05 May 2010

Speech Recognition & AI, again

The Cranky Professor | Speech Recognition Fail

The end of the pursuit of speech recognition for computers - and how it may mean that artificial intelligence will never get here either:
The accuracy of computer speech recognition flat-lined in 2001, before reaching human levels. The funding plug was pulled, but no funeral, no text-to-speech eulogy followed. Words never meant very much to computers--which made them ten times more error-prone than humans. Humans expected that computer understanding of language would lead to artificially intelligent machines, inevitably and quickly. But the mispredicted words of speech recognition have rewritten that narrative. We just haven't recognized it yet.
via Marginal Revolution
I also don't know anybody else in the AI community that thought speech recognition was the route to artificially intelligent machines.

At the risk of commenting before reading the rest of the article the good Professor links to, I think there may be some conflation going on between speech recognition and natural language processing, which has a much stronger link to AI.  (Though I don't think anybody outside of NLP thought it was, is, or will be the best route to strong AI either.)

~ ~ ~ ~ ~

Part of the reason people don't seem to see a lot of progress in AI is that the goal posts keep moving.  We get used to computers' new capabilities remarkably quickly, and we cease to see them as AI.  Game playing is a good example of this.  Getting computers to be competitive with humans in things like chess used to be a hugely active area in the field.  Now that they dominate humans though it's passe, it's just a bunch of state-space search algorithms and some optimized hardware.

A vast majority of our AI applications are still way behind even simple mammals in their capabilities.  But it's still important to keep in mind that they're also way ahead of what many people thought was possible a generation ago.

Go back to Turing's 1950 paper, "Computing Machinery and Intelligence," in which he introduced the now famous "Turing Test." He discussed nine possible objections to the notion that machines could be capable of thinking. One of the most well known, and common, is Lady Lovelace's Objection:
The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform. It can follow analysis; but it has no power of anticipating any analytical relations or truths.
A lot of people believed this in 1950. A lot of people believe this today. But we live in a world in which machines can independently rediscover Newton's Laws of Motion, develop patentable inventions, and design a whole new class of antenna now in use on satellites.

None of that is proof that machines will ever rival even lab rats in intelligence, but it is reason to be a little more optimistic about the future of AI than a lot of people are.  Not as optimistic as Kurzweil, Moravec, et al., mind you, but optimistic all the same.

No comments:

Post a Comment