LIST OF IMPORTANT OPEN PROBLEMS ABOUT INTELLIGENCE
===============Warren D Smith====May 2006=========
(*d problems are the most important ones.)
1*. Achieve a multiresearcher consensus as fast as possible on
the definition of intelligence and how the AI field should proceed by starting
standardized intelligence-measuring IQ-contests (I have already written
such a consensus statement in section 25 of my MDoI paper, the only problem
is to get people to sign it to make it *be* a consensus)
2. Understand more about the ME(FP) computational complexity class from section 14.
For example, the theory of NP has impressively grown to cover thousands of
problems and extended with the "PCP theorem" and "APX completeness." What are the
analogues, if any, for ME(FP)?
3. Re section 15, can you devise generators of all isomorphism classes of
trees of certain types (e.g, specified valence sequence, specified number of nodes
with ordered and nodes with unordered descendants) with N nodes, that run in
(which would be optimal) O(1) amortized time per tree
generated, and in which that O(1) also includes the cost of walking upward from all
modified nodes to the root? And now redo the same problem - but for DAGs not trees?
Sublinear amortized time would be a good achievement although falling short of
the ultimate goal of O(1) time.
4. Re sections 17 and 18: study examples of pairs of human mental tests which exhibit
negative centered correlations - investigate the Piagetian development of the
corresponding mental abilities in children.
5*. Write computer programs that are intelligence tests and are universal intelligent-entities,
enter them both in the IQ-contests, and write papers about, and prove theorems about
how they work.
6. Is there some way to define mathematically the "usefulness" or
"interestingness" of an algorithm?
For example, Gaussian elimination to solve systems of linear equations is an interesting
and useful algorithm - but an algorithm that just prints out some completely random-looking
string of symbols is probably not useful and not interesting. Well, that is fine as a subjective
judgement, but is there a way to make this objective and quantitative?