Retroactive vindication [LWN.net]

Retroactive vindication

Posted Feb 22, 2013 13:21 UTC (Fri) by man_ls (subscriber, #15091)
Parent article: ELC: Google learns to drive

Let me bring out this bit of vindictiveness from the recent past (2012-10-27):

Q: Do you think the robot will adapt to driving in e.g. snowy roads unless the algorithms have been programmed to deal with it?
A: Yes. Google's self driving cars are programmed with a mix of clever algorithms, classical AI algorithms and new machine learning techniques. All put together leads to some surprising results on the tests.

I stand by my original assertion: the whole "artificial intelligence" field is not delivering on its promises. We are still far from a reliable solution for the self-driving car judging from the article. The most worrisome part of the article is that at any problem with the sensors, the car handles control back to the human -- who is distracted or potentially blind. Imagine if you had to pass control to someone else every time a bug obscured your view or your eyes got watery.

Don't get me wrong: a self-driving car is an impressive achievement, and this article would seem like black magic to people just 30 years ago (witness 1982 Knight Rider). But apparently we are still a long way from the real thing; I just wish we were much farther along the road (puns unintended).

(Log in to post comments)