From: Matt Mahoney (firstname.lastname@example.org)
Date: Thu Jun 19 2008 - 07:55:12 MDT
--- On Wed, 6/18/08, Stuart Armstrong <email@example.com> wrote:
> But I can design "IQ" tests that can differentiate even advanced AI's.
> Something along the lines of "run a command economy" to maximise some
> result. Run and win an election campaign (against each other). Build a
> living copy of a certain human being from scratch, for the least cost.
> Find a shorter proof of the Shimura-Tanyama conjecture, using only
> basic symbols. Create a blockbuster movie, more sucessful than your
> competitor's (and any other). Be the first to build a reproduction of
> Manhattan on one of Uranus's moons. Select among all living humans,
> the six hundred who would self-organise into the most successful
> mini-economy if they were all abandonned on a desert island.
These are good examples, but these are not examples of agents evaluating the intelligence of more intelligent agents. In these examples, the intelligence is judged by a large group of humans which are collectively more intelligent than any individual. For example, I would probably find a shorter proof of Shimura-Tanyama to be incomprehensible without the work of thousands of mathematicians before me. Likewise, the success of an election campaign or movie can only be judged by millions of voters or movie goers.
One can argue that there are intelligence tests where it is easy to check the result. For example, factoring the product of two randomly chosen 1000 digit prime numbers, or finding a shorter route through 1000 cities, or a shorter program that outputs a particular snapshot of Wikipedia. We strongly believe that such problems are hard, but we have no proof.
We assume that a child who reads at age 2 or passes college courses at age 12 will grow up to be a genius, but we don't really know because there are no tests for adult geniuses. Some may become rich, some may pursue goals we don't understand, and some may have social problems and live isolated lives.
Some intelligence tests actually test for capabilities under controlled conditions that we believe correlate with intelligence. For example, we may find the ability to memorize lists of words or solve arithmetic problems correlates with intelligence in humans. However these types of tests cannot be applied to machines.
-- Matt Mahoney, firstname.lastname@example.org
This archive was generated by hypermail 2.1.5 : Wed May 22 2013 - 04:01:37 MDT