Friday, February 10, 2017

Tired of reading about AI from authors who really do not understand it

Just read another article that started, sensibly enough, with discussion of fake news and the problems with blind use of technology. The author then pivots toward discussing the great new advances in AI brought about by "Deep Learning". Maybe I need to read more but, last I checked, that was just a re-branding of back propagation and convolutional neural nets. Things that are not new.

My theory, currently is that the advances in AI have to do with Nuance engineers spreading out into industry, bringing technical secrets with them, about voice signal vectors [in particular 30 years of knowledge about what parts of the signal are most relevant], making it possible for Amazon to reproduce the technology in under 10 years, and same for Microsoft. IBM has been doing it longer than Nuance and, I assume, had their own expertise in voice signals.

Anyway, no, there are no such advances. Just another cycle of hype and much more data and more powerful computers...making it harder to see that the emperors clothes are still "new".

Added: but every day my reading of "Venture Beat" finds another example of an author talking about how AI is going to revolutionize [fill in the blank]. It is kind of tiresome because there is no free lunch and just like voice recognition, finding the proper feature vectors for a given subject area is not automated, it is not provided by statistical ("neural net") models of AI, and it may take several years to get started in any new subject area.

No comments:

Post a Comment