[ExI] AI Motivation revisited

[Eugen Leitl]

On Tue, Jun 28, 2011 at 07:40:53PM -0600, Kelly Anderson wrote:

> -Define intelligence very strictly.

Half of all job slots across the world occupied by nonhumans.

> -What is the minimum capacity required to achieve that?

Dunno about minimum, but 10^3 refreshes/s over 10^15 sites (~kBit)
(10^21 fat ops/s, including according communication crossection)
could probably do it. 

An off-shelf PC can encode some 10^6 sites, at maybe 10^-3 refreshes/s.
 
> The computational power required to defeat Kasparov in '96 is many
> orders of magnitude above the computational power required to play
> chess at that same level now. Who's to say that once we understand

Me.

> intelligence, we couldn't get some form of it to run on a lowly PC?

You can't understand intelligence by what is commonly considered
"understanding". This doesn't mean you can't recreate intelligence or
improve upon intelligence. 
 
> We will certainly have the first intelligence on supercomputers. Then

It's likely.

> we'll learn about it, then optimize it. After a while, who is to say

Let's say I'll give you full inspection and modification powers
to the hardware in your head. To what degree do you expect to
"understand" it and to do "optimizations" of it other than tweaking
fundamental parameters of the substrate?

> that we couldn't have some kinds of intelligence on a contemporary PC?
> But I don't think we know enough to determine that now.

I do think we do.
 
> If you look at the human brain, a lot of it processes sensory input,
> motor output, and other autonomic activities that don't have much to
> do with intelligence per se. Turing's view of intelligence, for

Intelligence is the sum of such activities. In order to compete
with humans across the board you need all of it.

> example, involves zero visual processing. So there is much less

Turing is dead. Wrong.

> computation required to achieve "intelligence" as defined by Turing
> than there is in the typical skull. Yet, an intelligence that could
> pass the Turing test would be completely incapable of distinguishing
> between a picture of a cat or a dog.

Turing's test is distinctly dated. Consider my metric above.
 
> I wonder if any of the Turing challengers ever sent ASCII pics to the
> contestants? :-) That might buy us ten more years before the computers
> win!

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE