Judges’ Queries and Presenter’s Replies

  • May 21, 2013 | 09:09 a.m.

    just to be sure I follow, do you propose that the use of less-informative signals improve coordination?

  • Icon for: Carlos Santana

    Carlos Santana

    May 21, 2013 | 10:22 a.m.

    No, sorry if that’s unclear. What I’m saying is that despite the fact that less-informative signals run the risk of decreasing coordination, in the right conditions they can improve fitness.

  • May 21, 2013 | 11:37 a.m.

    I’m curious about the role of the imposed costs. I would assume that costs are traded off against other factors in the model. Can you explain more about the extent to which costs play an important role?

  • Icon for: Carlos Santana

    Carlos Santana

    May 21, 2013 | 11:56 a.m.

    The main tradeoff is between the cost of having a complex strategy and the benefit of having signals carry more fine-grained information. In these models, the role cost plays is that, together with the ability to combine multiple sources of information, it allows signalers with less precise signaling profiles to compete.

  • May 21, 2013 | 06:46 p.m.

    I have two questions:

    First, could your please give some examples of cellular or developmental signals that illustrate (1) what ambiguous signals are, (2) what perfect signals are and (3) what “eavesdropping” is?

    Second, in predator-prey relationships, how does the evolution of camouflage relate to your discovery? What would your finding predict and does the prediction fit your understanding? (You can consider either predator camouflage or prey camouflage)

  • Icon for: Carlos Santana

    Carlos Santana

    May 21, 2013 | 08:17 p.m.

    I’m not an expert in cellular or molecular biology, but I can describe in more detail what conditions would have to hold to fit each part of your first question. If you know of specific examples, I’d be delighted to hear them. (2) A perfect signal would be one which was always transmitted in response to roughly the same condition, and which had roughly the same effect upon reception. A hormone with one very specific job to do would fit the bill. (1) An ambiguous cellular signal would be something that was used to trigger a wide variety of process. For example, if there were a hormone or neurotransmitter that had different effects depending on some factor in the receiving cell extrinsic to the signal itself, this would be an ambiguous signal. (3) Eavesdropping is the ability to make use of information from signals that an individual does not itself produce. An example would be a bacterium which alters its gene expression in response to the quorum sensing molecules of other bacteria, but does not produce those particular molecules itself. For the purposes of my model, eavesdropping is important because it serves as a sort of half-way house allowing partial poolers to invade perfect signalers.

    In regards to your second question, my work predicts nothing with regard to predator-prey interaction. I explicitly consider only the case where senders and receivers have common interest, because this is the only case where signaling game models tend to predict perfect signaling. Signals between individuals with significant competing interest are usually discussed in the literature in terms of “costly signaling” or the “Zahavi handicap principle”.

  • Icon for: Mary Gauvain

    Mary Gauvain

    May 21, 2013 | 07:01 p.m.

    Applying this model to human communication, how might intention of the communicator be involved? I am especially interested in how you might consider deception and its detection in this model?

  • Icon for: Carlos Santana

    Carlos Santana

    May 21, 2013 | 07:54 p.m.

    Since this model uses evolutionary game theory, signal meaning in the model is solely a product of selection pressure, and not of intention. Many economists, linguists and philosophers have used economic game theory to analyze the human case, where meaning is partially a function of speaker intention.( A classic discussion of deception and screening out deceivers in economic signaling games is Kreps and Sobel’s “Signaling” (1994).) One conclusion from this literature relevant to my research question is that when the sender and receiver have interest partially at odds imperfect signaling systems are almost guaranteed. If there is too little common interest, then “babbling” equilibria are favored, and all signals are meaningless because no agent can trust the sender.
    For this particular project I chose to look solely at the case where full common interest holds. We could adjust the model by changing the payoff structure so that senders can sometimes benefit from deceiving the receivers, and then use it to explore the evolution of deception and detection. But in such a case the failure to attain perfect signaling is almost trivially guaranteed, so including incentive to deceive would derail the purpose of my simulations, which is to understand why even in cases with little reason to deceive partial pooling emerges.

  • May 21, 2013 | 11:02 p.m.

    Interesting, and well described. I was pleased to see that you ran simulations to address the questions of invasion dynamics and evolutionary timescale. My question is about the relative contribution of both costs and context on your results. Do you have either results or intuitions about their individual effects?

  • Icon for: Carlos Santana

    Carlos Santana

    May 22, 2013 | 12:09 a.m.

    Thanks for giving me the opportunity to say more about the simulations than I was able to fit in the presentation.
    I can’t say much about contextual information except that if you remove it you don’t get the effect, and ambiguous signalers are always invaded by perfect signalers.
    As for cost, I did vary that parameter to see how robust the result was. For the invasion simulations, the rate of invasion by ambiguous signalers correlates with cost, as you would expect, but complete invasion of small populations (1000 individuals) still occurred most of the time in short timescales even with cost an order of magnitude lower than the simulations reported above (.005 vs .03).
    In the Babel simulations (randomized initial population with mutation) the result varied more with cost. The graphs on my poster are for the initial run at cost .03. I ran a simulations at cost .01 and there was very little advantage for ambiguous signalers; if I had graphed those results they would look more like the top right graph than the bottom right. I explored larger costs as well, and by about a cost of .1 almost no populations move towards a perfect signaling attractor.
    I apologize that these remarks are imprecise. I’m travelling right now and don’t have access to my results files.

  • May 22, 2013 | 09:55 p.m.

    Carlos, thank you. Even this is very useful and gives me a clear idea of the dynamics. Thanks again. Well done.

  • Further posting is closed as the competition has ended.

Presentation Discussion

  • May 21, 2013 | 01:02 a.m.

    Cool project, very interesting. I presume this is applicable to any variety of communicative signaling – auditory, visual, chemical? Any thoughts on how the ecological differences of various signal types might manifest themselves evolutionarily?

  • Icon for: Carlos Santana

    Carlos Santana

    May 21, 2013 | 10:32 a.m.

    Good questions. The applicability of this model to a particular target system depends on how much that real-world system deviates from the modeling assumptions and idealizations. Signaling games in general are silent about the signal medium, so they are applicable to auditory, visual, and chemical signals.
    Some of our idealizations, however, would need to be accounted for to increase the model’s aptness to certain phenomena. To take one example of how signal type and ecology would interact with modeling, my models are noiseless, but if a medium is particularly noisy (meaning that receivers have a significant probability of misreading a signal due to ecological interference) we would want to include that in the model. It’s possible that noisy signals could also lead to less than perfect signaling systems.
    Other examples of relevant modeling assumptions and idealizations include asexual reproduction and no preferential sorting among individuals. This might make the model more applicable to organisms fitting those criteria, and the types of signals they tend to use—auditory signals are much less common among species who reproduce asexually and have no social structure, for example.

  • Further posting is closed as the competition has ended.

Icon for: Carlos Santana


University of Pennsylvania
Years in Grad School: 2

Laziness as a communicative virtue: evolutionary game theory and ambiguous signals

In this research I use game-theoretic analysis and individual-based simulations to suggest that communication is often imprecise because ambiguity can be more efficient than clarity. By treating communication as a problem in coordination, evolutionary game theorists have demonstrated how natural selection leads to signal users being understood by other members of their species. Given the ability to make a variety of signals, populations in game-theoretic models will evolve towards using each of those signals to represent a state of the world. In these models, however, the communication systems that emerge are a little too neat: under the most realistic dynamics, simulated populations end up using perfectly precise signaling systems. This is unlike the real world, where communication is almost always imprecise. Theorists have explained this discrepancy between the models and the real world by arguing that imprecision comes from limited intelligence or conflict between communicators. While I accept these as partial explanations of why communication is messy, it seems to me that communication is often ambiguous merely because signal senders can get away with being imprecise without hurting their cooperative aims. To validate this intuition I show how a simple adjustment to classic signaling games can favor the evolution of ambiguity in simulated populations. I add two elements to the model: a cost for more complex signaling strategies, and the ability to combine information in signals with independent information. Analysis and simulation of the altered model shows that it leads to the predicted outcome of evolution favoring ambiguous signaling.