Computer says "no"

Every November, the Royal Society and the French Académie des sciences give out a prize to a scientist who has discovered how to do something innovative with computers. Naturally, it has to be a useful innovation. The 45 employees at the Department for Work and Pensions who have been disciplined after using their computers for shopping, pornography and "unauthorised downloading" need not apply.

It's not just government employees who abuse their computers. In May 2008, the head of the Max Planck Institute for Mathematics warned that
City traders were doing it, too. He argued that their lazy acceptance of whatever the computers said would lead to financial disaster.

No one understands "garbage in = garbage out" better than the scientists who have learned that their reputation can be ruined by placing unquestioning trust in the printout. To take a slightly ridiculous example, researchers at the UK's National Physical Laboratory have just fixed a 20-year-old problem in the official computer model of the shape of the outer ear. It was created to define a quality standard for the performance of headphones and mobile phones (you are forgiven for not realising how useful scientists can be). The input scan for the computer model was done at too low a resolution, however, and manufacturers have since been filling in the gaps in ways that reflect best on their brand. It ain't necessarily so, just because the computer says it is.

Innovation in scientific computing has done some wonderful things. It has allowed us to model the heart, giving us insight into how drugs affect cardiac rhythms and how heart attacks develop. It enables us to analyse medical images more accurately, providing earlier diagnosis of cancers. Thanks to computer models, we can see why misfolded protein gives rise to cystic fibrosis and predict the path of dangerous epidemics.

But in every case, the computer's predictions or declarations have to be checked against what scientists can observe happening.

In science, the only credible guide is real-world experiment. Across at the Channel, the French use a slightly different word for "experiment" - expérience. That is no coincidence: it emphasises that scientists learn from trial and realisation. Think, for example, of what the British economist John Maynard Keynes called the "folly and injustice" of the UK government's 1931 plan to beat the recession. The measures, which in effect shut down economic activity, didn't work, and led to the abandonment of the gold standard.

The gold-standard scientific approach - experiment or experience - suggests there is no reason to believe that the same approach to beating recession will work now, either - whatever the printout from the Treasury's computer might say.