Computer science and biological science have a lot to teach each other

Computing and biology

Arresting developments
Jul 12th 2007 | CAMBRIDGE
From The Economist print edition


Computer science and biological science have a lot to teach each other


THERE is, at the moment, a lot of interest in the idea of artificial life. The ability to synthesise huge screeds of DNA at will means the genomes of viruses can be replicated already, and replicating those of bacteria is not far off. But that merely copies what nature already manages routinely. David Harel of the Weizmann Institute in Israel reckons he can do better. He proposes to recreate living organisms inside a computer.
As with many biologists, his creature of choice is a worm called Caenorhabditis elegans. This tiny nematode (it is just a millimetre long) was the first organism to have its developmental pathway worked out cell by cell and the first multicellular one to have its genome sequenced completely. It is probably, therefore, the best understood animal in biology.
As he told “The next 10 years”, a conference organised by Microsoft Research in Cambridge, England, Dr Harel has been working on a computer model of C. elegans. He hopes this will reveal exactly how pluripotent stem cells—those capable of becoming any sort of mature cell—decide which speciality they will take on. He thinks that a true understanding of the processes involved will be demonstrated only when it is possible to build a simulation that does exactly—but artificially—what happens in nature. With colleagues at New York University and Yale University in America, he is modelling and testing the possibilities.
Indeed, he proposes to evaluate the result using an updated version of the Turing test. This was devised by Alan Turing, an early computer scientist, to identify whether a machine is capable of thought. The original test proposes that a person be presented with a suitable interface—say, a keyboard and a screen—through which to communicate. If the operator cannot tell the difference between talking to another person through this interface and talking to a computer, then the computer can be argued to be thinking. Dr Harel's version is a little more challenging. He wants to test whether scientists well versed in the ways of C. elegans could tell his computerised version from the real thing. So far, the distinction is obvious, but it may not always remain so.

Silicon biology

Stephen Emmott, who works for Microsoft Research, wonders whether to turn the whole approach on its head. Instead of looking at how computers can mimic creatures, he wants to build computers from biological components. People—and other creatures—are notoriously forgetful and not much good at number crunching compared with their silicon counterparts. But they do have compensating advantages. People excel at reasoning and make much better learning machines than do computers. Dr Emmott reckons that a biological computer might find it easier to cope with problems that have foxed the traditional, silicon variety for decades—such as how to recognise what it is that they see.
Working with Stephen Muggleton of Imperial College, London, he is developing an “artificial scientist” that would be capable of combining inductive logic with probabilistic reasoning. Such a computer would be able to design experiments, collect the results and then integrate those results with theory. Indeed, it should be possible, the pair think, for the artificial scientist to build hypotheses directly from the data, spotting relationships that the humble graduate student or even his supervisor might miss.
Indeed, Luca Cardelli, who also works for Microsoft Research, likens biological cells to computers. He points out that creatures are programmed to find food and to avoid predators. But exactly how the “wetware” of molecular biology works remains a mystery. Dr Cardelli is trying to discover whether it is more like the hardware of electronic circuits or the software of programming languages. He is using statistical techniques—in particular, a method called stochastic pi-calculus—to model how biological systems appear to change with time.
His colleagues, meanwhile, are examining how the spread of diseases such as malaria and AIDS can be thought of as information systems. They are using what used to be called artificial intelligence and is now referred to as machine learning to explore the relationships between the two. All of which raises some interesting philosophical points. If, say, a computer were used to diagnose a patient's symptoms and recommend treatment, and the result was flawed, could the computer be held responsible? Peter Lipton of the University of Cambridge, who ponders such matters, suggests that such expert systems could indeed be held morally responsible for the consequences of their actions (although the designers of such systems would not necessarily be off the hook). If so, then it is hard to see why computers should not be recognised for good work as well. If Dr Lipton is correct, the race has now begun to see whether the first artificial scientist to win a Nobel prize is based on silicon or biological material.


Copyright © 2007 The Economist Newspaper and The Economist Group. All rights reserved.