Computers don't have much sense of time. They can count nanoseconds, calculate missile trajectories, mix strands of music to pure digital harmony, but they can't begin to ponder whether the chicken preceded the egg, because the information that is fed to them is already dead.

The information computers consume is often littered with references to the time of the real world outside. But machines treat timing information as they treat tax figures - a series of numbers to be added and multiplied just any how. Machines treat all data as static, with time just another dimension to deal with, another column of interminable symbols. Which is all very well when a problem involves dozens or even hundreds of dimensions, one no more special than another. However, time is in our lives something special, recording change and implying cause and effect hidden within the other dimensions, cause and effect that form no part of most computation.

Causality is crucial. It is how humans organize much of their knowledge, forming the basis of their expertise and experience. Reasoning about before and after is what we spend most of our conscious, analytical brain on, yet it is not something that computers have been well equipped to do.

In fact, in the areas where such temporal reasoning in machines would most benefit us, there has been a deliberate movement away from all forms of reasoning, to neural networks that try to emulate the lower levels of the brain. But although it is likely that, and not completely understood how, the human brain tackles causality at its lowest levels, the computer simulations by and large avoid it altogether. Programs to predict market trends, for example, do so with minimal causal or temporal information - they treat input data as dead, identify patterns and extrapolate from the past without discovering underlying causes. Of course, more often than not the humans using these programs don't understand the underlying causes themselves.

Before Artificial Intelligence (AI) went out of fashion, when all human thought processes were going to be simulated by rational computers, forms of temporal logic were developed to get machines to understand time. Temporal logic emphasizes time - before, after, now - without which no data are complete. The relationship between data and the importance given to different elements depend on, and are calculated from, the temporal information that accompanies them like the beat on a musical score.

Unfortunately, temporal logic tends to get too complicated for even the human programmers to follow, which is why neural networks, sometimes souped-up with a dash of temporal processing, end up being used for such time- dependent tasks as balancing oil rigs on the seas or hedging against falling coffee prices.

But time is far too important to be left to the computers of billion-dollar corporations. In the age of intelligent agents and information floods it is strange that we do not have a computer-comprehensible way of incorporating time and causality into our knowledge bases. Only when we tell our machines the time, and teach them to work with it to our benefit, will the vast stores of information they hold truly come alive.