main
post story
rss feed
search
links
link to us
topics
about
main
|
 Posted by Aelhswith on Sunday August 21, @06:47PM
from the dept.
[article] Traditional measures of human intelligence would often be inappropriate for systems that have senses, environments, and cognitive capacities very different from our own. So Shane Legg and Marcus Hutter at the Swiss Institute for Artificial Intelligence in Manno-Lugano, have drafted an idea for an alternative test which will allow the intelligence of vision systems, robots, natural-language processing programs or trading agents to be compared and contrasted despite their broad and disparate functions.
http://www.newscientist.com/article.ns?id=dn7842
<
|
>
|
|
The Fine Print: The following comments
are owned by whoever posted them.
( Reply )
|
Re: IQ test for AI devices
by Shane Legg on Saturday August 27, @09:13AM
|
If anybody has any questions about this I'd be happy to answer them...
|
[
Reply to this ]
|
Re: IQ test for AI devices
by Rob Sperry on Monday September 05, @11:39PM
|
How long do you estimate before an initial test can be created?
Do you still plan on calling it Shiva?
Rob Sperry
|
[
Reply to this ]
|
Re: IQ test for AI devices
by Shane Legg on Saturday September 10, @04:03PM
|
I'd forgotten about the Shiva thing. In fact I don't really even remember exactly what I said. It was a long time ago now that I was talking to you about all this stuff! 2002?
Anyway, there are a few problems that need to be sorted out before an initial test can be created. The main problem is the choice of reference universal Turing machine. The invaiance theorem for Levin complexity means that the complexity of an environment only depends on the choice of reference machine up to some constant. For some applications this constant doesn't really matter, for example in Solomonoff induction. However here we are taking the 2^-Kt(x) and so this constant, however small, makes a big difference.
Normal IQ tests have the same problem: How to you combine an individual's performance in math and languages and ... into one score? How to you weight these different areas? There isn't an obviously "right" answer to this question. Unfortunately, the invariance afforded to us by Levin complexity also isn't strong enough to solve this problem.
What this all means is that we still need to define our choice of reference Turing machine. Or to put it rather informally, the problem of "what is intelligence?" and just turned into "what is complexity?", which might not be an easier question to answer.
I have a few ideas about how I might try to do this, but they aren't all that nice and clean unfortunately. I suspect that for higher levels of complexity the difference might not matter so much. Certainly with a very high level of universal intelligence for relative simple environments performance will be good for any reasonable choice of reference machine.
Also, perhaps intelligence as a function over possible environments might still be sufficiently smooth to be useful for things like artificial evolution.
|
[
Reply to this ]
|
Re: IQ test for AI devices
by Dennis Dupuis on Monday May 29, @11:12AM
|
Hi Shane and Marcus,
At Roboprize.com, we are constantly trying to evaluate whether or not an android model can
really think like a human, and I am looking
at a video right now, that seems to answer IQ test questions correctly, so I need to know if
a simple program can be written that "seems to
be passing an IQ test, but it is not really "thinking" like a human?
For example, how hard would it be to write a program that could answer sample IQ test like
"Some apples are red, but no apples are square,
so are there any green apples?"
or
"Larry is a brown haired man, and no brown haired men are Chinese. Is Larry Chinese?
I can't disclose more info yet, but maybe you can answer my general question as stated above?
Thanks ...
Dennis Dupuis, CEO
http://www.RoboPrize.com
(now at 6 Million Dollars)
|
[
Reply to this ]
|
|
Re: IQ test for AI devices
by questsin on Saturday December 24, @08:32PM
|
What about the opposite. An IQ test by an AI. take a look at questsin.net IQ Test. While you take the test, you get a relative score and the AI might learn a thing or two. Both win. Who knows maybe this AI would take the AI IQ test when its ready.
|
[
Reply to this ]
|
Re: IQ test for AI devices
by ts on Thursday February 23, @01:12PM
|
This is very interesting and useful. Where is the actual test to be found?
|
[
Reply to this ]
|
The Fine Print: The following
comments are owned by whoever posted them.
( Reply )
|
|
IQ test for AI devices
