Should we use AI to improve decision-making in the courts? This is the question that researchers from the prestigious Harvard University asked themselves as part of from a new study.
Surprising results
In fact, this technology is already used by judges to make certain decisions. The scientists therefore wanted to see things more clearly by comparing a judge’s bail choices with the recommendations generated by an AI. The impact of these tools on human decision-making was also scrutinized.
The authors set their sights on hearings conducted by a single judge in Wisconsin over 30 months from 2017 to 2019. According to Harvard Gazettethe results are surprising to say the least, since the AI was less efficient than the judge in predicting repeat offenses.
Likewise, it appeared that there was little difference between decision-making assisted by AI and that where the judge officiated without using this technology. Moreover, the latter refused the tool’s recommendations in 30% of cases.
Jim Greiner, professor of public law at Harvard Law School, and co-author of this research, comments:
This surprised me. Given the evidence we’ve cited that algorithms can sometimes outperform human decisions, it appears that this algorithm was tuned to be too harsh. He predicted that those arrested would behave badly, that they would do it too often, and therefore he recommended measures that were too harsh.
AI will not replace the human judge
This work does not, however, invalidate previous research conducted in this area. According to scientists, it is indeed possible to recalibrate the algorithm to make it more precise. A much easier approach than correcting human biases that can lead us to make bad decisions.
One of the other advantages of algorithms in judicial matters according to researchers is that it is possible to make them transparent. By opening access to source codes, we can better understand what leads to a decision in one direction or another.
The debate is in any case absolutely not settled, and we know that these systems remain highly criticized. In any case, there is absolutely no question of delegating legal decisions to machines, and the final decision remains that of a human being.
As a reminder, this is not the first time that we have spoken to you about the use of AI in a court. A startup thus attempted to deploy this technology instead of a lawyer, as we told you about here.