A computer that is used to determine if a criminal is likely to reoffend in courtrooms is ‘no more accurate than untrained humans’, claim scientists.
Used across the United States for 20 years, the algorithm decides if a criminal is too high-risk to be released on bail.
The researchers say that a group of untrained members of the public contacted through an online survey had the same success rate as the software.
Results of the study cast doubt on the accuracy of the machine used in more than one million cases.
A computer programme used in US courts has been found to be only as accurate at predicting reoffending as untrained humans that have answered questions in an online survey. The programme has been used for twenty years
Dartmouth College researchers looked at the accuracy of the Correctional Offender Management Profiling for Alternative Sanctions (Compas) software system used in US courts.
Scientists compared Compas success rates to the opinions of participants from Amazon’s Mechanical Turk crowd-sourcing marketplace.
The researchers found that the results from the untrained people were just as accurate as the algorithm.
Dr Hany Farid, a co-author of the paper and professor of computer science at Dartmouth College in New Hampshire, said: ‘The cost of being wrong is very high and at this point there’s a serious question over whether it should have any part in these decisions.’
The programme is designed to give objective assessments of an individual’s risk, without any human bias from the judge or jury.
Previous research has found that race, gender and employment status can all alter the fate of a defendant in a criminal trial.
Regular members of the public were given short descriptions of a defendant that included their sex, age, and previous criminal history.
They were then and asked whether they thought they would reoffend.
According to the study, the human results were accurate in 67 per cent of cases compared to the 65 per cent success rate of Compas.
In a second stage of the experiment, the study gave the participants just two pieces of information about a criminal – age and criminal history – and asked them to predict the likelihood of a person committing a further crime.
On the topic of this simple equation used to determine a persons fate, lead author Dr Farid said: ‘When you boil down what the software is actually doing, it comes down to two things: your age and number of prior convictions.
Similar to the technology used in the Tom Cruise film Minority report (pictured) the Corpus algorithm is used by law enforcement agencies to predict if a person will commit a crime
‘If you are young and have a lot of prior convictions you are high risk.’
Compas was developed in 1998 by Northpointe Inc, a subdivision of Equivant, a company with several US offices that specialises in tools and aids for the legal domain.
The details of the mechanism and parameters used by the computer software have been closely guarded since its inception.
That is something that Dr Farid believes should change.
‘As we peel the curtain away on these proprietary algorithms, the details of which are closely guarded, it doesn’t look that impressive.
‘It doesn’t mean we shouldn’t use it, but judges and courts and prosecutors should understand what is behind this,’ Dr Farid said.
Speaking to the Guardian, professor Seena Fazel agreed the workings of the software should be widely understood.
Professor Fazel said: ‘I don’t think you can say these algorithms have no value,
‘There’s lots of other evidence suggesting they are useful.’
Compas – Correctional Offender Management Profiling for Alternative Sanction – is used in US courts across the country and has been used in over a million cases so far to determine parole and bail conditions
Equivant published a statement in response to the damning claims made by the research.
In it the company criticises the techniques used by the researchers and refutes the claim that COMPAS uses 137 risk-assessment factors, claiming the algorithm actually only uses six.
‘The Compas risk assessment has six inputs only. Risk assessments and needs assessments are not to be viewed as one and the same,’ it says.
The company took objection to the methods used by the researchers.
‘We also question the testing procedure utilised with the brief 2-question risk scale.
‘This does not meet the test for external validation, thus adding further ambiguity to the findings.’
The developers say they will be requesting the data used in the referenced study to assess the research methods used by the researchers.