This article is fascinating. I had no idea that the Von Neumann machine was an extrapolation of a mental model. Even more fascinating, that the existince of a symbolic computation machine made the possibility of purely symobolic epistimology impossible. It's like they abstracted a lever higher than they wanted to and then made a cleaner, simpler implementation of it. And now with modern AI we are doubling down on that implementation and trying to build a new kind of intelligence no top of it.
It's fascinating that we've built a model of a computer, which was impractical; then we've modeled the brain with very high abstraction and then actually built a computer. Now we're back to modelling the brain because while the abstractions are universal they're not efficient enough.
We need to nitty gritty details on efficient ways of learning and computing approximately ("probably approximately correct"), quickly and efficiently.
Which is what I believe the article tries to convey as the last epiphany of Pitts.
Even Alan Turing, in "On computable numbers, with an application to the Entscheidungsproblem", practically invents turing machines modelling how a mathematician works: He has a pencil, some paper and a number of different states inside his brain.
Everyone was directly motivated at the time to solve Hilbert's Entscheidungsproblem [0] which was about mathematical proof not universal machines. Turing, along with a few other mathematicians, recognized that proofs involved notions of algorithms and computation and all, together, generalized these into notions of computation we have today—Turing Machines, Lambda Calculus, Recursive Functions. So it's not terribly surprising that Turing's model was a human one. It was exactly his goal (originally).
Comments
This article is fascinating. I had no idea that the Von Neumann machine was an extrapolation of a mental model. Even more fascinating, that the existince of a symbolic computation machine made the possibility of purely symobolic epistimology impossible. It's like they abstracted a lever higher than they wanted to and then made a cleaner, simpler implementation of it. And now with modern AI we are doubling down on that implementation and trying to build a new kind of intelligence no top of it.
It's fascinating that we've built a model of a computer, which was impractical; then we've modeled the brain with very high abstraction and then actually built a computer. Now we're back to modelling the brain because while the abstractions are universal they're not efficient enough.
We need to nitty gritty details on efficient ways of learning and computing approximately ("probably approximately correct"), quickly and efficiently.
Which is what I believe the article tries to convey as the last epiphany of Pitts.
Even Alan Turing, in "On computable numbers, with an application to the Entscheidungsproblem", practically invents turing machines modelling how a mathematician works: He has a pencil, some paper and a number of different states inside his brain.
Everyone was directly motivated at the time to solve Hilbert's Entscheidungsproblem [0] which was about mathematical proof not universal machines. Turing, along with a few other mathematicians, recognized that proofs involved notions of algorithms and computation and all, together, generalized these into notions of computation we have today—Turing Machines, Lambda Calculus, Recursive Functions. So it's not terribly surprising that Turing's model was a human one. It was exactly his goal (originally).
[0] http://en.wikipedia.org/wiki/Entscheidungsproblem