Lecture 25

For today

• 1. Write the first part of your final report, explaining the experiment you are replicating.

Today

• 1. Snow Day!

For next time:

1. Read today's notes

2. Do Notebook 12

   1. Read someone else's draft report and provide feedback (see instructions below)

3. Keep working on your project

Optional reading: Dawkins's Introduction to the 30th Anniversary Edition of The Selfish Gene

Draft feedback

You should get an email from me with the pairings for this assignment.

As soon as possible, please contact the person whose draft you are reading to get access to the draft.

At the same time, please schedule a time before our next class to meet and exchange feedback.

The draft should explain the experiment that's being replicated. It should answer the questions listed in the assignment.

Imagine you are trying to replicate the experiment; does the draft report give you enough information to do that? What is missing?

Particle-wave duality

Related to one of the themes of this class --- changing criteria about what makes a good theory --- particle-wave duality is an interesting case study.

In many discussions, duality is presented as something mind-bendingly weird, and a metaphor for all kinds of pseudo-profound ideas.

Frequently-used language includes "Light is both a particle AND a wave, at the same time!"

Or "Light is sometimes a particle and sometimes a wave."

Taking one more step, "Light sometimes behaves like a particle and sometimes like a wave"

Strangely, one seldom hears "Light is neither a (classical) particle, nor a wave, and that's only surprising if you thought it was obligated to be one of them."

An instrumentalist approach to physical modeling provides an interpretation of PWD that is entirely mundane:

• • If you model light as a (classical) particle, you will get things approximately right in some circumstances, and completely wrong in others.

  • If you model it as a wave, you will get things approximately right in some circumstances, and completely wrong in others.

  • And if you model it as (modern) particle, you get good answers in almost all circumstances. But it seems likely that we will find circumstances where that model fails too (if we haven’t already).

In this interpretation, we are making only weak claims about what light is; rather, we are making recommendations for model choice.

For more on this topic, you can read this exchange between Ted Bunn, a friend of mine who teaches physical at U Richmond, and me: https://blog.richmond.edu/phyame thing, different circumstancessicsbunn/2011/04/13/particle-and-wave/

Evolution of Cooperation

We have already discussed the problem of altruism.

And we discussed Axelrod's observations:

1) TFT generally does well in many populations.

2) More generally, successful strategies tend to be nice, retaliatory, and forgiving.

But TFT is a designed strategy competing with other designed strategies.

So that begs the question of whether it, and other nice strategies, are evolvable:

• 1. If they appear by mutation, do they propagate?

  2. If they dominate the population, do they resist invasion?

That's what the experiment in Chapter 12 is meant to address.

People have done experiments like this with lots of genotype-spaces, from

• 1. One-bit: always defect or always cooperate.

  2. Executable code.

I chose something in the middle, where every response is based on what the opponent did in the two previous rounds.

Strategy represented by 7 bits, so there are 128 strategies.

• • Can represent AllC, AllD, TFT, STFT, TTFT, a version of TFTT, etc (see Strategies for IPD)

  • But not Grim, or anything based on randomness or counting.

I wanted TFT to be a possible outcome, but didn't want to cook the experiment.

Results:

• 1. Nice strategies appear by mutation and propagate.

  2. But they are not immune to invasion.

  3. It looks like there is no stable equilibrium (or we don't get to it).

  4. Nevertheless, the long term average is "pretty nice".

  5. There is only a little retaliation (but maybe you only need a little).

  6. There doesn't seem to be a special gravity toward TFT.

Nevertheless, Axelrod's fundamental result holds up: in a wide range of evolutionary scenarios, a population of agents competing in IPD-like contests evolves toward "pretty nice".

How robust is this result? Well, I didn't have to try very hard.

As an exercise in Notebook 12, you will have a chance to extend my experiment and see if the results hold up.