Sunday, 26 February 2012

Are we helping an AI learn its ABCs?

Pause for thought re commenting verification. If this new system is getting us to identify hard-to-read words - see here - we may be part of something needing more discussion.

In this article, on a modified history of the computer, Freeman Dyson's son tells us:

In 2005 I visited Google's headquarters, and was utterly floored by what I saw. "We are not scanning all those books to be read by people. We are scanning them to be read by an AI," an engineer whispered to me.

It makes perfect sense, if we want to make the fictions real. But do we? And which AI? How soon? Whose vassal will it be - or is it? What do we all owe to a lifeform like this, and - getting back to the mundane - what does it mean for our use of the verification?

At novums we get a reminder of where this could be going re Isaac Asimov's "The Last Question", but that's one of the more optimistic views, and the literature has others.


Anonymous said...
This comment has been removed by a blog administrator.
G said...

that is a very scary day this stuff will become reality

Aaron E. Steele said...

Cyberdyne (I mean Google) will get no help from me. I've turned off word verification.

Porky said...

@ SAROE - I saw that via email notification and I think the point is a good one.

@ G - It could be reality already of course. How would we know, and how do even we define the change? The more we learn, the hazier the concepts seem to become.

@ Aaron E. Steele - I've been more or less expecting I'd do it since the last discussion, but holding out a little longer just in case more came to light. I didn't quite expect this though. One argument for keeping verification on in this context - if the unclear word theory is correct - is the fact it should help with the understanding of things that were meant to be understood, whoever the reader is - assuming the authors would want their work understood by those who might. And that author consent is of course a key issue. Then again even if we do agree with the controversial nature of the push to digitise everything, the argument doesn't count for much in the case of the AI, given that AI will have a much harder time working out a lot of what gets written elsewhere, on the net as a whole for example. The working it out could be beneficial in its development too, and in general, taking it all in good faith, the AI learning the sum of printed human knowledge and developing we might assume is a good thing. On the whole, while I do still wonder whether us turning verification off en masse is part of the point, I dislike the potential for the unagreed harnessing of our mental processing power more.