from democritus to signaling networks

60
from Democritus to Signaling Networks

Upload: others

Post on 18-Apr-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: from Democritus to Signaling Networks

from Democritus toSignaling Networks

Page 2: from Democritus to Signaling Networks

The word is the shadow of the deed.- Democritus

Page 3: from Democritus to Signaling Networks

DemocritusThe Laughing Philosopher

400BC

Democritus Laughing, by Hendrick ter Brugghen, 1628

Proclus states that Pythagoras and Epicurus agree with Cratylus, but Democritus and Aristotle agree with Hermogenes, the former that names arise by nature, the latter that they arise by chance.

Page 4: from Democritus to Signaling Networks

�In that gathering of men, at a time when utterance of sound was purely individual, from daily habits they fixed on articulate words just as they happened to come; then, from indicating by name things in common use, the result was that in this chance way they began to talk, and thus originated conversation with one another.�

VitruviusThe Ten Books of Architecture Bk2, Ch1

27 BC

Page 5: from Democritus to Signaling Networks

1. Can meaningful communication arise spontaneously?

Page 6: from Democritus to Signaling Networks

1. Can meaningful communication arise spontaneously?

2. Results robust?

Page 7: from Democritus to Signaling Networks

1. Can meaningful communication arise spontaneously?

2. Results robust?

3. Can the Game Evolve?

Page 8: from Democritus to Signaling Networks

1. Can meaningful communication arise spontaneously?

2. Results robust?

3. Can the Game Evolve?

4. Signaling on Networks

Page 9: from Democritus to Signaling Networks

1. Can meaningful communication arise spontaneously?

2. Results robust?

3. Can the Game Evolve?

4. Signaling on Networks

5. Can the Network Evolve?

Page 10: from Democritus to Signaling Networks
Page 11: from Democritus to Signaling Networks

The Simplest Lewis Signaling Game

Nature flips a fair coin to choose state 1 or 2.

Sender observes the state & sends signal A or B.

Receiver observes the signal and guesses the state.

Correct guess pays off 1 to both; otherwise nothing.

Page 12: from Democritus to Signaling Networks

1. EMERGENCE of MEANINGFUL SIGNALS

Page 13: from Democritus to Signaling Networks

Meaningful?

Page 14: from Democritus to Signaling Networks

“In the beginning was information.The word came later.”

Fred DretskeKnowledge and the Flow of Information

Information

Page 15: from Democritus to Signaling Networks

Emergence?

Dynamics

Page 16: from Democritus to Signaling Networks

Herrnstein-Roth-Erev

Reinforcement Learning

Probability of a specific choice is proportional to accumulated rewards from that choice in the past.

Herrnstein, R. J. �On the Law of Effect.� Journal of the Experimental Analysis of Behavior 13: 243-266, 1970.

Roth and Erev GEB 1995. Erev and Roth AER 1998.

Page 17: from Democritus to Signaling Networks

Simulation

Page 18: from Democritus to Signaling Networks

Theorem

Convergence to a signaling system with probability 1.

Argiento, Pemantle, Skyrms and Volkov. Stochastic Proc. Appl. (2009).

(Parallel results in Evolutionary Dynamics)

Page 19: from Democritus to Signaling Networks

Experiments

Subjects in an economics lab learn to signal spontaneously.

Blume, DeJong, Kim, Sprinkle (1998) Experimental evidence on the evolution of meaning of messages in sender-receiver games. Am. Econ. Rev.

Blume, DeJong, Neumann, Savin (2002) Learning and communicationin sender-receiver games: and econometric investigation. J. Appl. Econ.

Page 20: from Democritus to Signaling Networks

2. Robust?

Page 21: from Democritus to Signaling Networks

Nature’s die is not fair

N states, N signals, N acts

N states, M signals, N acts

Page 22: from Democritus to Signaling Networks

Nature’s die is not fair- pooling equilibria may be stable

N states, N signals, N acts- partial pooling equilibria (some stable)

N states, M signals, N acts- signaling systems may not be conventions

Donaldson, Lachmann, Bergstrom (2007) Journal of Theoretical Biology

Page 23: from Democritus to Signaling Networks

Here Dynamic Analysis is more nuanced

Efficient signaling can emerge, but is not guaranteed.

Hu, Skyrms, Tarres “Reinforcement Learning in a Signaling Game”

Hofbauer and Huttegger “Feasibility of Communication in Binary Signaling Games” JTB (2008) [Also 3by3by3 Games (2015)]

Huttegger, Skyrms, Tarres and Wagner “Some Dynamics of Signaling Games” PNAS (2014)

Page 24: from Democritus to Signaling Networks

Experiments Track Theory

Brunner, O.Connor, Rubin, Huttegger “David Lewis in the Lab – Experimental Results on the Emergence of Meaning” Synthese (2014/2018)

Efficient signaling can emerge, but is not guaranteed.

Page 25: from Democritus to Signaling Networks

3. Can the game itself evolve?

Page 26: from Democritus to Signaling Networks

Inventing New Signals

• Alexander, Skyrms, Zabell (2011) Dynamic Games and Applications.

Page 27: from Democritus to Signaling Networks

Previous Urn Model + “Mutator”

1. No New Signal Tried2. New Signal Tried, but unsuccessful.3. New signal tried with success.

Page 28: from Democritus to Signaling Networks

Learning to Signal with Invention

3 by 3 by 3 states equiprobable

2 by 2by 2 probabilities probability .9

Page 29: from Democritus to Signaling Networks

Self-Assembling Games

Barrett & Skyrms BJPS 2017

Page 30: from Democritus to Signaling Networks

4. Signaling on Networks

Page 31: from Democritus to Signaling Networks

Lattice

• Kevin Zollman “Talking to Neighbors” PhilSci (2005)

Page 32: from Democritus to Signaling Networks

Regional Meaning on a Lattice

Also Pre-play signaling in Stag Hunt Games promotes Coopertion

Page 33: from Democritus to Signaling Networks

Small World Networks

Elliott Wagner “Communication and structured correlation.” Erkenntnis (2009)

Page 34: from Democritus to Signaling Networks

Signaling Chains 1

“Signalling chains with probe and adjust learning”

- Giorgio Gosti

Connection Science 2017

Page 35: from Democritus to Signaling Networks

Signaling Chains 2

“Broken Telephone: An Analysis of a Reinforcement Process”

-Johnathan Kariv

PhD Thesis MathematicsPenn

Page 36: from Democritus to Signaling Networks

Dynamic Networks

Page 37: from Democritus to Signaling Networks

5. Can the Network Itself Evolve?

Reinforcement:

Skyrms and Pemantle (2000) “A Dynamic Model of Social Network Formation” PNAS

Page 38: from Democritus to Signaling Networks

A RINGBala and Goyal (2000) “A Non-cooperative Model of Network Formation” Econometrica(Best Response with Inertia)

Skyrms and Huttegger (2013) “Emergence of a Signaling Network with Probe and Adjust” in Cooperation and its Evolution.

Huttegger , Skyrms and Zollman (2014) “Probe and Adjust in Information Transfer” Erkenntnis

Page 39: from Democritus to Signaling Networks

Stars, etc.

“Self-Assembling Networks” Barrett, Skyrms, Mosheni BJPS forth.

Page 40: from Democritus to Signaling Networks

There is lots more to do.

Page 41: from Democritus to Signaling Networks

Thank you.

Page 42: from Democritus to Signaling Networks

Beyond Common Interest

Page 43: from Democritus to Signaling Networks
Page 44: from Democritus to Signaling Networks

Opposed InterestsA1 A2 A3

S1 -1, 1 .5, -.5 1, -1

S2 1,-1 -1, 1 .5, -.5

S3 .5, -.5 1,-1 -1, 1

Page 45: from Democritus to Signaling Networks

Chaos(structurally stable)

Wagner BJPS 2012, Sato Akiyama, Farmer, PNAS 2002.

Page 46: from Democritus to Signaling Networks

Mixed Interests with Differential Signaling Costs

Cycles also occur here in a non-trivial way in:Spence Signaling Game

- Noldeke &Samuelson J. Econ. Th. (1997)- Wagner Games (2013)

Sir Philip Sydney Game - Huttegger & Zollman Proc.Roy.Soc.(2010)

- But signaling system Equilibria are also possible.

Page 47: from Democritus to Signaling Networks

Information Transfer

may emerge spontaneously withoutcommon interest

in and out of equilibrium

Page 48: from Democritus to Signaling Networks

Cycles around Hybrid Equilibrium

Page 49: from Democritus to Signaling Networks

DeceptionSearcy and Nowicki The Evolution of Animal Signals: Reliability and Deception in Signaling SystemsHarper and Maynard Smith Animal SignalsSkyrms SignalsGodfrey-Smith review of SignalsBergstrom “Dealing with Deception in Biology”Fallis "Skyrms on the Possibility of Universal Deception", Philosophical Studies, (forthcoming)McWhirter "Behavioral Deception and Formal Models of Communication" in the British Journal for the Philosophy of Science (forthcoming)

Page 50: from Democritus to Signaling Networks

Propositional Content

Birch “Propositional Content in Signaling Systems” Philosophical Studies.

Godfrey-Smith et.all.

Special Journal Issue

Page 51: from Democritus to Signaling Networks

Costless Pre-Play Signaling

Robson (1990) “Efficiency in Evolutionary Games” Journal of Evolutionary Biology.

Skyrms (2002) “Signals, Evolution and the Explanatory Value of Transient Information” Philosophy of Science

Santos, Pacheco, Skyrms (2011) “Co-evolution of pre-play signaling and cooperation” Journal of Theoretical Biology

Page 52: from Democritus to Signaling Networks

Perturbation: mutation

Pooling equilibria collapse to a single point.Is it dynamically unstable, stable, strongly stable?

It depends. (Hofbauer and Huttegger JTB 2008).

If

a sink, otherwise a saddle. (for small mutation rates).

Page 53: from Democritus to Signaling Networks

Synonyms and Bottlenecks: Simulations

N Partial Pooling

3 9.6 %

4 21.9 %

8 59.4 %

Barrett Phil. Sci. 2007

Page 54: from Democritus to Signaling Networks

Transient Information in Pre-play Signaling: Stag Hunt

-- Skyrms PhiSci 2002

Page 55: from Democritus to Signaling Networks

Structural Stability

A Dynamics (given by a vector field) is

Structurally Unstable

if an arbitrarily small change in the vector field yields a qualitatively different dynamics.

Page 56: from Democritus to Signaling Networks

Arbitrarily small difference?

At each point in the simplex, for each component, there is a numerical difference. Take the maximum.

Take the least upper bound of these numbers.

This is the distance between the vector fields.

Page 57: from Democritus to Signaling Networks

Qualitatively Different?

Two vector fields are qualitatively the same, i.e.(topologically equivalent) if there is homeomorphism of the simplex to itself that takes the orbits of one into the orbits of the other (preserving sense of the orbits).

Page 58: from Democritus to Signaling Networks

Selection-Mutation Dynamics

Hofbauer (1985) J. Math. Bio.

Page 59: from Democritus to Signaling Networks

Functional Analysis

Aczél, J.; Daróczy, Z. On Measures of Information and Their Characterizations

Page 60: from Democritus to Signaling Networks

Complex Signals• Franke (2013) “Compositionality from Reinforcement

Learning” Proc. G.I.R.L.• Barrett (2009) “Evolution of Coding in Signaling Games” Th. &

Dec.• Steinert-Threlkeld (forth.) “Compositional Signaling in a

Complex World” JLLI.• Nowak and Krakauer (1999) “The Evolution of Language”

PNAS.• Batali (2002) “Negation and Acquisition of Recursive

Grammars…” in Briscoe (ed.)• Barrett, Skyrms & Cochran (2018) “Hierarchinal Models …”

ms.