Wednesday 21 July 2010

Playing with brains - a write up

As I mentioned previously, I went along to a talk entitled Playing with Brains on Monday evening. It was given by Peter McOwan, Professor of Computer Science at Queen Mary, University of London. I guess it was very much a lecture for the ley person but I found it very interesting and for a fleeting moment considered taking up studying part time. I won't, though, I simply don't have enough spare time and I know my motivation would slip. All the same, I will look out for more events run by the same people.

The talk took place in a fairly small room in a pub and while it filled up slowly it ended packed, and I was surprised to find a much greater female to male ratio than I usually see at tech meetings.

Peter McOwan has a background in technology and psychology and he used the session to give an overview of current understanding of how brains work and how the knowledge can be applied to AI with liberal use of illusion and computerised examples. The first we were shown was the checkerboard illusion in which a shadow is cast over a checkerboard and the two squares upon it are marked, one we instinctively know to be a black square, the other a white one. In fact both are the same shade of grey but out brains will not accept this without demonstrative proof. McOwan points out that we are getting very good at teaching computers what is right and what is wrong, but getting computers to get things wrong in the same ways that we do remains somewhat ellusive. He explains that many of our perceptions are based on evolutionary need and repeated examples reinforcing our understanding, so in the same way that this just seems true, so too are we 'programmed' to see faces everywhere and to understand that the light source comes from above even when it doesn't. He reinforced the idea of what a face looks like according to our brains by showing the rotating mask illusion. We know that noses stick outwards and aren't sunken inwards and so that is what we are forced to see.

Much of this was stuff I was already passingly familiar with, but he also covered some history of understanding brains. I had never realised that ancient science understood brains to be containers of liquids squirted around our body, causing mixtures of cognition, imagination, emotion and memory. I was familiar with the basics of phrenology, but hadn't linked the head maps and bumps to the actual shape and mapping of brains in more modern times and they are disturbingly similar.

McOwan talked at length about how we know what we know, and how we accept that it may all be proven incorrect further down the line as we take our understanding further. One of the biggest sources of examination is what he termed "nature's experiments", that is, those who have suffered brain injury and now perceive things differently as a result. He talked in particular about a patient who cannot recognise movement, and moved on to discussion of further routes of examination that include monitoring brain blood flow using FMRI and depolarising sections of the brain to effectively turn them off. This latter work has been performed on people he knows and he reports that they talk about it being a very trippy experience, one he's reluctant to volunteer for himself. I asked him to expand on these experiments and he explained how areas of the brain that deal with speech or movement can be disabled and the "victim" finds themself unable to interact normally.

Moving on we covered further illusionary and test examples where McOwan demonstrated how the brain fills in missing details so we overlook changes between two pictures, or fill in the gaps when listening to an MP3 file. Computers, meanwhile, can spot these differences, and use them. Further, they can be "trained" or appear to learn and while they can fake many things others are more easily perceived to be broken or wrong. He talked about how computer generated images that are based on human painting styles can fool art critics, but this seemed to hold little weight to my mind. Elephants can paint pictures that people laud as great examples of artwork before they know their origins, after all. He talked about companion robots used around the world and the work he does for Lirec and touched briefly on what counts as a cyborg (a man wearing glasses has enhanced mechanical ability, after all) before time drew short and he opened the floor to questions.

Overall, it was a fascinating talk and well worth attending. It's clearly worth looking out for more from Science London.

No comments:

Post a Comment