Thursday, February 11, 2010

"How Long Till Human-Level AI?"

H+ magazine recently carried an article regarding a small-scale survey of experts (twenty-one of the attendants at the 2009 Artificial General Intelligence conference) on the question of when general artificial intelligence will arrive, and specifically when it will attain four milestones-Turing-level intelligence, the intelligence of a third-grader, the intelligence needed to do Nobel Prize-quality work, and finally the key Singularitarian outcome of superhuman intelligence.

Interestingly, fifteen of the twenty-one-seventy percent-of them predict a computer will pass the Turing test by the 2040s.

A significant percentage answered "never," however, particularly in regard to the question of when a computer would achieve superhuman intelligence-nine of the twenty-one answering in this way, the greatest unanimity the survey finds on any point. However, that still leaves the doubters of that particular outcome in the minority, eleven of the twenty-one predicting this actually happening by the 2040s. (Incidentally, the second most popular guess was that computers would be doing Nobel-quality science by the 2020s, with a full third of the respondents giving that answer.)

Of course, this is a limited examination of the views of a small, pre-selected group (these all being AI specialists rather than a more general sampling of computer scientists), and this is a particularly tricky kind of prognostication, so that my guess would be a likelihood on their part to err on the side of overoptimism rather than the reverse. Nonetheless, that such a view is common is well worth noting, and the discussion (as well as the magazine more generally) well worth a look from those interested in the issue.

The Return of Neo-Medievalism?
2/10/10
$123 Trillion by 2040?
2/10/10
A Kurzweil Reboot?
2/8/10

No comments:

Subscribe Now: Feed Icon