dark knowledge

Reading about Geoffrey Hinton, an academic computer scientist who has focused on artificial intelligence (AI), I came across several terms I had never heard.

I’m no scientist but I am aligned with him in believing that we understand little about the brain. We understand so little that the idea of being able to create AI seems rather foolish.

But Hinton’s algorithms might already be behind some aspects of our lives.

I had heard of neural networks (the artificial and biological types) and now I find out that they have been rebranded as “deep learning.” These artificial neural networks sound like sci-fi but if you use mobile devices then you have especially embraced them via Google, Facebook, Microsoft etc.  Their neural nets take all the data we throw at them and make predictions from it. It is big business. (The article points to a company called DeepMind that Google spent $400 million on just to get their memory integration method.  They bought some of Hinton’s time too.)

Does it sound more like The Matrix films or when the Terminator said “My CPU is a neural net processor” than real life? Wait – are you carrying a little “learning computer” in your pocket these days?

Deep learning goes back a ways. Hinton was writing ten years ago about “deep belief networks” which improved themselves as they ran, and that became artificial neural networks, which became deep learning.

I can’t quite imagine life on Google’s campus, and it’s even harder for me to imagine Hinton (who works there half the year) playing with neural nets there and going after what he calls “dark knowledge,” which sounds a bit creepy. It reminds me of some occult studies.

If I watch/listen to him lecture about it, I don’t get frightened, just confused. Machines learning has been a staple of sci-fi for a long time and it still scares most of us.

I’d try reading a Wikipedia entry on dark knowledge, but there isn’t one yet. That’s a bit scary too.