Arend Hintze, Department of Integrative Biology, Department of Computer Science and Engineering, and BEACON Center for the Study of Evolution in Action at Michigan State University, presents his computational analysis of evolutionary processes at the Michigan State University Bioeconomy Institute on 10-12-16.

1. Taking an ancient path toward Artificial Intelligence
Arend Hintze
Department of Integrative Biology
Department of Computer Science and Engineering
BEACON Center for studying evolution in action
Michigan State University

2. The idea of computation
Whitehead
&
Russell
~1910
2a = b
true true
“rule”
a = b/2
divide by 2
true
true
true
axiom
formal process of proof:

3. Principia Mathematica
axiom
true
true
true
true
true
true
true
true
true
Kurt Gödel:
I am true, but not provable

4. Turing and the Enigma
“XTYENSODNEADF…“Attack at dawn…”
“Need refueling soon”
“TEHDIENENEEE”
possible mappings:

5. universal computation
every possible
mapping
all possible mappings:
“AAAAAAA”
“AAAAAAB”
…
“ZZZZZZZY”
“ZZZZZZZZ”
Any:
“AAAAAAA”
“AAAAAAB”
…
“ZZZZZZZY”
“ZZZZZZZZ”
Any:
Konrad Zuse

6. your brain is just a process …
find:

7. the 60s, 70s, and 80s (and maybe the 90s)
machine
intelligence

8. AI after 2001…

9. software engineering
how to find:
reverse engineering!

10. expert system vs. artificial intelligence (that I want)
• specific goal
• well defined
boundary cases
• single purpose
• very vague goal
• no boundary case
• multi purpose
• does change during its
lifetime
vs

11. engineering vs. evolution

12. What to evolve?

13. What to evolve?
grammar
L system
code ANN genetic
programming
Markov
Model
-unpredictable
mutational
effect
range
-almost all
mutations
deleterious
-exploding
number
of weights
to optimize
-fixed topology
-over growth
-hard
to train
-no bound
on hidden
states
-Baum Welch
Fix:
evolvable code
Fix:
NEAT
Fix:
pruning

14. Probabilistic Logic Gates
1
1
1
1
1
90% 1
10% 0
FUZZY MORE MODULAR

15. evolvable encoding

16. does it scale?
282 (302)
~100,000
~250,000
~1,000,000
~71,000,000
~86,000,000,000

17. one Markov Gate is a group of neurons
≠=
MARKOV GATE

18. Darwin vs. DARPA
Synapse ~97% accuracy
250.000 neurons
~1.000.000.000 transistors
dedicated neuromorphic chip
Markov Brains ~93%-96%
accuracy
~1000 deterministic gates
hardware from late 70s
~4500 transistors intel 8088
Dave Knoester Samuel Chapman

19. neuron vs. Markov Gate
neuromorphic chip Markov Brains
bio neuron 1:1 1:100
neuron 250.000 gates 1000
transistors 1.000.000.000 4000
n:t 1:4000 1:40
goal 344.000.000.000.000 344.000.000.000

20. 10,000,000,000
100,000,000,000
1.000,000,000,000
10.000,000,000,000
100.000,000,000,000
2016 2024 2032 2040 2048
~350.000.000.000
~350.000.000.000.000
US debt $17,872,947,693,177
US debt $19,688,773,606,117

21. one question answered:
What to evolve?
one to go:
How to evolve it?

22. How to evolve? How did it evolve?
Evolutionary Biology
Cognitive Science
Psychology
Behavior Biology
Evolutionary Neuroscience

23. How to evolve something?
R

24. How to evolve something?

25. population n=100
evaluate
performance best
performance
worst
performance
highest chance
to make offspring
lowest chance
to make offspring
next generation
mutation

26. after 2000 generations

27. after 14.000 generations

28. finally after 49.000 generations

29. How to do Biology with this?

30. an evolutionary experiment
predation response, foraging, mating ….

31. 2d swarming agent
and prey agents are log
layer includes 12 sensors
of 158 and a range of 100
dators). Moreover, each la
agent. Specifically, the p
that is only capable of s
have a dual-layer retina,
fics and the other senses
single predator active du
a predator-sensing retina
Regardless of the nu
slice, the agents only kn
that slice, but not how
coarse-grain visual syste
starlings [42]. For exam
slice has two prey in it
for that slice activates. S
left has both a predator
grey triangle) agent in it, s
retina
Markov
network
L R
01
0
0
0
0
0
0
0
1
1
1
0
1
0
0
0
0
0
1
0
1
0
0
Figure 1. An illustration of the predator and prey agents in the model. Light
grey triangles are prey agents and the dark grey triangle is a predator agent.
on October 4, 2rsif.royalsocietypublishing.orgDownloaded from
Randal Olson

32. normal predator confusable predator
R. S. Olson, A. Hintze, F. C. Dyer, D. B. Knoester, and C. Adami,
Predator confusion is sufficient to evolve swarming behavior (2013)
Journal of the Royal Society Interface 10: 20130305.
plus many more ….
evolution of swarming as a predation response
requires a confusable predator

33. where to go from here?
memory
entrainment
learning
neuro correlates:
IIT, phi, Representations
AI in computer games
complex
environmentshuman decision making

34. Memory, learning, entrainment
L R F N
A x
B x
C x
D x
L R F N
A x
B x
C x
D x

35. Hierarchies and organized groups
division of labor dominance hierarchy
Evolution of autonomous hierarchy formation and maintenance
A. Hintze, M. Miromeni, ALIFE 14: Conference Proceedings, 2014

36. multi agent environments

37. psychology
rational decision maker evolved decision makervs.
A. Hintze, R. Olson, C. Adami, R. Hertwig, Risk sensitivity as an evolutionary adaptation. Scientific reports 2014
P. Kvam, J. Cesario, J. Schossau, H. Eisthen, A. Hintze, Computational evolution of decision-making strategies
Cognitive Science Conference Proceedings 2015
A. Hintze, N. Phillips, R. Hertwig, The Janus face of Darwinian competition, Scientific reports 2015
A. Hintze, R. Hertwig, The Evolution of Generosity in the Ultimatum Game , Scientific Reports 2016

38. value judgement task: Heuristic vs. Optimal Strategy
generations
performance
no strategy
heuristic
optimal strategy

39. more virtual environments

40. AI in computer games

41. the future?

42. Thank you!
Laura
Smale
Barbara
Lundrigan
Fred Dyer
Tim Pleskac Ralph Hertwig
Kay Holekamp Heather Eisthen
Chris Adami

43. Modular Agent Based Evolution Framework
https://github.com/ahnt/MABE