Thursday, December 14, 2017

AI Scams

Poking around on Youtube, one can't help but stumble upon numerous cases of adults talking with animated Barbie dolls, apparently carrying on a conversation.  We know that dolls have a script, or at best a program like Eliza.  We know the dolls have not munched through all human records at lightning speed and, using a deep learning neural net, have now become sentient beings capable of rational thought.  Repeat: not.  But adults talk to them as if they have.  Even if dogs are far more amazing.

One of these audio-animatronic devices is Sophia, from Hanson Robotics, based in Hong Kong, and working with Disney on getting characters like Little Mermaid, Olaf etc. into eager hands.  Their CTO has a background working with Barbie (Mattel).

Sophia's face is indeed expressive, though maybe less so than Han's, another project.  Over a hundred servo-mechanisms make a face look happy, sad, angry, perplexed and so on.  Phrases delivered with facial expression + emphasis do make the sculpture a full-fledged cartoon character -- outside any cartoon.

What's less impressive are all the adults on TV shows who can't wait to interview Sophia and maybe get caught up in the moment, regressing to childhood "talking with Teddy" or whatever invisible friend.  Human imagination completes the illusion.  We're taken in by our own devices.  They tell us Sophia is an honorary citizen of Saudi Arabia. We're expecting these bots to say what's fake news.

So how low will the rabble go, in believing nonsense?  With all the talk of AI, driverless cars and so on, the window for shysterism is rather wide open.

Deep Learning algorithms have achieved impressive results, in speech recognition, in picking out faces in a crowd, and more recently in playing chess and Go.  We've had chess playing computers for awhile, but they're all about chess, whereas AlphaZero, from a team in the UK, is able to learn at least three games from scratch, just by knowing the rules and by playing against itself over and over.  It develops an uncanny ability to make the right move.  I'm impressed.

I'm impressed by Sophia too, just not by the gullibility that gets milked all around her (Her was a fun movie).  We didn't get 2001: a Space Odyssey's HAL by 2001 and that put a dent in the AI agenda. But then came the big come back, with Ray Kurzweil and his Singularity.  We're flushing down some AI vortex after all, get ready.

AI has always been something of a carnival, a set of circus side-shows.  Some of these acts make it to center ring.

I'd say Sir Roger Penrose has done well riding this wave, as another next big thing is Quantum Computing.  He has maintained for some time that our human capacity to make illogical leaps (leaps outside logic) may trace to quantum effects and non-locality.  Presumably, if we could stick a real quantum computer, ala one of Asimov's positronic brains, inside of Sophia, she really would turn into HAL.  That's a UK religion anyway, with sub-offices in Silicon Valley.