Could Judges be replaced by algorithms

An algorithm is a set of rules to be followed in problem-solving operations. A computer programme can take a number of variable inputs, follow a certain set procedure, and yield a particular result. So it could be said to operate like a decision-maker, such as a judge, who is given a set of facts, applies the rule of law to them, and makes a particular decision.

The idea of replacing judges with algorithms is not a new one. In 1977, Anthony D’Amato wrote a piece in the Georgia Law Review entitled “Can / Should Computers Replace Judges?” He suggested that there could be benefits to society from the saving in costs arising from the use of computers, as well as the removal of uncertainty from certain court decisions. He wrote: ‘A diminution in the trappings of law may signify greater equality before the law and greater delivery of the equal protection of the law to poor people.’

Notwithstanding the enormous developments in computing in the intervening forty years, it does not seem that any country has taken the bold step of passing its legal decisions to computers. A high-profile case in the US state of Wisconsin has led many to suggest that such an approach was dangerous. In Wisconsin v. Loomis (2017), an ‘algorithmic assessment program’ called Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) was used by the court in a sentencing decision.

The COMPAS programme was not of a legal nature, however, but was designed to predict the rate of recidivism based on publicly available information. The main objection to the programme was that its methodology was a proprietary trade secret that could not be accessed by the offender prior to sentence. The Wisconsin Supreme Court held that it did not violate Mr Loomis’s due process rights, provided that any report that relied on COMPAS included a number of warnings for the judge who considered it.

Could an algorithm be developed that actually assisted, for example, sentencing decisions, or personal injuries assessments? Several thousand sentencing decisions are made in Ireland every year, that rely on a relatively predictable considerations, including: the offence itself; the maximum available sentence; the impact of the offence on the victim(s); the age and stage in life of the offender; any history of substance abuse; any previous offences; the other people likely to be affected by the sentencing decision; the sentences handed down to any co-accused; whether (or when) a plea of guilty was entered; whether the offender shows remorse; and whether he or she is likely to re-offend. Unlike the UK, where there are sentencing guidelines, in Ireland there is often perceived to be quite a disparity in the sentences handed down for particular offences.

It seems probable that a programme could be developed that assimilated all of the sentencing decisions made each year, and the various considerations involved, then applied them to each new case. Such an algorithm could be of value to judges, and to the solicitors and barristers in the case, if it were publicly available and had a comprehensible methodology. For example, having considered the ‘inputs’ in a particular sentencing decision, it could identify the five or six previous decisions most relevant to the case at hand.

Such a programme might have other advantages. It would probably identify ‘outlier’ judges who were particularly light or harsh in their sentencing decisions. It might also indicate whether there were any bias in the system based on ethnicity, class or gender.

The danger would be if any decision were left to the programme itself. Nobody wants decisions of such magnitude to be made by a computer programme alone. While an algorithm could assist in the decision-making process, the decision itself would have to be made by a person.

First Published in Law Ireland in May 2018