Science and Technology: The crowd that thinks for itself: A computer model which mimics mob behaviour is helping police to develop tactics for command and control in times of public disorder. Phil Hilton reports

WHEN the first beer can flew over my head and bounced off an armoured police van, it seemed inevitable that it would be followed by a shower of similar missiles. As an unwitting bystander at the 'Battle of Trafalgar Square' in March 1990, I was witnessing the type of mob momentum that has become the subject of a software program capable of recreating an evolving riot in all its fast-moving complexity.

The program, developed by Roger Hartley and his team at the University of Leeds Computer Based Learning Unit, can mimic the crowd behaviour at the poll tax riots or, more recently, on the streets of Sweden and allow police commanders to test strategies for dealing with crowds.

The academics attended marches and drew on the experience of seasoned police officers and reports of public events to construct a model of a crowd that behaves realistically, with a mind of its own.

My first experience of large-scale public disorder illustrated a paradox exploited by the software system: when a crowd turns away from the rules of civilised behaviour, it tends to adopt some other fairly rigid pattern. The way that the poll-tax rioters followed the example set by an aggressive minority, that a nearby building site became a refuge for protesters and that shops were looted have all been incorporated into a crowd simulation designed to train police officers in public-event command and control.

The mood of the 'software' crowd can mimic the style of event that the trainer wishes his students to experience. Once the crowd's disposition has been determined, it makes its own decisions according to a series of rules and probabilities. The trainee making command-and-control decisions does not see the action - that happens 'off-stage' - but is kept informed as the system generates reports similar to those that would be sent from units attending actual incidents.

The Leeds team was confronted with the problem of creating a crowd that would present the student with unpredictable behaviour but not act unconvincingly. Mr Hartley considered using a stockpile of scenarios, but such a system would have been limited by the library of incidents and the tutor's capacity to manage the situations.

His ingenious solution employs an evolving network of behaviours, stimulated by external circumstances. He has built into a random program - which works like a dice game - the possibility of unforeseeable group behaviour, limited by rules to maintain plausibility. The crowd is divided into groups of about 20 people, the average number that can be watched efficiently at any one time. Each has a set of possible behaviours - at present, there are 16 varieties - ranging from marching peacefully to pelting the police with missiles or attempting to gain control of key buildings.

Every 15 seconds, each group considers whether to change its tactics, on the basis of pre-set probabilities restricted by surrounding circumstances. For example, a group walking along may encounter a neighbouring group sitting down in protest against an arrest. The former would then have to decide whether to join the latter, or to continue to be co-operative.

Its choice is limited by a hierarchy of actions, which do not allow it suddenly to start burning buildings when it has been fairly docile only seconds earlier; it has to work itself up - through, for example, standing around menacingly and shouting abuse - before it is allowed to become physically disruptive.

The group decisions are influenced by buildings on the route - such as a police station or government office - and the actions of police and other players in the unfolding drama. Once certain behaviours are adopted, the 15-second decision cycle is slowed down. If a group decides to join a sit-down protest, it has to remain seated for a minimum period.

Mr Hartley's interacting groups, although derived from observed behaviour, closely correspond to psychological theories of crowd dynamics. Ivan Hill, a criminologist who teaches a crowd-management course at Sunderland Polytechnic, is impressed by the crowd simulator's ability to mimic conformity.

He teaches football stewards to locate ring-leaders in violent crowds, and emphasises that the urge to follow dominant characters and run with the pack can override an individual's own personality. 'It is possible for an individual to cease to exist and be subsumed by the crowd,' he says. 'A civilised bank manager can end up in the middle of a group, contagious behaviour sets in, and he can become as violent as everyone else.'

Mr Hartley felt that it was important to build in the possibility of an event being hijacked by aggressive elements. 'In a march setting, only the first group will know the route; the rest of the crowd will just follow. So if false leaders come along, the body of the march may be inclined to go with them.' Similarly, a violent group may start a ripple effect.

The software police, although governed largely by the trainee, have a degree of independence in making arrests or forming cordons. Inevitably, this places them in conflict with the crowd, and one side has to win these confrontations. Here, the computer switches to a built-in referee program that acts rather like the hand of God.

The referee takes into account the strength of each side and makes a decision using the random dice programme. Thus, as in life, even if a trainee has theoretically placed enough officers to block a key street, his force may suffer an unexpected defeat.

Mr Hartley believes that, to be credible, luck has to play a part in the proceedings. 'If you made it completely deterministic, it would be too predictable,' he says. 'There can be an unlucky bounce. In real life, the best side does not always win.'

The Leeds simulator, which is being evaluated by the Metropolitan Police, could provide a safe means of developing command-and-control skills. As the model is tuned to become more and more life-like, real crowds, already accustomed to hi-tech surveillance, may find that Big Brother is not only watching them but also has a pretty good idea what they are going to do next.

(Photograph omitted)