Dreamers of dreams

Whatever downers this year has brought, it has been an upper in the science of the mind, thanks to blockbuster proof of the efficacy of deep neural networks. For about half a century, a debate has been under way about the human mind. Is it like a computer? Or just a messy-round-the-edges semblance of such a rational machine?

The New York Times had the story this week in a long-read article by Gideon Lewis-Kraus. The faction who reject the computational view are generally termed connectionists, since they propose that the nuances in the connections joining what we have learned with what we perceive are sufficient to explain thought.

The only way to scientifically prove this is feasible would be to build a synthetic device that works the same way to achieve human-like results. This year, both Google and Baidu succeeded in doing it.

Lewis-Kraus puts this in the context of a stockmarket investment opportunity in artificial intelligence, which is rather like saying the Enlightenment was a historic opportunity to invest in dictionary publishing. What's really happening here is that we are in the midst of developing a new paradigm for understanding ourselves or "what the brain might be up to" as Geoffrey Hinton puts it in this interview.

My research has been built around the hypothesis that humans partly reason with the help of spatial mechanisms in the brain. A diagram (and good layout generally) helps us to make sense of ideas, because it harnesses spatial thought. Like many revolutionary new views of the mind, this does not fit well with the rationalist view of the mind that has risen since the Enlightenment.

We are still immensely far from understanding the mind, but the practical benefits of this year's connectionist experiment make it far less likely that the mind is like a computer, and far more likely that it is an assembly of reasoning effects that simulate pure reason. A neural network cannot shut out irrational deductions, but it could integrate a very mixed bag of inputs.

This may even make us more open to older, pre-Enlightenment ideas such as the classical concept of memory, the western medieval theory of symbols and the idea that we are not natively rational, but learn to be rational. Cognition may not even be limited to one brain, but be distributed across individuals. We are not logical machines. We are dreamers of dreams.

No comments :

Post a Comment