I liked the Steve interview, if only because it proves that his project is still going. Steve Grand is one of the very few people in the UK who is actually trying to push the AI envelope a bit.

What I would say though is that it ain't over till the fat lady sings, or in this case until the ape-like robot points at the banana and not the apple. Just because you think something should work in principle and you may have tested some of the steps in isolation it doesn't necessarily follow that the thing will actually work when you put it together.

Object recognition is notoriously tricky and a lot of researchers get away with it by adopting the PNAMBeC (Pay No Attention to that Man Behind the Curtain) approach. They usually tend to specifically engineer the robot's environment so as to make the recognition process trivial, for example by showing the object against a single coloured background, or by deviously putting critical assumptions into the software such as what colour the background will be or what direction the light source will be coming from. Having read a lot of machine vision papers I'd say that such assumptions often aren't explicitly stated, leading to false expectations about what a given algorithm can or can't do.

From experiments with my own robot Rodney I'd say that being able to recognise the banana is really a critical facilitating technology. If I could get my own robot to reliably recognise just a handfull of objects under normal conditions (cluttered backgrounds and variable lighting) that would allow it to start making reasoned judgements about what it can see, and ask intelligent questions using its background knowledge from Push Singh's OpenMind database.

- Bob

Vision is huge. Got any AI/vision sites I should be looking at?

ROB.T.

Steve came and gave a talk about the more technical details of his approach to vision with Lucy last week and, I must say, he seems to be on to something. He has completely ignored the normal visual processing pipeline type structure that everyone else seems to be working in and is basing all his work on 'simple' coordinate transforms. It's a facinating way to go about it but he is starting to get results. I think he explains it all in a new book of his. Anyhow, huge respect to him - he has no job, no funding, no income other than the books and a few donations. The real modern day pioneer :)