Will computers hold key to mental hospitals?
- 02 November 1991
- From New Scientist Print Edition. Subscribe and get 4 free issues.
Panels of experts which must decide whether to release potentially dangerous psychiatric patients from custody may in future seek advice from a computer. The Dangerous Assessment Database is a computer program which seeks to predict the likelihood that a released patient will be a danger to society.
Offenders in special psychiatric hospitals are considered for release by a panel of assessors which can include judges, psychiatrists, psychologists and nurses. Recent well-publicised cases of people being attacked by patients shortly after their release, such as that of Emma Brodie, highlight the weaknesses in the present procedures.
Eleven-year-old Emma Brodie was stabbed to death in April by Carol Ann Barrett, a complete stranger, in a shopping centre in Doncaster. Two days earlier Barrett had been released from the psychiatric unit of Doncaster Royal Infirmary.
At the same time, the judgments often err on the side of caution. For every genuinely dangerous person released, research suggests three individuals are kept locked up when they may be safe to discharge.
Ian Stringer, a psychologist who has worked for many years with dangerous offenders at the Rampton Special Hospital in Nottinghamshire and other institutions, developed the system in collaboration with a company called Risk Decisions. The company makes software for risk management in oil exploration and finance.
The system takes the form of a questionnaire program consisting of a about 1000 questions. By quizzing assessors the program gathers detailed information about the individual covering areas such as family background, childhood behaviour, and response to therapy. The information is weighted for reliability on a scale from hearsay, at the low end, to official reports where there have been attempts to corroborate evidence.
In addition, the program gathers information about the assessors themselves in an attempt to bring to the surface any underlying factors which might influence their judgment. 'People making judgments view information in different lights according to their training, their experience, their belief systems, and the contextual issues surrounding the case,' says Stringer.
The answers to the questionnaire are fed into the database. The risk analysis program uses a technique known as genetic algorithms, which mimics the process of evolution to find solutions to complex problems. Normal computer algorithms - the sets of instructions computers follow to solve a problem - are inadequate to tackle complex issues such as dangerousness because of the number of factors involved. Genetic algorithms do not seek exact solutions but aim instead for a 'best result'.
The best result is achieved by starting out with a 'population' of facts about the offender's life, such as a history of antisocial behaviour as a child. These are derived from the assessors' replies to the questions and only those facts which appear consistent between various assessors are included. The facts are also given a score depending on their importance and reliability.
These facts are represented as strings of numbers and are distributed randomly in a hypothetical array analogous to a population of living cells. The population then 'evolves' through a number of 'generations', following rules which define a fact's interactions with its neighbours and whether a fact replicates, simply survives or becomes extinct - only the fittest facts survive. Facts can undergo random changes analogous to genetic mutation and can also 'breed' with neighbouring facts by splitting in half and swapping over half its string of numbers.
After a number of generations a small selection of the most important facts will emerge. Stringer believes that the patterns of facts will 'help us make qualified decisions on the likelihood of dangerousness and what sort of situation is likely to increase dangerousness in an individual.'
The project is now beginning to collect data and systems are being installed in a number of institutions. Clive Hollin, senior lecturer in psychology at Birmingham University and research psychologist at the Glenthorpe Youth Treatment Centre, says that a computer system that gathered information relevant to a decision in an orderly way with weightings for different indicators 'might ease the process of coming to a decision'.
While it is 'science fiction' to think that a computer system could accurately predict human behaviour, Hollin thinks empirical research could show that the database 'may increase the accuracy of decision making - and that's valuable in itself.'