When something isn’t working the way you think it should, find something different. And few will argue that bail (or bond, as some refer to it instead) isn’t broken, keeping innocent people in jail pending trial for their inability to pay. This has myriad impacts, most notable among them being guilty pleas to crimes that never occurred or weren’t committed by a defendant because it gets him out now rather than sitting in the can for the next year and a half awaiting a trial that will never happen.

So why impose bail? You can’t let people arrested not appear for court in the future. That would subvert the system and reduce the majesty of the law to a farce. (For the sarcasm impaired, that was sarcasm). Under some bail regimes,* it’s also for the safety of the community, since a presumptively innocent defendant might be cut loose, re-offend and ruin a night’s sleep for the judge who let him out, causing the judge’s worst picture to appear on the front page of a tabloid and all dinner invitations to be rescinded. No judge wants to suffer such humiliation. And why take the chance, just to cut some likely mutt a break by adhering to the law.

The solution most adored is the one least subjective, and that provides the most plausible deniability to those involved, so if the shit hits the fan, the defendant who’s cut loose goes out and murders a white family, has sex with their dog, calls police mean names and then snorts crack, it won’t be the judge’s fault. Blame the algorithm!

San Francisco, which applied a fixed bail scheme** that is the subject of a constitutional challenge, is trying a new plan.

That could change now that San Francisco has introduced a new scoring system to help judges determine bail based on a defendant’s risk of committing another crime or failing to appear in court. Judges can still make their own bail decisions if they don’t agree with the system’s results.

Instead of the Sentence-o-Matic 1000, it’s the Bail-o-Matic 1000.

The new scoring system, developed by the Laura and John Arnold Foundation, uses nine factors to estimate risk, including a defendant’s age, whether the charge is a violent offense, prior convictions and previous failures to appear in court. It does not take into account race or gender, and judges still have the final say in setting bail.

Before you point out the obvious, that the scoring system says it doesn’t take race or gender into account doesn’t make it so. Both prior convictions and previous failures to appear are proxies for race, gender and poverty. Is that a problem?

This tool is in use in about 30 other cities and states, several of which have seen encouraging results. After Lucas County, Ohio, which includes Toledo, began using the Arnold Foundation tool, the share of defendants released without bail rose to 28 percent from 14 percent. The percentage arrested while out on bail fell to 10 from 20, and the percentage who failed to show up for trial dropped to 29 from 41. The number of people jailed in Mecklenburg County, N.C., dropped by 20 percent in the year after the county began using the tool. Crime did not increase.

People who love algorithms also love statistics. And from these stats, it would appear that use of this system shows better results than before. Of course, since we have no clue what generated the “before” results, all we know is that it’s better in the aggregate. The stats are undifferentiated, so they may be skewed by the high number of disorderly conduct arrests, where people are being released, while the far lower number of serious felony charges aren’t producing the same numbers or outcomes.

The use of scoring systems in criminal justice has raised concerns in the past. An analysis by ProPublica showed that an algorithm developed by the company Northpointe to aid in sentencing decisions wrongly predicted that black defendants would commit future crimes almost twice as often as it made the same mistake with white ones. While any algorithm that relies on criminal history could replicate inequalities in the criminal justice system, the places using the Arnold system have not found the same race-based problems.

This is classic question begging. The statistics seem to show that its good in the aggregate, but then, the criteria used have always been available for judges in making a bail determination, plus whatever arguments are raised at arraignment addressing the specifics of the case, the defendant, his history, etc. What makes these nine factors the solution when they weren’t the solution before?  The only addition to the process is scoring, the basis for which is unexplained. Is that the missing fix?

Basing bail decisions on a cold assessment of risk brings the system back to its original purpose, proponents say — keeping those who could be dangerous off the streets, regardless of their ability to post bail. Still, the rollout of the algorithm hasn’t been without tension, and some are frustrated with what they are seeing in San Francisco.

Prosecutors and defense attorneys say they have encountered results from the assessment they do not agree with. Meanwhile, they say, some judges are often refusing to follow the release recommendations.

Policy wonks love an approach that gives the appearance of working in the aggregate. Empiricists wrap themselves in statistics and feel all warm and comfy. Trench lawyers concern themselves with one life at a time. So what if 99 people were happy, when you’re standing next to the 100th person? What if that 100th person was charged with a violent crime that ups his score because the cop who arrested him knows that it will prevent him from getting bailed out, even though there is no evidence that he committed the crime?

The relief valve is that a judge can overrule the scoring system, so the judge can take anomalies into account. But then, the judge is back to exposing himself to the lynch mob if he makes the wrong call.  And if the judge was inclined to take such arguments into account in the first place, there would never have been a need for an algorithm to fix the problem and replace judicial discretion.

What the scoring system does is provide judges with an objective factor to which they can point to cover their butts for failure to have the guts to do their job. Whether the system is good, bad or otherwise, the alternative of judges who refuse to take a risk, and a media that will crucify a judge if a released defendant commits a serious crime, gives rise to systems with built-in excuses.

Maybe this scheme is good. Maybe it’s awful, but its one clear virtue is that it gives judges cover to avoid the townsfolk and their pitchforks.  Is that really the best the legal system can manage? Maybe so.

*In New York, the sole criterion for bail is to assure the return to court. Former Chief Judge Lippman, as part of his solution to reduce the number of people held in lieu of bail, proposed adding danger to the community as an additional reason to impose bail. If this confuses you, since it would have the exact opposite effect, you’re not alone.

**A fixed bail scheme is a list of offenses with presumptive bail based on the nature of the offense alone. In other words, a person charged with Crime A would have bail set at $1000, regardless of whether the defendant was the Pope or Charles Manson.