Yes! That’s how the character Greg Burgess learned to write software that could learn, leading to his creation of the self-aware networked AI in the book ‘The Adolescence of P-1′.

I hunted up the book and read it because it was reviewed in Byte magazine back in the early ’80s. A very good read.

Anybody remember what book this was from?

@Mic: Where would you find matchboxes big enough to hold the hundreds of beans representing all possible moves for a go position?

First of all, sorry for being a dick last night. I was a lot harsher than I meant to be, on rereading it. I must have been overtired.

You have a point. Maybe randomness does change the game.

I’m used to dealing with cognition, not computation, and “learn from your mistakes” does seem like a cognitive task.

Since neurons are just tiny analogue computers, I’m tempted to say that anything that’s a cognitive task is also a computation. But, I could be mixing up the definitions. I guess it all depends on whether your theory of computation allows for probabilistic outcomes. (Quantum computing?)

Tangentially relevant quotation:

“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” –Dijkstra

Alex Dodge

As far as I can tell the machine has two functions: “Output random number from 1-N”, and “Store a given value for N at a given address”. If the machine is capable of computation, I’m not sure what that computation would be, since that first function is not computable (and of high entropy)! The second function is just memory.

Thanks for the reply, I enjoyed thinking about this! (Also I still use a slide rule.)

“However… I cannot find a way to make it conditionally branch or copy the contents of one bit of memory to an arbitrary other bit, so my vote is ‘Not a computer’.”

Your definition of computer makes no sense. It performs a computation. It is a computer. It doesn’t have to be Turing complete. A slide rule is a computer.

Alex Dodge

That was my first thought. I had marvin’s book, ‘The hangmans noose….’. He talks about the matchbox bean engine in that book too. I specifically remembering him talking about a kid who built one but one and ate a jelly bean each time the “computer” lost….

I haven’t found a way to do it with two machines playing each other repeatedly either, whatever I run in my head just ends up with both machines ending up with memory contents that repeat with fixed period regardless of initial state.