Oct 14, 2009

Machine Reading

A Machine Reading system would be able to parse any documents and extract all the info, like people, places, time line, any associations between those things and more. It would also have to understand innuendos, ambiguities, sarcasms and backwards logic to understand the real meaning of a text. It would be nice for many application domains like book publishing, web search, forums policing or spying.

Are you nervous now?

A program like that doesn't exist right now. This looks to me like a perfect NP-complete problem. My guess is that it's worth a PhD at the very least... But for that you have to be able to publish your results. Why is that relevant? Because, it looks like the first application of that type is not going to be from the private sector but from the military. Yes, DARPA just gave a 5 year contract of nearly 30 million dollars to BBN Technologies to 'build a prototype' of a fully functional machine reading AI.
Oh, and yes you will need a security clearance to work on the project.

How about now?

For my part, I wish Goolge was building a thing like that and would make it an Open Patent. I also wish I was working on the project but that's another story.

No comments:

Post a Comment