The Bayesian Brain

October 30, 2008 – 0:35 by Mikko Hämäläinen

Some time ago there was an interesting article in New Scientist, about new theory on how human brain works. It turns out that high level brain functions can be modeled using Bayesian statistics and the brain in fact uses statistical data from prior experiences to predict the current sensory input. Being a layman, this rises a few issues at least for me. I’m not going to quote the article here, please read it first yourself.

The observed reality is only an approximation

If the human brain really does not actually ever have a complete and accurate view on the environment (or reality) as the whole sensory process is based on calculating probability and minimizing error between observation and predicted input.

The brain functions follow causal processes

If prior events effect on how we observe the environment, causality is a fundamental part of how our brain works. If our whole neural network functions via Bayesian statistics, then everything we process, even our thinking, happens in accordance with causal processes – probabilistic causation comes to my mind first. This is pretty much what Russell actually proposes in his theory about causal lines.

The brain can be simulated more easily than was thought

While computers have a hard time crunching massive neural networks, the basic infrastructure of the brain, they are quite efficient in statistical analysis. If you have ever used a spam filter, you actually have Bayesian statistics running in your software that does two things that a human would: 1) it learns along the way b) it predicts using partial or fuzzy data. Therefore it is quite possible to be able to simulate some limited brain functions without millions of CPUs. And as processing power increases, such simulations can be done by anyone or a community of people just like SETI@home. Singularity, anyone?

But what is most fascinating for me is that once again macroscopic structures stem from microscopic ones. Bayesian probability can be applied in quantum mechanics – as QM is mostly about probabilities since we do not have the resolution to really accurately measure full quantum states. And due to Heisenberg uncertainty principle, we can not even if it technically was possible. And as atomic particles and natural forces are causal, and as it seems so is the brain in this theory.

Also one more interesting thing in this theory is that it gives you some advice on how to become an intuitive decision maker. Intuition in this case is to have a lot of alternative statistics available in order to make a quick and most correct estimation of given situation. So you should get rid of tunnel vision and observe a broad set of information even if it is not of direct interest to you or your business. The thing with statistical probabilities is that the correct solution could come from a “wrong” place and it would come spontaneously. That is what intuition is all about.

PS. I did not even go to the free energy principle proposed by Friston. It also underlines that actually our interpretation of quantum physics has it’s counterpart on higher level structures like classical physics that has its higher level counterpart in chemistry which has a counterpart in molecular biology and so forth. Kind of makes me think that in the end, everything is hierarchical following the same rules from Planck scale to infinity.

  1. 2 Responses to “The Bayesian Brain”

  2. Excellent post, thanks!

    By Eero on Nov 3, 2008

  1. 1 Trackback(s)

  2. Dec 16, 2008: SciAm: Blurring the Boundary Between Perception and Memory | mikolas

Post a Comment

mikolas is proudly powered by WordPress Entries (RSS) and Comments (RSS). Theme by Bob