Re: Maximizing vs proving friendliness

From: Tim Freeman (
Date: Thu May 01 2008 - 20:44:44 MDT

From: Matt Mahoney <>
>I believe your algorithm is correct.

Wow, that's a first. Thanks for digging through it.

>It looks like your program is based on AIXI^tl: enumerate all utility
>functions on time bounded Turing machines of increasing complexity
>until a machine is found that predicts a training set of observed
>human behavior.

More or less. The main differences are:

I threw out some speed optimizations in AIXI^tl to get something
simpler and slower.

I use the speed prior to get something with fewer parameters to tweak.

Then it's working on a more complex problem than AIXI^tl so it grew
new and different complexity and parameters to tweak.

>Have you estimated its run time?

If "atrocious" is an estimate, then yes, otherwise no. Much worse
than AIXI^tl, if given a comparable problem. Hutter can say something
interesting about the runtime of AIXI^tl because he have a precomputed
table and a theorem proving pass that I left out.

>Earlier I estimated the human genome complexity is bounded by 10^7
>bits. If we assume that 10% of it encodes the brain, then the
>complexity of the "hardcoded" algorithm (complexity at birth) is 10^6
>bits. Your program also has to guess the laws of physics (409 bits*)
>as well as human behavior, so you need to test 2^1000409 choices.

So we agree on the "atrocious" part. :-).

>If you had enough computing power to run your algorithm, then you would
>have solved AI long ago.

Agreed, it's more like a specification for an AI rather than an AI,
since the computer won't ever be available.

>A faster algorithm (only 2^818 steps*) would
>be to enumerate all laws of physics until a universe supporting
>intelligent life is found.

Where's the connection with humanity in that scheme? You might get a
connection with some imagined form of intelligent life, but there's no
reason to expect us to be like them.

I don't know how to write code that even in principle could test for
the presence of intelligent life in a simulated universe, so I
wouldn't know how to get started with that.

Tim Freeman      

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:02 MDT