# Full Text: MVEE:  A Framework for Evolutionary Studies

> Extracted from `2018_MVEE.pdf`

---

## Page 1

MVEE: 
A Framework for 
Evolutionary Studies.
Daniel Ari 
Friedman
    2018
?
?

## Page 2

Research Question:
How can we formalize the evolution of 
heredity, environment, and phenotype 
through time and across biological levels?

## Page 3

Research Question:
How can we formalize the evolution of 
heredity, environment, and phenotype 
through time and across biological levels
……..accounting for……..
plasticity & meta-plasticity,  
interactions within and among scales,
scientific knowledge/measurement/modeling constraints, 
ecological variability over behavioral/developmental/evolutionary time,
diversity & open-endedness of evolving systems (bio/social/computational)   ?

## Page 4

Research Question:
How can we formalize the evolution of 
heredity, environment, and phenotype 
through time and across biological levels?
 Goal:
●Extend Variational Neuroethology (Ramstead et al. 2017) 
to specify a tractable general framework for all 
Evolutionary studies, Biological and Otherwise.
●
This would allow us to integrate current data across systems       
and suggest new measurements/experiments/systems.

## Page 5

This Talk:
●Background theory...Part 1
○
Claim 1: EcoEvoDevo is trapped in a locally-optimizing regime 
because it is trapped in an incoherent G-P-E & F phrasing.

## Page 6

This Talk:
●Background theory...Part 1
○
Claim 1: EcoEvoDevo is trapped in a locally-optimizing regime 
because it is trapped in an incoherent G-P-E & F phrasing.
●Background theory...Part 2
○
Claim 2: The data, theory, math, and philosophy already exist to 
synthesize a tractable generalized model of evolution.

## Page 7

This Talk:
●Background theory...Part 1
○
Claim 1: EcoEvoDevo is trapped in a locally-optimizing regime 
because it is trapped in an incoherent G-P-E & F phrasing.
●Background theory...Part 2
○
Claim 2: The data, theory, math, and philosophy already exist to 
synthesize a tractable generalized model of evolution.
●MVEE - A Way Forward….
○
MVEE sits between VNE and EcoEvoDevo. 
○
MVEE is a Synergetic framework for Evolutionary studies.
○
MVEE == Variational Neuroethology + EcoEvoDevo + Collective Behavior + ML

## Page 8

*[Page 8 appears to be blank or image-only]*

## Page 9

Theoretical 
Background
Part 1

## Page 10

“Fitness”?

## Page 11

*[Page 11 appears to be blank or image-only]*

## Page 12

Theoretical 
Background
Part 2

## Page 13

We need a way to 
integrate sub-theories.
2018
2011

## Page 14

2018

## Page 15

*[Page 15 appears to be blank or image-only]*

## Page 16

2011
A-B
A + B
kab
First issue: Complex biological 
patterns, no “rules”…. 
(e.g. not just “entropy” or ψ)
Second issue:Which rules 
apply for a specific situation? 
How do we compute or apply 
these rules/patterns? 
???
???
???

## Page 17

h2
x1
x2
x1
s
s
R
R = h2s
 
Simple state models in Biology….
Theoretical assumptions are violated in nature (H-W, additivity)
No insights/inferences into mechanisms or counterfactuals
Hard to apply to real data (or limited plausible sample size)

## Page 18

R = h2s
Is like a “F=MA” of Biology. 
Like Fundamental Theorem of NS, or Price Eq., or Indirect/Kin... 
State models don’t make assumptions about underlying processes.
We don’t think that F=MA is (or, “needs to be”) hard-coded into 
our universe. F=MA is a pattern that applies to varying degrees 
of accuracy for certain spatial temporal scales. 
What is “hard-coded” below F=MA?

## Page 19

2016
Recent work 
on Gauge 
theories and 
Information 
Entropy in 
Biology….
2017
2017
2016

## Page 20

2016
2017
2017
2017
2017
2016
Recent work 
on Gauge 
theories and 
Information 
Entropy in 
Biology….

## Page 21

2017
2017
2016
2017
Recent work 
on Gauge 
theories and 
Information 
Entropy in 
Biology….

## Page 22

RBF 
“just a coincidence”??
VNE
Variational 
Neuro-
Ethology
Ramstead, 
Badcock & 
Friston
2017

## Page 23

VNE is an Evolutionary Systems Theory about 
how hierarchically-nested Markov blankets 
performing active inference inextricably 
(statistically and mechanistically) link internal 
and external states through action and 
perception among, and across, biological 
spatio-temporal scales.
VNE applies to biological 
systems across many 
orders of magnitude of 
spatio-temporal variation.

## Page 24

RBF 2017: “Given the success of this explanatory framework in biology, 
we suggest that Tinbergen’s levels of inquiry might be apt to elucidate 
structural laws that supplement the general principles provided by the 
FEP….the FEP describes a [general biological modeling imperative], 
while Tinbergen has offered a distinctive but complementary framework 
that allows us to develop substantive explanations for the phenotypic 
traits and behaviors of any given species or organism…”
Contemporary
Historical
How?
(Proximate)
Why?
(Ultimate)
Mechanism 
Causal Explanation
Function 
Use or Survival Value 
Development 
Ontogeny, Plasticity
Evolution 
Selection and Drift

## Page 25

RBF 2017 work through only one example in their paper: 
How the “Hierarchically Mechanistic Minds” EST can be used 
within the Variational Neuroethology framework to study all 
scales of human socio-biocultural evolution in Humans.

## Page 26

Other people are already critiquing and building on VNE. 
As of 
1/3/2018
12 comments

## Page 27

1. The case study in VNE on 
human cognition is still not 
operationalized in a way that 
allows usage of actual data:
2. And it is even more unclear 
how the VNE approach could 
be deployed in other, arbitrary 
evolutionary systems…….. 
Especially using real data!

## Page 28

2013
Processes, Interactions, Context (1992,2011,2014,2016) 
Ⓕⓡⓔⓔ Ⓔⓝⓔⓡⓖⓨ
“Read Dennett (1995) and Noble (2016) and J&L (2005) 
& Akçay & Van Cleve (2016) and EFK (2000) and .....” 
Pragmatic…Empirical…Quantitative…Pluralistic...
Make predictions.…..Accommodate Complexity…
“Keep it rational and measurable”

## Page 29

The data, theory, and philosophy already exist to 
synthesize a tractable generalized model of evolution!
 
The challenge lies in finding how to 
integrate the VNE with classical 
evolutionary theory and apply it to real 
systems using limited empirical data. 
Claim 2:

## Page 30

*[Page 30 appears to be blank or image-only]*

## Page 31

MVEE- A Way Forward...
Natural
Selection
Modern
Synthesis
EcoEvo
Devo
1859
1920’s
W1959?
W-E2003?
Lamarck…
Darwin…
1806

## Page 32

EcoEvoDevo
T4?’s EST EcoDevo

## Page 33

EcoEvoDevo
VNE
FEP T4?’s EST EcoDevo

## Page 34

EcoEvoDevo
VNE
FEP T4?’s EST EcoDevo
MVEE

## Page 35

EcoEvoDevo
VNE
MVEE
FEP T4?’s EST EcoDevo
Synergetics
Collective 
Behavior
ML
Hakan
Bucky
Hakan
TFlow

## Page 36

Multilevel Variational Evolutionary Ecology 
(MVEE)
“Variational” in the Darwinian sense refers to “Natural Variation”, for 
example in Lewontin 1983. However “Variational” also is technically used 
to refer to variational (ensemble) Bayesian models, for example in the 
Variational Free Energy Principle. MVEE draws on both these definitions.
MVEE also has another embedded meaning:
Minimum Viable Evolutionary Explanation. 
As it turns out, MVEE is the framework for formulating MVEE’s!

## Page 37

●Is MVEE consistent with prior Biological theory?
○
Yes, I believe MVEE is at worst equivalent to EcoEvoDevo at best.
●Is MVEE a Gauge Theory?
○
Yes, it only deals with observables and follows RBF 2017.
●Can MVEE deal with theoretical (analytical) evolution? 
○
Yes, and may provide atemporal formal solutions to these systems. 
●Can MVEE deal with empirical (open) evolution?
○
Yes, and at least can provide statistically-optimal estimates and 
maximally-informed hypotheses about real biological systems. 
●What is the computational architecture of MVEE?
○
By using a single integrated machine learning framework on pre-existing 
biological data (TensorFlow/compute-graph), MVEE jointly and optimally 
considers coarse-grained state and process models across scales.

## Page 38

G1
E1P1
G2
E2P2
No “Fitness”....
Only measurable things:
G, E, P, and Time
Time

## Page 39

STATE MODELS
PROCESS 
MODELS
Collective Behavior
Decent. Algorithms
time
G1
E1
P1
G2
P2
E2

## Page 40

G1
E1
P1
G2
P2
Use what 
works!
E2
STATE MODELS

## Page 41

G1
E1
P1
Evo. Genetics
G2
E2
P2
Env. Change
Quant. Genetics
Developmental &

## Page 42

G1
G2
Epigenetics
Niche Construction
(plasticity or 
transgenerational)
Phenomenological 
models of DNA 
evolution (GTR)
E1
E2
P1
P2

## Page 43

G
E
P
Development

## Page 44

G1
E1
P1
Development
G2
E2
P2
time
Evolution of Development…..
Within generation = 
“Meta-Plasticity”
Between generations = 
“EcoEvoDevo”

## Page 45

Collective Behavior
Local computation
Predictive/Adaptive
Multilevel Feedback loops
Robustness & Evolvability
Emergence/Self-Organization
What is the Algorithm?
What does each agent perceive? 
What can each agent do? 
How do group outcomes arise?
PROCESS 
MODELS

## Page 46

Formalizing a decentralized process implies 
formalizing the environmental dynamics:
2014
2016
Also see Collective 
Behavior Appendix

## Page 47

Biology ∈ ℝGEP 
Generalized, Dynamic, multiscale
Tensor representation of Any Evolving System. 
Free from “Fitness”!
t 
GEP1
GEP2
GEP3
GEPn

## Page 48

And more...

## Page 49

E = MVEE⊗t
Evolution is (described by)
Multilevel Variational Evolutionary Ecology 
tens⊗rized through time

## Page 50

Summary
In Two Breaths

## Page 51

G
P
F
D
E
E
1.
EED 
is a 
mess...
?

## Page 52

MVEE draws on many sources, especially VNE, 
to potentially overcome some of the fundamental 
weaknesses in current evolutionary studies. 
2.
Biology ∈ ℝGEP 
t 
E = MVEE⊗t

## Page 53

E = MVEE⊗t
&
Biology ∈ ℝGEP t

## Page 54

*[Page 54 appears to be blank or image-only]*

## Page 55

Hankey, Igamberdiev, 
Hu/Petoukhov2, Islami, 
Rosen, Longo…...

## Page 56

Appendix 1
Collective 
Behavior

## Page 57

Appendix 2
Synergetics

## Page 58

Appendix 3
Gauge and
Coarse-Graining.

## Page 59

*[Page 59 appears to be blank or image-only]*

## Page 60

Appendix 1
Collective 
Behavior

## Page 61

Define “Behavior”:
Pragmatically, Pluralistically, 
Algorithmically, Empirically, 
Parsimoniously, etc
A multilevel, 
Coarse-Grained, 
Mechanism-Flexible 
taxonomy is our manifold 
map to effectively navigate 
the multilevel (G-E-P)t 
Tensor construct.

## Page 62

Heredity
Metabolism
Reproduction
Homeostasis
Response to Stimuli
Development
 Adaptation/Evolution
Aperiodic Crystal Structure (S. 1944)
Negentropy Generation (S. 1944)
“Hallmarks of Life”

## Page 63

1992

## Page 64

Heredity
Metabolism
Reproduction
Homeostasis
Response to Stimuli
Development
 Adaptation/Evolution
Aperiodic Crystal Structure (S. 1944)
Negentropy Generation (S. 1944)
“Hallmarks of Life”

## Page 65

“Algorithms of Life!!”
Maintain Information/Material
Process Resources
Inter-Generational Change
Process Information
Intra-Generational Change
Local organization (e.g. Stigmergy)
Physical-Informational Nexus
Heredity
Metabolism
Reproduction
Homeostasis
Response to Stimuli
Development
 Adaptation/Evolution
Aperiodic Crystal Structure (S. 1944)
Negentropy Generation (S. 1944)
Resilience, Robustness, Persistence
Learning & Inference (Implied from above)
“Hallmarks of Life”
Eco
Evo
Devo

## Page 66

The “Algorithms of Life” are coarse-grained 
or simulated algorithmic models, providing 
biologically-plausible manifolds to explore the 
GEP tensor over any time scale….
The compute graph (TensorFlow) 
model allows arbitrary, modular, and 
tractable use of current data/models.

## Page 67

The “Algorithms of Life” are coarse-grained 
or simulated algorithmic models, providing 
biologically-plausible manifolds to explore the 
GEP tensor over any time scale….
In other words,  we use information about the species/time-specific processes to 
preferentially explore functional manifolds that are reachable under informed 
models but implausible under incorrect models (e.g. Parsimony vs. GTR+I+G)......
We know that bone morphogenetic processes and allometric scaling allow us to 
understand biases of morphological evo. (Gould’s Antlers, Wagner, Turing) 
...................We need something like this for collective behavior!

## Page 68

Contemporary
Historical
How?
(Proximate)
Why?
(Ultimate)
Mechanism 
Causal Explanation
Function 
Use or Survival Value 
Development 
Ontogeny, Plasticity
Evolution 
Selection and Drift
TREE, 2013

## Page 69

*[Page 69 appears to be blank or image-only]*

## Page 70

Appendix 2
Synergetics

## Page 71

1975
1979
+ Hakan (1983) :
"Nonequilibrium Nonlinear 
Statistical Physics"
Complexity:
"Self-Organization"
"Chaos and Order"
(Phase transitions, etc…)

## Page 72

RBF’s K-Shell Class of Tensegrity 
Structures exhaustively and 
hierarchically describes structural 
dynamics among K items. 
G
E
P

## Page 73

Time
An MVEE object is a formalized dynamic multiscale Synergetic equilibrium: 
G
E
P
G
E
P
Class 3 + 1 
Tensegrity system 
(open ended Szathmary et al. 2016/2017, Infinite Semiosis/Novelty) 
 
States
Processes

## Page 74

An MVEE object is a formalized dynamic multiscale Synergetic equilibrium: 
F
Class 4 
Tensegrity system 
(closed-form Nowak / Game Theory, limited syntactic Novelty)
(Mechanism under strong closure) 
G
E
P
Class 3 + 1 
Tensegrity system 
(open ended Szathmary et al. 2016/2017, Infinite Semiosis/Novelty) 
 
Formal, timeless relationships
as in Newtonian physics
G
E
P
G
E
P
Time
States
Processes

## Page 75

*[Page 75 appears to be blank or image-only]*

## Page 76

Appendix 3
Gauge and
Coarse-Graining.

## Page 77

If you printed and crystallized the single-stranded DNA of all possible 
sequences of a 100-basepair genome, the required volume would be: 
?

## Page 78

If you printed and crystallized the single-stranded DNA of all possible 
sequences of a 100-basepair genome, the required volume would be 
10 billion times the volume of Jupiter. 
(1.76e+25 km^3 == (1.1e-26 m^3 displaced per 100-bp DNA helix) * (1.6e60 oligomers, representing the 4^100 combinations in sequence space)). 
Every specific genomic configuration is novel. 
The same could probably be said for Phenotype or Environment, which are both 
endlessly dynamic and infinitely describable (cue Borges story…..). So…..
We use abstract lower-dimensional state spaces (manifolds).

## Page 79

Broadly we can consider two types of theory -- State and Process theories:
State = What?
PV = nRT……..E = mc2……...R = h2S…. 
These are State theories because they describe how aspects of the system (measured or calculated) 
are instantaneously related to one another. They are predictive and mechanistically agnostic. 
Process = How? 
Natural Selection….Thermodynamics…..Semiotics….
Theses are Process theories because they describe algorithmic, procedural, or mechanistic 
relationships and becomings. They are not predictive unless coupled to a state theory as well. 
For example, R = h2S is a simple state model following the process model of Darwin 1859.

## Page 80

G
E
P
F
D i m e n s i o n a l        r e d u c t i o n ….
(Coarse-Graining of Data)

## Page 81

“Dirac defines gauge as under-determination 
of the variables’ evolution, and observes 
that….only gauge-invariant quantities can be 
physical, by definition...”
2013
“Gauge”= 
“Measurement”
Rovelli 2013

## Page 82

2013
“Dirac defines gauge as under-determination 
of the variables’ evolution, and observes 
that….only gauge-invariant quantities can be 
physical, by definition...”
Rovelli 2013

## Page 83

Coarse-graining
Mainstream:

## Page 84

Fringe:
2013
2016
Coarse-graining

## Page 85

Coarse-graining
I personally don’t care about philosophical issues of top-down causation but I am bolstered 
by knowing that some people heartily endorse this principle as transcendent. 
I am interested in how we can formalize optimal 
State and Process model/data coarse-graining, 
given our available information about the states 
and processes of the world.

## Page 86

For the scale of one ant colony over one day: 
State: Dynamic model (any type) with colony-, time- and environment-effects, e.g. Prab.2012
Process: an agent-based model like e.g. G.G.2013 + Davidson.2016
For the scale of a population of colonies over a summer:
State:  An ensemble characterization of foraging dynamics using the State model above. 
Calculated on a realized ecological trajectory, or over a distribution of trajectories. 
Process: A parametric or non-parametric descriptor of ensembles of the Process model 
above, allowing for natural variation in colony-specific “sensitivity”, iterated over a summer.
For the scale of populations over intergenerational time: 
State:  Something like Fisher’s Fundamental Theorem or R = h2S (but see Ewens/Lessard 2015 and Plutynski 2006), 
applied to the simulated/actual Phenotype-Fitness mappings from the above State model. 
Process: A life history table showing how colony behavior, colony developmental stage, and 
ecological context influence the likelihood that a colony leaves offspring….AND….. a 
mechanistic neurophysiological model explaining how statistical heritability arises via vertical 
transmission of molecular variation........Make Darwin Proud Again!!! 
Example state/process + coarse-graining for “Pogo Foraging”

## Page 87

2017
2016

## Page 88

*[Page 88 appears to be blank or image-only]*


---
*Extraction method: pymupdf*