Q&A: Microsoft’s hunt for a HIV cure

Stuart Turton Read more October 7, 2009

Microsoft’s chief scientist Chris Bishop explains the link between Xbox Live and the search for a cure for HIV

Professor Chris Bishop is possibly the most qualified man ever to appear in PC Pro. He has a BA in Physics from Oxford, a PhD in Theoretical Physics from the University of Edinburgh, and is the chair of the Institute for Adaptive and Neural Computation. Now he’s the chief research scientist at Microsoft’s Cambridge research lab, tasked with developing the next generation of artificial intelligence.

We caught up with him at a recent open day at the company’s research centre to find out what he – and his staff – are devoting their considerable brain power to.

Q Artificial intelligence seems to have stalled in recent years. Is this actually the case, or are we just being impatient?

A We’re on the third generation of machine intelligence right now, and it’s emerged only in the last five years. The classic example [of AI] is getting computers to recognise images; knowing this is a table and this is a chair. The rule would say, the vertical piece is a leg. That isn’t always true, though, so you create an exception to the rule, and then an exception to the exception. This sort of AI is very brittle and constantly breaking, so we’ve abandoned that approach.

Today, all research is based on statistics. If you want a machine to learn faces you collect lots of images, some of which have faces in. You label the faces and present them to a computer that does a huge amount of computation to work out what differentiates faces from non-faces. If it works the machine creates its own rules, and can then generalise to new situations – it isn’t thrown off by new lighting, or different-coloured shirts, or other environmental factors.

Q So, is this technology being used in the wild?

If you’re constantly getting thrashed, we’re not going to have many people playing for very long

A Xbox Live uses it to match up players with similar skill levels. After five games of Halo, our Bayesian system is beginning to work out how good you are, whereas you have to play about 50 games under the old system to get the same accuracy. Commercially, that’s important, because if you have to play 50 games before the system knows who to match you against and you’re constantly getting thrashed, we’re not going to have many people playing for very long.

Q The Xbox seems an unlikely breeding ground for artificial intelligence research.

A Think of the discovery of X-rays. If somebody had said “go out and find me a way of mending broken bones”, nobody would have taken a glass tube, put electrodes in it and applied 50,000 volts.

We’re exploring this general concept of Bayesian influence in an academic setting. You do the research, which produces results, and then you find out how to apply it. So apart from the Xbox stuff, we’re also using it to understand how gene-environment interactions affect childhood asthma, and our colleagues in Redmond are using it to try to develop a vaccine for HIV by looking at how the protein coat on HIV evolves.

Q How does Microsoft feel about research that isn’t directly making money?

A We’re funded as a corporate tax, which means we don’t have to go before product groups and seek funding. That’s crucial, because in many industrial labs where they do have that model it causes the research to become advanced development, and very focused on the ship cycle of that product. Therefore, you don’t make these big discoveries.
People criticise blue-sky research as being useless and just for fun. But it isn’t about that. It’s about looking at new things and finding unexpected things and then finding out how that’s applicable to different projects. Good research is 60% about asking the right questions, and 40% answering them. That’s what we do.