Monthly Archives: May 2016

An update to the last post. Pro Publica just published this report indicating racial bias in the software used to predict recidivism. So of course, the robots are only as non-racist as their programming.

Also, someone asked me how we could solve some of these problems without resorting to AI. Here’s a very rough sketch:

1) Have judges use a sentencing checklist that determines maximum sentences based on relevant factors, and make the checklist public, and its application public, so that independent auditors and the public can see if it was properly adhered to in each case. This helps remove the judge from the problem by making lists of features related to each crime that set the sentencing. That way, a judge who is hungry cannot give a harsher sentence; a harsher sentence cannot go to a black youth rather than a white one; a black youth will not be “tried as an adult” when a white youth in the exact same criminal situation would not be. Of course, such a list is difficult to put together, but it’s not impossible to at least improve on current guidelines.
2) Give better instruction to jurors on the unreliability of eyewitness testimony. For example: currently it’s ok to tell jurors that if a witness seems certain, they should take that into account. However, memory research (cf Loftus et al) shows that subjective certainty has almost no bearing on accuracy of eye witness reports.
3.) Make all emotional appeals inadmissable. Eliminate the opening arguments in criminal cases. Demand that closing arguments be limited to relation between evidence and verdict.
4.) Both the defendant and the victims, if there are any, should be invisible to the jury. Follow the mode of modern orchestra auditions so that the jury can’t see if the victims and defendant are black or white, male or female. Using a screen and a voice-disguising device (vocoder) would eliminate a great deal of bias in cases.
5.) Make it the law that there must be equally matched lawyers. If there is a criminal case, both the prosecutor and the defense attorney should come from the same office, a general legal office, and the role of prosecutor/defense should be determined by coin flip in each case, or should alternate so that everyone does an equal amount of both. Make sure each side is absolutely equally funded; no spending money the other side doesn’t have. Both defense and prosecution should come from public funds, and be well funded and have the same social access to judges, police and forensic specialists
6.) Eliminate the plea bargain. This is a tool designed to get poor people to go to jail because they can’t afford attorneys, or can’t spend time in jail waiting for a trial. Notably, plea bargaining is illegal or severely restricted in most countries already. The U.S. has one of the most extensive, and highly abused, plea bargaining systems in the world.

Judges are known for disliking mandatory sentencing laws; generally, they hold that a human being is better able to understand the nuances of a case, and shouldn’t be prevented from using his or judgment to give more or less lenient sentences depending on circumstance. But initial research, though far from conclusive, has indicated that judges are notoriously bad at this sort of thing. Instead of allowing proper considerations to govern sentencing, factors like how long it’s been since the judge has eaten[1], the race of the suspect[2], and the sex of the victim[3], seem to have much more effect on sentencing than such proper measures as the severity of the crime and the likelihood of re-offense.

Similarly, the justice system is notoriously biased against poor defendants, and juries are terrible at distinguishing trustworthy testimony from untrustworthy testimony[4], relying upon such factors as the appearance of the testifier, and testifier’s command of English and emotional reactions.

Recently, IBM developed an artificially intelligent “lawyer.[5]” It doesn’t do much besides the legal scutwork, the boring research and paper-sorting that lawyers tend to find tedious and deadening. So on that level, it’s probably doing what people going back to Marx hoped mechanization would do, which is take away the worst jobs (although Marx and Russell’s hope that this would free people up for lives of leisure requires a large-scale political economy project which may or may not materialize.) But what if we could create, a la the Deep Blue project, a super-lawyer that found the best legal strategies and could argue the strongest case.

Combine this with an AI judge and jury and, at least potentially, there could be an increase in justice: the AI judge or jury would not sentence based on its hunger or its emotional manipulation by tears or the heartfelt sincerity of (often faulty) eye-witnesses. If both defendant and prosecutor had access to the same AI law programs, there wouldn’t be an issue of a wealthy defendant having an unfair advantage, and a poor defendant an unfair disadvantage. And with a good AI jury, we could avoid the sorts of legal tricks that rely on irrelevant appeals in order to win cases.

It’s a way off, but it at least is a possibility that we can hope for automated justice that lacks the rather nasty implications of letting biased, prejudiced, easily manipulated, and cranky-because-hungry humans decide matters of such importance.

[1] http://www.pnas.org/content/108/17/6889.full

[2] http://www.sentencingproject.org/publications/shadow-report-to-the-united-nations-human-rights-committee-regarding-racial-disparities-in-the-united-states-criminal-justice-system/

[3] http://cad.sagepub.com/content/early/2014/04/10/0011128714530826.full.pdf?ijkey=BRpzSvqlj4KILoG&keytype=ref

[4] http://www.nytimes.com/roomfordebate/2011/08/31/can-we-trust-eyewitness-identifications/juries-dont-understand-eyewitness-testimony

[5] http://futurism.com/artificially-intelligent-lawyer-ross-hired-first-official-law-firm/