficient in producing informational products of higher quality over time.
Operations
Accountability and use of funds is taken very seriously in many of the CeSci funding
channels, especially where funding flows through multiple hierarchies. Even where funding
flows through less steps, such as in NGOs, demonstrating success in order to raise funds
constitutes its own stream of work requiring a substantial allocation of attention and
funding. Documentation and records management in operations serves to fulfill the
40

## Page 41

"evidential function", pro forma evidence of action and use [127]. These records should be
systematic and need to meet relevant standards for reliability, integrity, compliance, and
comprehensiveness [128]. It is here that DeSci offers challenges and opportunities relative
to traditional systems [20]. Where CeSci organizations rely heavily on manual processes
in order to meet documentation requirements, DeSci is predisposed toward generating
records by merit of reliance on formalized, computational protocols. However, these
advantages will only be realized if there is a directed effort to ensure
automatically-generated records are adapted to map to the myriad CeSci requirements.
Legal
Many of the business-, technical-, and operations-related documentation
requirements extend to legal use. It is in this domain that records serve their "warranty
function", clarifying intents, deliverables, requirements, and other agreements [127].
Project documentation and both personal and organizational identity become key
obstacles to DeSci's compatibility with CeSci, especially where teams are emergent or
may change in structure over time. Extensive work is required in this domain in order to
overcome legal compatibility challenges, given that no amount of advantages of efficiency,
cost, quality, or impact can allow DeSci to circumvent problems related to CeSci funders
being able to exercise their options related to legal recourse.
Technical
Technical aspects of CeSci funding blend with operations and legal aspects, as
would be the case with most bureaucratic systems. The synthesis between these two
sections (operations and legal) is where the potential bridge between DeSci and CeSci is
currently most precarious, given that an inability to meet legal requirements will eliminate
this bridge entirely. However, it is also where the bridge has the potential to outperform
traditionally-funded organizations, as meeting operating and legal requirements
computationally would, as previously suggested, offer a number of advantages.
Social
The proverbial "pipeline" of government and institutional funding to research and
development can be monolithic, baroque, and at times quite obscure [129]. Those involved
in the channeling of funds have standard operating procedures encoded both in terms of
business, operations, legal, and technical surfaces, as well as in social norms and in
narratives. The preference for DeSci to offer more accountability or other efficiencies for
research is compatible with the preferences and requirements of those who currently
occupy positions which channel funds in CeSci. Case studies, post-mortems, and high
levels of visibility in both successes and failures, as well as test-runs in incubators or
regulatory sandboxes, could assist in communicating the benefits of DeSci to potential
stakeholders. Further, finding ways to allow DeSci systems to be flexible enough to map
41

## Page 42

directly to CeSci standards for accountability could also facilitate DeSci integration. For
example, even though DeSci allows for and can encourage purely horizontal teams, one
can start with teams which can: assign a higher authority and offer them power of direct
oversight, allow them to measure and assess compliance and performance, and be
responsible for reporting and conveying that information to funders (even where this could
be automated) [130] to help to introduce CeSci standards to DeSci without requiring
excessive accommodation.
BOLTS and Reliability for Science Systems
It is important to provide reliability in all of these BOLTS dimensions, as new
technologies and systems can easily fail if there are repeated bad experiences or
unexpected incompatibilities. Even if the technical and legal aspects of DeSci funding
out-perform CeSci funding, failure of DeSci to provide a smooth integration from the social
and business perspectives could delay or prevent wide adoption. For example, traditional
funding agencies often come in at middle-to-late stages of research projects, which is why
so much preliminary work is required for a NSF/NIH grant. It is often said grants are not
“for ideas”, but rather they are for seriously-developed avenues of specific tractable
research. In DeSci, some more speculative vehicles might exist (e.g. a “Learner’s Fund”),
but we cannot expect these new vehicles to back-propagate the risk-tolerance of DeSci
into legally- and organizationally-bound CeSci funding agencies.
A priority of both CeSci and DeSci is to ensure reliability of outcomes and high
levels of accountability, in complex scientific commons. If a team has been funded,
disintegration or failure to produce work doesn’t just mean a negative outcome for the
team, it means a negative outcome for the funder, which has the potential for network
impacts on trust within the entire market. This means that teams within the DeSci space
would not only need to vastly outperform CeSci in terms of cost, time-to-impact, and
efficiency, but would also need to effectively be perceived as high-reliability organizations
in order for the market to survive. Given that DeSci teams are likely to be emergent, they
may not be able to rely on intimate trust between one another, and instead will have to rely
heavily on shared Ontology, Narrative, Formal documents, and Tools [23]. The team
working environment should be designed to create the best possible conditions for the
cognitive security of the individuals on teams, such that they stay on task, maintain
consistency in terms of goals and intents, and understand their impact on the team and
the future of the market as a whole. This type of system design also means high cognitive
security on the part of the funder, which means high visibility and high quality narrative
information management tools at their disposal [41], the ability to do retroactive analysis of
track-records of individuals and collections of individuals, the ability to rapidly compare
projects, and the ability to negotiate cost as a basis to reduce risk based on current
42

## Page 43

information. Additionally, transparency ensures that when things do go wrong, it reduces
the negative sentiment of the funder on the network, and might be mitigated by insurance
options (e.g. community stake in research) or in-house recourse options (e.g. sanctions
and penalties on negligent and bad-faith actors)
While we detailed the BOLTS conflicts related to funding processes as DeSci
practices are integrated into existing CeSci practices, this framework can be applied to
several other aspects of this integration such as the publication of knowledge artifacts,
scientific review of knowledge artifacts, communicating research results at conferences,
analyzing data, training and mentoring new researchers, and developing scientific
software. CeSci and other fields already have successful aspects which should be
preserved as DeSci is integrated, such as specific mentorship, individual, and peer
development programs. As a framework that furthers the holistic design of
transdisciplinary systems for science, we we suggest that Active Inference provides
additional insights and directions related to DeSci, which have been outlined in the this
work.
43

## Page 44

Discussion
This paper evaluated the opportunity to implement decentralized science in Web3
technologies from the perspective of Active Inference.
We first highlighted the distinction between two approaches to scientific research,
respectively Centralized Science (CeSci), where research is driven from the top-down
perspective of centralizing institutions, and Decentralized Science (DeSci), where it is
driven from the bottom up through situated sensemaking by communities of practices.
Both approaches may work together, without either approach dominating the other. We
argued that the DeSci stance can facilitate the governance of scientific knowledge
(understood as an epistemic common good) by providing both the incentive and the
opportunity to build a detailed understanding of various systems and problems in an
integrated language.
We consequently reviewed potential avenues to implement DeSci in the era of
Web3. We underlied a major limitation to marketization and tokenization, one of the main
mechanisms of Web3 technologies, namely that creating real world value for the token
renders it vulnerable to speculative investment and hoarding (and other corruptive
influence). We discussed the potential roles and benefits of modern affordances, including
online teams, blockchain, tokenization, and smart contracts. Finally, we discussed several
practical aspects of integrating DeSci concepts and practices into CeSci, such as
business, operations, legal, technical, and social (BOLTS) aspects of this transition.
Finally, we discussed the role of the Active Inference framework in describing
scientific ecosystems. As a formalism relating the basic activity of dynamical systems to
epistemic inference, we argued it is a natural fit to describe science as a socio-institutional
system. We outlined an Active Entity Ontology for Science (AEOS), intended to define key
Entity Classes, Entity Types, Areas of Concern, Roles, Affordances, and Action-Perception
States involved in conducting scientific work, whether CeSci or DeSci, with examples
illustrating particular embodiments of this ontology. We hope that further development of
research within the scope of AEOS would yield fruitful progress in our ability to design and
evaluate possible forms of scientific organization.
Some specific next steps include:
●
In the open source Active Blockference package [131], we are developing Active
Inference models that are implemented in the complex system modeling framework
cadCAD [132], with a goal of enabling description, analysis, and simulations of
complex cognitive ecosystems of “shared intelligence” [28].
44

## Page 45

●
Exploration of Active Inference ontologies and relationships with category theory
[116] and other formalisms.
●
Creation of Graphical User Interface for DeSci (DeSciCAD), to enable the real-time
use of AEOS in science ecosystem design and operation. For example, a user
could select entity types, assign them roles, connect the entities with
affordance-based edges, and engage in simulation and parameter optimization. The
front end of such a design interface would ideally be accessible (via visual, textual
and other modalities) and enable productive creation of composable Active
Inference simulations.
●
Exploring system designs in AEOS that scaffold and incentivize different kinds of
research such as basic/theoretical, translational/applied, and quantitative/qualitative
work. One specific application would be to use AEOS in behavioral modeling of
incentive gaps, such as getting people to publish data even if they don't publish the
paper, or finding ways to accept modular contributions, so that in the case of
retraction or deciding not to publish, pieces can still be made available and citable.
We hope that this work has provided a useful understanding of DeSci with respect
to the current CeSci system in terms of what is possible now and in the future, and an
interactive modeling framework in AEOS which we will continue to develop and apply.
45

## Page 46

Author contributions using CRediT – Contributor Roles Taxonomy [133]:
●
Conceptualization (AGC, JAB, AKS, SED, VBK, AC, DAF)
●
Funding acquisition – No funding was acquired specifically for this project.
●
Investigation (AGC, JS, SED, DAF)
●
Project administration (DAF)
●
Visualization (VBK, DAF)
●
Writing – original draft (AGC, AKS, AC JS, SS, SED, VBK, RJC)
●
Writing – review & editing (DAF, SAS, RJC, AC, SED, AGC, VBK, IM, SS, AKS, JS
CT, ASV)
Works cited
1.
Weaver W. Science and complexity. Am Sci. 1948;36: 536–544. Available:
https://www.ncbi.nlm.nih.gov/pubmed/18882675
2.
Gershenson C. The Implications of Interactions for Science and Philosophy. arXiv [nlin.AO]. 2013. pp.
781–790. doi:10.1007/s10699-012-9305-8
3.
Wright C, Bechtel W. - Mechanisms and psychological explanation. In: Thagard P, editor. Philosophy of
Psychology and Cognitive Science. Amsterdam: North-Holland; 2007. pp. 31–79.
doi:10.1016/B978-044451540-7/50019-0
4.
Chan ME, Arvey RD. Meta-Analysis and the Development of Knowledge. Perspect Psychol Sci.
2012;7: 79–92. doi:10.1177/1745691611429355
5.
Lawrence MG, Williams S, Nanz P, Renn O. Characteristics, potentials, and challenges of
transdisciplinary research. One Earth. 2022;5: 44–61. doi:10.1016/j.oneear.2021.12.010
6.
Head ML, Holman L, Lanfear R, Kahn AT, Jennions MD. The extent and consequences of p-hacking in
science. PLoS Biol. 2015;13: e1002106. doi:10.1371/journal.pbio.1002106
7.
Rivera H. Fake Peer Review and Inappropriate Authorship Are Real Evils. J Korean Med Sci. 2019;34:
e6. doi:10.3346/jkms.2019.34.e6
8.
Scott JC. The Trouble with the View from Above. 2010 [cited 30 Jan 2022]. Available:
https://www.cato-unbound.org/2010/09/08/james-c-scott/trouble-view-above/
9.
Fanelli D. “Positive” results increase down the Hierarchy of the Sciences. PLoS One. 2010;5: e10068.
doi:10.1371/journal.pone.0010068
10.
Vazire S. Implications of the Credibility Revolution for Productivity, Creativity, and Progress. Perspect
Psychol Sci. 2018;13: 411–417. doi:10.1177/1745691617751884
11.
Munafò MR, Nosek BA, Bishop DVM, Button KS, Chambers CD, Percie du Sert N, et al. A manifesto
for reproducible science. Nature Human Behaviour. 2017;1: 1–9. doi:10.1038/s41562-016-0021
12.
Reproducibility and reliability of biomedical research. [cited 26 Feb 2022]. Available:
https://acmedsci.ac.uk/policy/policy-projects/reproducibility-and-reliability-of-biomedical-research
13.
Hills TT. The Dark Side of Information Proliferation. Perspect Psychol Sci. 2018; 1745691618803647.
46

## Page 47

doi:10.1177/1745691618803647
14.
Barker Bausell R. The Problem with Science: The Reproducibility Crisis and What to do About It.
Oxford University Press; 2021. Available:
https://play.google.com/store/books/details?id=oHEWEAAAQBAJ
15.
Bar-Yam Y. Multiscale variety in complex systems. Complexity. 2004;9: 37–45. doi:10.1002/cplx.20014
16.
Bruni LE, Giorgi F. Towards a heterarchical approach to biology and cognition. Prog Biophys Mol Biol.
2015;119: 481–492. doi:10.1016/j.pbiomolbio.2015.07.005
17.
Feinerman O, Korman A. Individual versus collective cognition in social insects. J Exp Biol. 2017;220:
73–82. doi:10.1242/jeb.143891
18.
Rossi F, Bandyopadhyay S, Wolf M, Pavone M. Review of Multi-Agent Algorithms for Collective
Behavior: a Structural Taxonomy. IFAC-PapersOnLine. 2018;51: 112–117.
doi:10.1016/j.ifacol.2018.07.097
19.
Mohanty D. Ethereum Architecture. In: Mohanty D, editor. Ethereum for Architects and Developers:
With Case Studies and Code Samples in Solidity. Berkeley, CA: Apress; 2018. pp. 37–54.
doi:10.1007/978-1-4842-4075-5_2
20.
Nabben K. Decentralised Autonomous Organisations (DAOs) as Data Trusts: A General-purpose Data
Governance Framework for Decentralised Data Ownership, Storage, and Utilisation. 2021.
doi:10.2139/ssrn.4009205
21.
Schneider N, De Filippi P, Frey S, Tan JZ, Zhang AX. Modular Politics: Toward a Governance Layer for
Online Communities. Proc ACM Hum-Comput Interact. 2021;5: 1–26. doi:10.1145/3449090
22.
Gencer AE, Basu S, Eyal I, van Renesse R, Sirer EG. Decentralization in Bitcoin and Ethereum
Networks. arXiv [cs.CR]. 2018. Available: http://arxiv.org/abs/1801.03998
23.
Vyatkin A, Metelkin I, Mikhailova A, Cordes RJ, Friedman DA. Active Inference & Behavior Engineering
for Teams. 2020. doi:10.5281/zenodo.4021163
24.
Fields C, Levin M. How Do Living Systems Create Meaning? Philosophies. 2020;5: 36.
doi:10.3390/philosophies5040036
25.
Veissière SPL, Constant A, Ramstead MJD, Friston KJ, Kirmayer LJ. Thinking through other minds: A
variational approach to cognition and culture. Behav Brain Sci. 2019;43: 1–97.
doi:10.1017/S0140525X19001213
26.
Albarracin M, Demekas D, Ramstead MJD, Heins C. Epistemic Communities under Active Inference.
Entropy . 2022;24. doi:10.3390/e24040476
27.
Kiefer AB. Psychophysical identity and free energy. J R Soc Interface. 2020;17: 20200370.
doi:10.1098/rsif.2020.0370
28.
Friston KJ, Ramstead MJD, Kiefer AB, Tschantz A, Buckley CL, Albarracin M, et al. Designing
Ecosystems of Intelligence from First Principles. arXiv [cs.AI]. 2022. Available:
http://arxiv.org/abs/2212.01354
29.
Bruineberg J, Kiverstein J, Rietveld E. The anticipating brain is not a scientist: the free-energy principle
from an ecological-enactive perspective. Synthese. 2018;195: 2417–2444.
doi:10.1007/s11229-016-1239-1
47

## Page 48

30.
Constant A, Clark A, Friston KJ. Representation Wars: Enacting an Armistice Through Active Inference.
Front Psychol. 2020;11: 598733. doi:10.3389/fpsyg.2020.598733
31.
Hipólito I. Cognition Without Neural Representation: Dynamics of a Complex System. Front Psychol.
2021;12: 643276. doi:10.3389/fpsyg.2021.643276
32.
Guénin--Carlut A. Thinking like a State - Embodied intelligence in the deep history of our collective
minds. 2021 [cited 25 Jan 2022]. doi:10.31219/osf.io/dxnzt
33.
Buterin V. The meaning of decentralization. Medium com. 2017;6.
34.
Nabben K. Is a DAO a Panopticon? Algorithmic Governance as Creating and Mitigating Vulnerabilities
in “Decentralised Autonomous Organisations.” 2021. doi:10.2139/ssrn.3907693
35.
Kiayias A, Lazos P. SoK: Blockchain Governance. arXiv [cs.CR]. 2022. Available:
http://arxiv.org/abs/2201.07188
36.
Graber J. Web3 is Self-Certifying. In: Medium [Internet]. 23 Dec 2021 [cited 26 Feb 2022]. Available:
https://jaygraber.medium.com/web3-is-self-certifying-9dad77fd8d81
37.
Benet J. IPFS - Content Addressed, Versioned, P2P File System. arXiv [cs.NI]. 2014. Available:
http://arxiv.org/abs/1407.3561
38.
IPFS powers the distributed web. [cited 15 Feb 2022]. Available: https://ipfs.io/
39.
Tenorio-Fornés Á, Tirador EP, Sánchez-Ruiz AA, Hassan S. Decentralizing science: Towards an
interoperable open peer review ecosystem using blockchain. Inf Process Manag. 2021;58: 102724.
doi:10.1016/j.ipm.2021.102724
40.
Hamburg S. A Guide to DeSci, the Latest Web3 Movement. In: Future [Internet]. 9 Feb 2022 [cited 25
Feb 2022]. Available: https://future.a16z.com/what-is-decentralized-science-aka-desci/
41.
Cordes RJ, Applegate-Swanson S, Friedman DA, Knight VB, Mikhailova A. Narrative Information
Management. Zenodo. COGSEC; 2021. doi:10.5281/zenodo.5573287
42.
Szablowski D. “Legal enclosure” and resource extraction: Territorial transformation through the
enclosure of local and indigenous law. The Extractive Industries and Society. 2019;6: 722–732.
doi:10.1016/j.exis.2018.12.005
43.
From Theory to Practice. The MIT Press; 2007. Available: http://www.jstor.org/stable/j.ctt5hhdf6
44.
Wilson DS, Ostrom E, Cox ME. Generalizing the core design principles for the efficacy of groups. J
Econ Behav Organ. 2013;90: S21–S32. doi:10.1016/j.jebo.2012.12.010
45.
Paul W.B. Atkins, David Sloan Wilson, Steven C. Hayes. Prosocial. 2019. Available:
https://www.newharbinger.com/9781684030248/prosocial/
46.
Wilson DS, Philip MM, MacDonald IF, Atkins PWB, Kniffin KM. Core design principles for nurturing
organization-level selection. Sci Rep. 2020;10: 13989. doi:10.1038/s41598-020-70632-8
47.
Beck J. What is Web3? Here Are Some Ways To Explain It To A Friend. In: ConsenSys [Internet]. 12
Jan 2022 [cited 22 Feb 2022]. Available:
https://consensys.net/blog/blockchain-explained/what-is-web3-here-are-some-ways-to-explain-it-to-a-fri
end/
48

## Page 49

48.
Edelman G. The Father of Web3 Wants You to Trust Less. Wired. 29 Nov 2021. Available:
https://www.wired.com/story/web3-gavin-wood-interview/. Accessed 26 Feb 2022.
49.
ĐApps: What Web 3.0 Looks Like. [cited 22 Feb 2022]. Available:
https://gavwood.com/dappsweb3.html
50.
Kim HM, Laskowski M, Nan N. A First Step in the Co-Evolution of Blockchain and Ontologies: Towards
Engineering an Ontology of Governance at the Blockchain Protocol Level. arXiv [cs.CY]. 2018.
Available: http://arxiv.org/abs/1801.02027
51.
Graeber D. Debt : The First 5000 Years. In: Melville House Books [Internet]. 2011 [cited 18 Feb 2022].
Available: https://www.mhpbooks.com/books/debt/
52.
clepp. Molecule’s biopharma IP-NFTs — A technical description. In: Molecule Blog [Internet]. 6 Aug
2021 [cited 15 Feb 2022]. Available:
https://medium.com/molecule-blog/molecules-biopharma-ipnfts-a-technical-description-4dcfc6bf77f8
53.
Bartoletti M, Pompianu L. An empirical analysis of smart contracts: platforms, applications, and design
patterns. arXiv [cs.CR]. 2017. Available: http://arxiv.org/abs/1703.06322
54.
Khan SN, Loukil F, Ghedira-Guegan C, Benkhelifa E, Bani-Hani A. Blockchain smart contracts:
Applications, challenges, and future trends. Peer Peer Netw Appl. 2021; 1–25.
doi:10.1007/s12083-021-01127-0
55.
Shorish J, Stephenson M, Zargham M. A Practical Theory Of Fungibility. WU Vienna University of
Economics and Business; 2021 May. Report No.: 8137. Available:
https://ideas.repec.org/p/wiw/wus051/8137.html
56.
Nandwani A, Gupta M, Thakur N. Proof-of-Participation: Implementation of Proof-of-Stake Through
Proof-of-Work. International Conference on Innovative Computing and Communications. Springer
Singapore; 2019. pp. 17–24. doi:10.1007/978-981-13-2324-9_3
57.
poapxyz. POAP. [cited 1 Mar 2022]. Available: https://www.poap.xyz/
58.
Peters G, Panayi E, Chapelle A. Trends in Cryptocurrencies and Blockchain Technologies: A Monetary
Theory and Regulation Perspective. 2015. Available: https://papers.ssrn.com/abstract=3084011
59.
Carson KA. Exodus. 2021. Available:
https://theanarchistlibrary.org/library/kevin-a-carson-exodus?v=1624008858
60.
Duniter. In: Duniter [Internet]. [cited 1 Mar 2022]. Available: https://duniter.org/
61.
McConaghy T. Towards a Practice of Token Engineering. In: Ocean Protocol [Internet]. 1 Mar 2018
[cited 19 Jan 2022]. Available:
https://blog.oceanprotocol.com/towards-a-practice-of-token-engineering-b02feeeff7ca
62.
Analysis D. Table of contents. Int J Radiat Oncol Biol Phys. 2003;57: 1519–1532.
doi:10.1016/j.ijrobp.2003.10.003
63.
Tools for the Web3 Data Economy. In: Ocean Protocol [Internet]. [cited 22 Jan 2022]. Available:
https://oceanprotocol.com/
64.
Williams S, Berman L, Raybould I, Uemlianin I. Arweave: A protocol for economically sustainable
information permanence. [cited 15 Feb 2022]. Available: https://www.arweave.org/yellow-paper.pdf
49

## Page 50

65.
Smart contract research forum. In: Smart Contract Research Forum [Internet]. [cited 15 Feb 2022].
Available: https://www.smartcontractresearch.org/
66.
DeSci Foundation. Why we need to fundamentally rethink scientific publishing. In: Medium [Internet]. 1
Dec 2021 [cited 15 Feb 2022]. Available:
https://desci.medium.com/why-we-need-to-fundamentally-rethink-scientific-publishing-43f2ae39af76
67.
Trovò B, Massari N. Ants-Review: A Privacy-Oriented Protocol for Incentivized Open Peer Reviews on
Ethereum. Euro-Par 2020: Parallel Processing Workshops. Springer International Publishing; 2021. pp.
18–29. doi:10.1007/978-3-030-71593-9_2
68.
Avital M. Peer Review: Toward a Blockchain-enabled Market-based Ecosystem. Communications of the
Association for Information Systems. 2018;42: 28. doi:10.17705/1CAIS.04228
69.
van Rossum J. The blockchain and its potential for science and academic publishing. Inf Serv Use.
2018;38: 95–98. doi:10.3233/isu-180003
70.
The ResearchCoin Whitepaper. In: ResearchHub [Internet]. [cited 15 Feb 2022]. Available:
https://www.researchhub.com/paper/819400/the-researchcoin-whitepaper
71.
Tenorio-Fornés A, Jacynycz V, Llop-Vila D, Sánchez-Ruiz A, Hassan S. Towards a Decentralized
Process for Scientific Publication and Peer Review using Blockchain and IPFS. 2019.
doi:10.24251/HICSS.2019.560
72.
Jan Z, Third A, Ibanez L-D, Bachler M, Simperl E, Domingue J. ScienceMiles: Digital Currency for
Researchers. Companion Proceedings of the The Web Conference 2018. Republic and Canton of
Geneva, CHE: International World Wide Web Conferences Steering Committee; 2018. pp. 1183–1186.
doi:10.1145/3184558.3191556
73.
Ethereum. Ethereum Whitepaper. In: ethereum.org [Internet]. [cited 14 Feb 2022]. Available:
https://ethereum.org/en/whitepaper/
74.
Explanation of DAICOs. In: Ethereum Research [Internet]. 6 Jan 2018 [cited 30 Jan 2022]. Available:
https://ethresear.ch/t/explanation-of-daicos/465
75.
Tamari R, Friedman D, Fischer W, Hebert L, Shahaf D. From Users to (Sense)Makers: On the Pivotal
Role of Stigmergic Social Annotation in the Quest for Collective Sensemaking. Proceedings of the 33rd
ACM Conference on Hypertext and Social Media. New York, NY, USA: Association for Computing
Machinery; 2022. pp. 236–239. doi:10.1145/3511095.3536361
76.
Science Fund. [cited 28 Feb 2022]. Available: https://www.sciencefund.io/
77.
El Damaty S. Rich in Data, Poor in Wisdom: Science Needs a Decentralized Data Commons. In: OpSci
[Internet]. 4 Dec 2021 [cited 27 Feb 2022]. Available:
https://pulse.opsci.io/rich-in-data-poor-in-wisdom-science-needs-a-decentralized-data-commons-98c7ff
db56a1
78.
Ocean Protocol Foundation with BigchainDB GmbH. Ocean Protocol: Tools for the Web Data Economy.
Available: https://oceanprotocol.com/tech-whitepaper.pdf
79.
Datum. Datum-WhitePaper.pdf. Available: https://datum.org/assets/Datum-WhitePaper.pdf
80.
Dataspace. datapace_whitepaper.pdf. Available: https://www.datapace.io/datapace_whitepaper.pdf
81.
The Problem of Data: Data Management and Curation Practices Among University Researchers. In:
50

## Page 51

CLIR [Internet]. 13 Dec 2016 [cited 18 Feb 2022]. Available:
https://www.clir.org/pubs/reports/pub154/problem-of-data/
82.
Gorgolewski KJ, Auer T, Calhoun VD, Craddock RC, Das S, Duff EP, et al. The brain imaging data
structure, a format for organizing and describing outputs of neuroimaging experiments. Sci Data.
2016;3: 160044. doi:10.1038/sdata.2016.44
83.
Gregory K, Groth P, Scharnhorst A, Wyatt S. Lost or found? Discovering data needed for research.
Harvard Data Science Review. 2020. doi:10.1162/99608f92.e38165eb
84.
Kitano H. Nobel Turing Challenge: creating the engine for scientific discovery. NPJ Syst Biol Appl.
2021;7: 29. doi:10.1038/s41540-021-00189-3
85.
The Turing AI scientist grand challenge. In: The Alan Turing Institute [Internet]. [cited 16 Feb 2022].
Available: https://www.turing.ac.uk/research/research-projects/turing-ai-scientist-grand-challenge
86.
SEBoK. [cited 24 Jan 2022]. Available:
https://sebokwiki.org/wiki/Guide_to_the_Systems_Engineering_Body_of_Knowledge_(SEBoK)
87.
ISO/IEC/IEEE 15288:2015. In: ISO [Internet]. 2020 [cited 23 Feb 2022]. Available:
https://www.iso.org/standard/63711.html
88.
Essence 1.2 - SEMAT. [cited 23 Feb 2022]. Available: http://semat.org/essence-1.2
89.
Tumiel J. Spinning up in active inference and the free energy principle. In: jared tumiel [Internet]. 14 Oct
2020 [cited 21 Jan 2022]. Available: https://jaredtumiel.github.io/blog/2020/10/14/spinning-up-in-ai.html
90.
Friston K, Heins C, Ueltzhöffer K, Da Costa L, Parr T. Stochastic Chaos and Markov Blankets. Entropy.
2021;23. doi:10.3390/e23091220
91.
Parr T, Pezzulo G, Friston KJ. Active Inference: The Free Energy Principle in Mind, Brain, and
Behavior. MIT Press; 2022. Available:
https://scholar.google.ca/scholar?cluster=1067966391259043561&hl=en&as_sdt=0,5&sciodt=0,5
92.
Sims M. How to count biological minds: symbiosis, the free energy principle, and reciprocal multiscale
integration. Synthese. 2020. doi:10.1007/s11229-020-02876-w
93.
Ramstead MJD, Kirchhoff MD, Constant A, Friston KJ. Multiscale integration: beyond internalism and
externalism. Synthese. 2021;198: 41–70. doi:10.1007/s11229-019-02115-x
94.
Friston K. A free energy principle for a particular physics. arXiv [q-bio.NC]. 2019. Available:
http://arxiv.org/abs/1906.10184
95.
Kuchling F, Friston K, Georgiev G, Levin M. Morphogenesis as Bayesian inference: A variational
approach to pattern formation and control in complex biological systems. Phys Life Rev. 2019.
doi:10.1016/j.plrev.2019.06.001
96.
Fields C, Levin M. Scale-Free Biology: Integrating Evolutionary and Developmental Thinking.
Bioessays. 2020;42: e1900228. doi:10.1002/bies.201900228
97.
Friston K. Active inference and agency. Cognitive neuroscience. 2014. pp. 119–121.
doi:10.1080/17588928.2014.905517
98.
Pezzulo G, Rigoli F, Friston K. Active Inference, homeostatic regulation and adaptive behavioural
control. Prog Neurobiol. 2015;134: 17–35. doi:10.1016/j.pneurobio.2015.09.001
51

## Page 52

99.
Hipólito I, Baltieri M, Friston K, Ramstead MJD. Embodied skillful performance: where the action is.
Synthese. 2021;199: 4457–4481. doi:10.1007/s11229-020-02986-5
100. Marković D, Goschke T, Kiebel SJ. Meta-control of the exploration-exploitation dilemma emerges from
probabilistic inference over a hierarchy of time scales. Cogn Affect Behav Neurosci. 2021;21: 509–533.
doi:10.3758/s13415-020-00837-x
101. Da Costa L, Parr T, Sajid N, Veselic S, Neacsu V, Friston K. Active inference on discrete state-spaces:
A synthesis. J Math Psychol. 2020;99: 102447. doi:10.1016/j.jmp.2020.102447
102. Vasil J, Badcock PB, Constant A, Friston K, Ramstead MJD. A World Unto Itself: Human
Communication as Active Inference. Front Psychol. 2020;11: 417. doi:10.3389/fpsyg.2020.00417
103. David S, Cordes RJ, Friedman DA. Active Inference in Modeling Conflict. 2021.
doi:10.5281/zenodo.5759807
104. Schoeller F, Miller M, Salomon R, Friston K. Trust as Extended Control: Active Inference and User
Feedback During Human-Robot Collaboration. ArXiv. 2021 [cited 21 Sep 2021]. Available:
https://www.semanticscholar.org/paper/7c39921b6050d7be82e18df2a31973f2c3f864ea
105. Heyes C. Précis of Cognitive Gadgets: The Cultural Evolution of Thinking. Behav Brain Sci. 2019;42.
doi:10.1017/S0140525X18002145
106. Muthukrishna M, Doebeli M, Chudek M, Henrich J. The Cultural Brain Hypothesis: How culture drives
brain expansion, sociality, and life history. PLoS Comput Biol. 2018;14: e1006504.
doi:10.1371/journal.pcbi.1006504
107. Graeber D, Wengrow D. The dawn of everything -A new history of humanity. 2021 [cited 25 Feb 2022].
Available: https://uujec.org/Portals/0/The%20Dawn%20of%20Everything.pdf
108. Doyle MJ, Marsh L. Stigmergy 3.0: From ants to economies. Cogn Syst Res. 2013;21: 1–6.
doi:10.1016/j.cogsys.2012.06.001
109. Correia L, Sebastião AM, Santana P. On the role of stigmergy in cognition. Progress in Artificial
Intelligence. 2017;6: 79–86. doi:10.1007/s13748-016-0107-z
110. Ramstead MJD, Constant A, Badcock PB, Friston KJ. Variational ecology and the physics of sentient
systems. Phys Life Rev. 2019. doi:10.1016/j.plrev.2018.12.002
111. Constant A, Ramstead MJD, Veissière SPL, Campbell JO, Friston KJ. A variational approach to niche
construction. J R Soc Interface. 2018;15: 20170685. doi:10.1098/rsif.2017.0685
112. Friedman D, Tschantz A, Ramstead MJD, Friston K, Constant A. Active inferants: The basis for an
active inference framework for ant colony behavior. Front Behav Neurosci. 2021;15: 126.
doi:10.3389/fnbeh.2021.647732
113. Guénin--Carlut A. Cognitive agency in sociocultural evolution. 2022. doi:10.31219/osf.io/x7yr4
114. Coelho FC, Brandão A. Decentralising scientific publishing: can the blockchain improve science
communication? Mem Inst Oswaldo Cruz. 2019;114: e190257. doi:10.1590/0074-02760190257
115. Brian Arthur W. Economics in Nouns and Verbs. arXiv [econ.GN]. 2021. Available:
http://arxiv.org/abs/2104.01868
116. Fields C, Friston K, Glazebrook JF, Levin M. A free energy principle for generic quantum systems. arXiv
52

## Page 53

[quant-ph]. 2021. Available: http://arxiv.org/abs/2112.15242
117. Boik JC. Science-Driven Societal Transformation, Part I: Worldview. Sustain Sci Pract Policy. 2020;12:
6881. doi:10.3390/su12176881
118. Rubin S, Parr T, Da Costa L, Friston K. Future climates: Markov blankets and active inference in the
biosphere. J R Soc Interface. 2020;17: 20200503. doi:10.1098/rsif.2020.0503
119. Friston KJ, Parr T, de Vries B. The graphical brain: Belief propagation and active inference. Netw
Neurosci. 2017;1: 381–414. doi:10.1162/NETN_a_00018
120. van de Laar TW, de Vries B. Simulating Active Inference Processes by Message Passing. Frontiers in
Robotics and AI. 2019;6: 20. doi:10.3389/frobt.2019.00020
121. ForneyLab.jl: Julia package for automatically generating Bayesian inference algorithms through
message passing on Forney-style factor graphs. Github; Available:
https://github.com/biaslab/ForneyLab.jl
122. Friston K, Douglass DS, Iennaco de Vasconcelos ML, Sillett S, Sganzerla L, Tickles D, et al. Transcript
of: Karl Friston, 1st Applied Active Inference Symposium, Active Inference Lab, June 21, 2021. 2021.
doi:10.5281/zenodo.5797041
123. Friston K, Da Costa L, Sajid N, Heins C, Ueltzhöffer K, Pavliotis GA, et al. The free energy principle
made simpler but not too simple. arXiv [cond-mat.stat-mech]. 2022. Available:
http://arxiv.org/abs/2201.06387
124. Kirchhoff M, Parr T, Palacios E, Friston K, Kiverstein J. The Markov blankets of life: autonomy, active
inference and the free energy principle. J R Soc Interface. 2018;15: 20170792.
doi:10.1098/rsif.2017.0792
125. Weightman G. The History of the Bar Code. In: smithsonianmag.com [Internet]. 23 Sep 2015 [cited 28
Jan 2022]. Available: https://www.smithsonianmag.com/innovation/history-bar-code-180956704/
126. Silverman L. Barcodes: A Brief History. In: corp.trackabout.com [Internet]. 19 Nov 2015 [cited 28 Jan
2022]. Available: https://corp.trackabout.com/blog/barcodes-brief-history
127. Dikopoulou A, Mihiotis A. The contribution of records management to good governance. The TQM
Journal. 2012;24: 123–141. doi:10.1108/17542731211215071
128. ISO. ISO 15489 - Information and documentation - Records Management. 2001.
129. Cordes RJ. The Pipeline. The Pipeline: EdTech, Knowledge Management, and NATSEC. COGSEC;
2021.
130. Cox RJ, Wallace DA, Wallace D. Archives and the Public Good: Accountability and Records in Modern
Society. Greenwood Publishing Group; 2002. Available:
https://play.google.com/store/books/details?id=sH5BaIWT7o0C
131. Smékal J, docxology, Singh A, Active Inference Institute, Monticone P.
ActiveInferenceInstitute/ActiveBlockference: v0.1. 2022. doi:10.5281/zenodo.7114729
132. cadCAD – A Python package for designing, testing and validating complex systems through simulation.
[cited 15 Feb 2022]. Available: https://cadcad.org/
133. CRediT - Contributor Roles Taxonomy. In: CASRAI [Internet]. 14 Sep 2019 [cited 28 Feb 2022].
53

## Page 54

Available: https://casrai.org/credit/
54


---
*Extraction method: pymupdf*
