December 19, 2009

Emergent humanoid robotics

hanson einstein face [see below]

There are no doubt many patent hurdles and other impracticalities to hinder the following ideas. I just play with things and then move on, which is why I'm a dilettante proofreader rather than something more focused, but also why I'm [probably] happier this way.

Three different robot stories. The first is from Hanson Robotics, seen below in this recent TED clip. They try and make robots with human faces. It looks like animatronic / Disney stuff, but there seems to be more going on, or almost going on. The model shown in the video [Einstein, not the doll face in the frozen shot] can mirror what people are doing - you smile, it smiles. There seems to be a pretty fine level of control of the motorized musculature underneath the realistic looking skin, and although it obviously needs more work, the future has a lot of time.

What caught my attention was this mirroring, which is based on being able to read faces, and so these machines should be able to read microexpressions and basically be loaded up with all of Paul Ekman's work.

The next video won't embed, but it's from an article at BotJunkie called Robots Learn to Look Shifty. Basically the robot, under certain conditions, can use eye movements to give cues that human subjects respond to but don't consciously pick up. The two projects are separate, but they could be easily be joined together.

And the third video, below, shows how motion capture can be used to create a more natural moving humanoid robot, with this one particularly good at swiveling its hips.

The point is that robotics seems to be full of seemingly isolated projects that aim to do one small thing very well, and usually one thing that on its own is cool but not that useful. But one day all of this technology could be integrated into one machine, which would obviously be connected to the Internet so that it could run many apps and access all data. Out of all of these projects something bigger will emerge.

I had a post about emergent AI a while ago, based on the idea that there's no center of consciousness in the brain, but that we emerge from the interaction of many simple processes [Minsky's society of the mind], so that the full complexity of their interactions undertakes a phase transition to another order of simplicity in consciousness, which can only hold a few bits of data at a time and is essentially creating useful half-truths out of a torrent of data that has been entirely reconstructed in the brain. The astonishing hypothesis is just how things are.

What's true of robotics is true in the field of AI - a lot of researchers working on small things that one day may come together and then something else occurs. But it struck me the other day that computers do not need an unconscious and perhaps therefore not a consciousness. In the small space required for cunning we don't act entirely on instinct, but we also need to filter all the data down to the essentials. A computer wouldn't need to do this - bandwidth isn't a problem, it can hold a lot of information in its working memory and access it all in practically an instant. So the nature of consciousness, if it emerged, would have to be different. I quoted Hofstader a little while ago, and I'll do it in full again, as it serves my purpose here:
When and only when such a loop arises in the brain or in any other substrate, is a person - a unique new "I" - brought into being. Moreover, the more self-referentially rich such a loop is, the more conscious is the self to which it gives rise. Yes, shocking though this might sound, consciousness is not an on/off phenomenon, but admits of degrees, grades, shades. To put it more bluntly, there are bigger souls and smaller souls.
Douglas Hofstadter, Godel, Escher, Bach, P-6
The point being would such loops arising on a large, distributed computer connected to sensors, RFID tags and so on have access to a bigger soul? I don't see why not, and then that would be available everywhere, in all machines plugged into the mind.

Blade Runner - Voight-Kampff Test

No comments: