The Emergent Fool

Two Books On Cognitive Science You Should Read

Posted by kevindick in Cognition, Psychology, Science, Society on March 17th, 2009 | View Comments

As I mentioned in this post, one of the three primary planks of my worldview is that, “…the human brain is a woefully inadequate decision making substrate.” I started adopting this posture in graduate school and have refined it with constant input from the cognitive psychology and neurobiology literature over the years. Luckily, you don’t have to put in that kind of time. Simply go out and read Rational Choice in an Uncertain Worlds by Hastie and Dawes and The Accidental Mind by Linden.

The former is an overview of all the logical mistakes the mind makes when trying to reach a decision. Khaneman and Tversky’s Choices, Values, and Frames is considered more seminal (and Tversky was one of my favorite professors in graduate school), but Hastie and Dawes is both more approachable and more complete in my view.

The later is an overview of how the brain is put together and operates at the biological level. There are a couple of really, really dry chapters on the biochemistry of never signal propagation that you just have to get through. But the rest of it is pretty enjoyable.

If you read these books, you’ll undertsand why I’m very skeptical of “trusting my instincts” in any situation that isn’t a fairly close parallel to something encountered in the ancestral environment. However, this knowledge has also made me optimistic in a weird way. Given the micro-level capabilities of our brains, it seems like we shouldn’t be able to get very much done, but our civilization is actually quite remarkable. So the whole is substantially greater than the sum of its parts. There must be something in the dynamics of society that allows us to overcome, in some haphazard way, our individual cognitive limitations.

Related posts:

kevindick

Nah, it's probably not any better. It's just The One that everyone in the field has read. Like I said, if you're interested in the information, Hastie and Dawes is the best use of your time. But if you're interested in signaling that you're in the field, you should also read C, V, & F. I'm not particularly interested in signaling for this but thought I should warn people.
Daniel

I've only just started "The Accidental Mind" but it immediately strikes me as the best general overview (from a biological perspective) on the brain I've read. Linden knows what's up.

Kahneman and Tversky's work is certainly seminal, but do you think that Choices, Values, Frames, is any better than their other edited collections of essays?
rafefurst

@Alex, not sure what you mean by endgame being neural network type scientists, can you please explain?

You can deal with probabilistic inputs by running the simulations over again from the beginning numerous times with different initial conditions. Some forms of simulation also have randomness as continuous input (part of the structure of the simulation), for example, agents react in a probabilistic manner, not deterministic. Does this answer your question or are you talking about something else?

As for predictability, one thing we need to accept about the world is that it's not as predictable (even in principle) as we'd like it to be or believe it to be. Quantum physicists have accepted this in their field of study, but complexity arguments can show this for other real-world systems as well.
kevindick

OK Rafe, then we're on the same page. I agree that a single distribution on a complex outcome loses a lot of the interesting dynamics.

Though, it is convenient for summarizing that outcome and reasoning about it further given computational limitations.
Alex Golubev

Ok, i read up on preferential attachment and small world networks. I find it interesting that you think we need to give up completeness and certainty. So an "everchanging hypothesis" in a sense. It's hard for me to understand, because i'm not very familiar with complex system simulations and how they're set up, but it sounds like the endgame are neural network type "scientists". Also, i'm not sure how we can have probabilistic inputs when the probability input itself might change throughout the simulation. is that allowed/possible in the simulations you're suggesting? And it sounds like we WOULD actually continue to spin our wheels until scientists start adopting simulation sof complex systems. Unfortunately, the theory cycle from discovery to failure takes years and sometimes decades so i really hope this methodology is adopted by the scientific community soon. this is a great opportunity for econ and finance. (unfortunately i still think we'll be spinning our wheels if any process is dynamically random. no? isn't it possible that certain things in the world are caused by random butterflies flapping their wings, which allows for very little predictability)
rafefurst

I don't think we really disagree here. I think all reasoning about complex systems should be probabilistic (Bayesian) not deterministic, and most structural claims should be probabilistic as well (e.g. when saying that two nodes are "linked" or two agents "interact", these should be qualified probabilistically).

The problem I have is with claims that a single distribution can adequately characterize a very complex system. Sure, you can *observe* that wealth follows a power law distribution, but this tells you very little about the underlying dynamics and causal factors, and social scientists often will extrapolate all sorts of wacky claims about a system based on an observed distribution. Better to simulate the system, see if you get the same distribution -- your first clue that the simulation is capturing something meaningful -- and let the simulation do the extrapolation for you.
kevindick

I agree with most of what Rafe said. Except for probabilistic methodologies becoming archaic. His and my conceptions of probability are rather different.

In fact, to do complex simulations correctly, I contend that you should use probabilistic reasoning. Part of your simulation should be sampling from various probability distributions to determine the behavior/characteristics of the simulation elements. Then you should use probabilistic analysis to examine the simulation results.

In fact, if one were to try and sum up the admittedly multidimensional space of Rafe's and my philosophical differences on causal worldview I think it would be that I think probability is more important than complexity while Rafe thinks the reverse.
rafefurst

1. I do believe what "the public knows" or believes feeds back in a meaningful way to the system dynamics, and partly because of this it's a moving target. There are however some distributions that are more realistic than Gaussian for Econ/Finance, such as Pareto (i.e. power law) The underlying reason can be traced to preferential attachment, which leads to small world networks, whose link structure follows a power law. That said, any distribution is a massive simplification and obfuscates quite a bit of relevant information about the system. And as systems get more complex, looking at distributions becomes less meaningful. Therefore if the study of economics/finance remains rooted in symbolic and probabilistic methodologies (as opposed to, say, simulation) it will become more and more archaic, regardless of what distributions they are using.

2. I agree that that the scientific method as it's currently practiced is incomplete and hopeless when it comes to understanding systems as complex as the current economy. But it can be fixed. What's missing is the emergent and multicausal perspectives to complement the reductionist. You can still make testable, falsifiable predictions in a complex systems methodology, but you have to give up on completeness and certainty. The number one thing that could change science for the better is if simulation were accepted as more important than mathematical formulae in "proofs" when it comes to most systems of study.
Alex Golubev

Rafe, i read your comment on the Money Illusion regarding the gaussian distribution becoming highly outdated. This is somethign i was thinking about just the other day. Couple of ideas: 1. Black Swan has become a household term last year. Econ/Finance used to assume a normal distribution for practicually all analysis. Now we (the public) know it doesn't apply to most phenomena outside of nature. Will this cause another distribution shape change? by what mechanism? is it oscillating or always increasing? (i haven't really made much progress on any of these to note here). 2. This is a more general scientific method idea. A hypothesis has to written in a form that is testable and can be proven wrong. What if the natural phenomenon is either RANDOM or CHANGING (differential equations). Is scientific method helpless here? what if soft studies like economics and sociology are just spinning their wheels discovering somethign that almost instantly becomes obsolete as soon as the general public becomes aware of the idea? is the study more like a religion then? or can we celeberate the fact that we've change the society for the better... or worse by discovering something about our behavior that we then changed. Seems like you might have something to say about both ideas. I would be very interested.