Sunday, January 18, 2009

Old School Artificial Intelligence and Determinism...

I used to work in the MIT Artificial Intelligence Laboratory. Many people there believed in the philosophy of determinism, especially as embodied by Causal Determinism. From their Computer Science vantage point, they thought of the universe as a giant Turing Machine. In this mind set, conclusions about Turing Machines can be applied to the universe at large. This leads to faith that reductionism can solve any problem because once you know the microscopic properties of a system, you can predict any complex behavior. Modern physics is rife with this philosophy. I formerly subscribed to this viewpoint until reading Ernst Mayr, Carl Woese and Stuart Kauffman. Carl Woese's article "A New Biology for a New Century" helped me see the light.

In response to any discussion of the universe as a Turing Machine:

All observations of quantum behavior have so far shown that particle motion at quantum scale is truly stochastic. The physical universe is itself fundamentally non-deterministic. The determinism in finite state or Turing models (infinite or not) is simply an abstraction that exists in the human mind. It is a useful heuristic for making decisions in an environment where the standard deviation of physical parameters is so small that it can be ignored without ill effect.

Technically, even a digital computer is not fully deterministic. At any given clock cycle there is a finite probability that the next state will not be the intended one. Embedded systems designers have been aware of this for a long time, which is why they always include watchdog timers in their designs. Even in a Pentium, the probability of failure is high enough that in continuous operation, it will probably fail once every few years.

No comments:

Post a Comment