One of my favorite computer geeks, Jake DiMare, posed a question the other day in response to the confusion over the language of 18 U.S.C. § 3553(a), which provides the considerations a federal judge is required to take into account in imposing sentence. The law includes both the parsimony/Goldilocks clause, that a sentence shall be no greater than necessary to accomplish the stated goals, and the list of goals.
Judge Richard Kopf raised the question of whether the language was so devoid of meaning and guidance as to render § 3553(a) worthless. Judge Mark Bennett responded “Sentencing requires us to weigh that which cannot be measured,” to which Judge Kopf replied: “Let’s be honest then and declare that sentencing is entirely a matter of discretion…” If so, this raises the specter of sentencing being so arbitrary and capricious, so captive to any judge’s whim, as to be a total crapshoot.
Jake offered this reaction:
Was there ever a task in the courtroom more ripe for automation?
He went on to explain:
I can tell you those of us in the tech field are programming computers to do some pretty fantastic things in the pursuit of deeply personalized experiences…And there will come a time when somebody has created the sentence-o-matic 1000…Even capable of accepting and calculating the value of mitigating circumstances.
As a representative of the ignorant masses, I find comfort in the notion that everyone would be given sentences using the same criteria, and never again subjected to the whimsy of some of the judges.
While I dismissed it tersely in the comments, it wasn’t because the point wasn’t good, but the response went far beyond what could be offered off the cuff in a comment. Rather, it was a serious enough question as to require a more detailed response.
This “problem,” that judges impose disparate sentences on seemingly like-situated defendants, has long been a vexing problem. It was one of the foundational arguments for the Sentencing Guidelines, to create greater consistency in sentencing across the country, so that a judge sentencing a defendant in a drug conspiracy in Wichita would impose a sentence reasonably the same as one in Brooklyn. Consistency was the goal, and from a substantial distance, it appeared to achieve that goal.
The problem was that it failed miserably to accommodate the myriad personal details that comprise the heart of sentencing. Indeed, it precluded judges from doing so, forcing lawyers into striving mightily to come up with arguments about why their defendant’s circumstances fell outside the “heartland” of the guidelines. Most of the time, these arguments failed. One size fits all sentencing was imposed, and those who feared mercy slept well at night.
Is there much to hate about leaving sentencing to the whims of judges? You bet. But the alternative to bad isn’t necessarily good. On the other side, it can always get worse, as the continuum goes both ways. This is where the Sentence-O-Matic 1000 comes in.
How much do we all love the algorithmically directed advertisements rammed down our throats on websites connected to Google? Not only do they suck, but they’re misdirected most of the time, despite huge access to our data which ought to allow Google to do a far better job of it. Perhaps Google doesn’t find it worth the effort, or perhaps this is the best it can do at present.
But the point is that Google (just an example, don’t get caught up in Google details) can only use past data to drive future predictions, but it does so, by definition, without context, which similarly, by definition, means that it’s just a stab in the dark. That may be good enough for marketing, but sentencing is a bit more serious an endeavor.
Ask anyone engaged in criminal law what criteria should be considered on sentence, and the only honest answer is that the devil is in each individual defendant’s details. Sure, there are the surface measures, the nature of the crime, for example, but that’s just where the inquiry starts. It ends up in a different place in every instance, because every person is different, lived a different life, suffers different causes for different reasons. Whether there are 20 or 20,000 criteria for data input, they differ from person to person.
Add to this morass that the interpretation of the criteria varies enormously by nature of the person inputting the data. Who quantifies sincerity, remorse, intellectual ability, sociopathy, traumatic brain injury, childhood suffering, just to name a few obvious issues? The ability to convert even the obvious concerns into quantifiable measures renders the output suspect. GIGO.
While faith in the utility of technology by its fanboys is understandable, given that’s what you work with, rely upon and have chosen to dedicate your careers to, you can also appreciate that there is no arguing with binary thinking. Ultimately, it must make choices, based upon inputs, then crank out an answer. So while it may well produce consistency, a la the Guidelines, its answers will suffer from the same arbitrary and capricious problems as they did under the Guidelines, and the opposite (but equally whimsical) answers judges provide under § 3553(a).
As Judge Bennett noted, “Sentencing requires us to weigh that which cannot be measured.” If it can’t be measured (and it can’t), then it can’t be input into the Sentence-O-Matic 1000. And even if it was input to the best a bunch of programmers could manage, it couldn’t be subject to the arguments that real life human beings require to address the individualized circumstances that exist in every case and every sentence, because that would obviate the goal of consistency and return us to the whimsy of the judge.
While sentencing based upon a judge’s assessment of the facts, circumstances and arguments of each case and each defendant may be a terrible method, fraught with caprice, personal prejudice, and flagrant inconsistency, it would be unfathomable that any criminal defense lawyer would prefer a system that precluded any argument for leniency based upon the individual defendant before the court.
So we’re back to the old adage, sentencing by whim of an individual judge is the worst system possible, except for all the others. Or as Mencken said, “for every complex problem, there is a solution that quick, easy and completely wrong.” I hate the capriciousness of judicial sentencing. I dread the consistency of the Sentence-O-Matic 1000 far more.
great post it had me wondering what would be your solution to the
kobayashi maru?
I love the kobayashi maru analogy. That’s the point, not every problem has a solution. Sometimes, were stuck doing the best we can under circumstances that will invariably suck. The best we can do is find the least offensive method, because there is no good answer.
If you wanted sentencing to become more consistent nationally, while still allowing for the understanding of nuances, you could build such a system.
One idea would be: Once the sentencing phase is reached, instead of using just the trial judge, have each sentence independently judged by 4 Judges that are randomly chosen across the country plus the original Judge. As long as all 5 reach a conclusion within a predetermined limit of difference, the sentence is the average of the group.
If there are 1 or more outliers, then the sentence would go to a second review. This might include having the judges meet virtually to understand the difference. If they cannot agree after review, maybe the process starts over with a new panel.
In the end, this would promote consistency across the country and keep a single judge from being too harsh or lenient. The problem is that it would be way too expensive as each judge would have to be involved in the entire sentencing process.
This was a post about the use of computers for sentencing. What part of that made you think I was soliciting random, untenable ways to achieve sentencing consistency?
I completely agree with a few things:
1. The stakes are exponentially higher in your world than they are in marketing.
2. The last few years of fumbling around with personalized digital experiences often leads to laughably inaccurate outcomes.
3. Therefore, automated sentencing is probably a bad idea today.
My theory was/is there will come a time when the hardware, algorithms, and the programming can be trusted to do a better job with sentencing than a judge. (And it’s worthy of conversation because this time will come well before many other roles fulfilled by a judge or jury could be handled in such a manner. With proper investment…Ten years from now)
But what’s more interesting to discuss is the first time a representative of the Sentence-O-Matic 1000 comes to a defendant about to face his destiny and says: Based on our analysis of your situation your outcome is likely to be X. However, if we align the color of your skin/sexual orientation/religious preference/hair color with this judge’s preferences your outcome is likely to be favorable in the following ways…I think you can see where I’m going with this.
If the technology got to a point where it could be demonstrated the outcome for a specific defendant was likely to be BETTER with automated sentencing, would you at least support the notion of giving that defendant the ability to choose between the two?
No.
I can easily see how this will eventually happen, just like I can easily see Minority Report happening. Because technology can solve everything. Binary thinking is great where it works. The shame is binary thinkers are incapable of understanding this.
How well has the Intoxilyzor 5000 worked out? Technology is magic, trust it. Everything will work out. We promise.
What frightens me most about your comment is that it fails to reflect any grasp of the problem of measuring that which can’t be measured. You’re like the Reinvent Law folks, so filled with blind optimism that lack the capacity to appreciate the disastrous consequences.
I believe tech is a critical part of the future. I fear the fight will with those who love it so much they can’t grasp it’s functional limits, and won’t realize the depths of their error until you’re surrounded by ruined lives. Your inability to grasp even the issue here demonstrates the point.
Isn’t part of the problem that the participants don’t agree at the outset on the relative circumstances that should be considered? The prosecutor and report writer may not agree with my assessment of the defendant’s upbringing or mental health obstacles or any of the other criteria we talk about. That’s what the judge does – decide among our competing visions for the precise mix of rehabilitation/deterrence/retribution. The fact that I put rehabilitation first is itself a judgment that would be contested by many. And it would not necessarily be a mix that would or should be the same for all defendants.
Any computer program would require values to be attached to particular inputs, and those values would be subjective. I.e. – Defendant x has cognitive difficulties of 5 (some arbitrary number out of some arbitrary scale) and that itself is a subjective evaluation that has to be teased out of competing judgments.
Is it fair that one defendant is evaluated by a judge who might be more receptive to rehabilitative arguments than her neighbor the next court over? No, but at least we criminal defense lawyers can advise our clients of the risks based on our experience, and that too is an unfair source of risk to our clients.
Maybe its just my human-centric bias, but I would like to think that I can be more useful to my clients than a computer program. On the other hand, Big Blue could probably calculate a better strategy for defense than I can – as I am certainly no Grand Master. So I will revise my objection – if defendants can get the benefit of a full Big Blue supercomputer criminal defense, then maybe it will end up ok. But I would imagine that those investing the heavy money in “solving” this problem won’t have defendants’ best interests at heart.
Law is more art than science. Science has its place, and it’s a very useful place, but it’s still not art.
Why is consistency the goal? If judges are supposed to weigh that which cannot be measured, disparities in outcomes are to be expected.
A sentence-o-matic 1000 containing nothing more than a suitably tuned random number generator might pass the judicial equivalent of the Turing test: An observer could not discern whether the sentence came from a judge or the machine.
Since that approach is unlikely to please most people, perhaps the real requirement is not consistency, but that judges compose a plausible story about how they choose each sentence.
Consistency is not a goal in itself, but a proxy to show that sentencing is not a product of judicial whims. If the same defendant, under the same circumstances, was sentenced by ten different judges, the sentence should (within I suppose some reasonable parameters) be the same. If not, then sentencing would be arbitrary, and that would undermine the integrity of the process.
Since one defendant can’t be sentenced by ten judges, the system instead looks to ten defendant, using criteria that the government believes makes them comparable, to see if the sentences are consistent, to validate the methodology of sentencing.
I recently heard a federal probation officer’s presentation on “evidence based” risk assessment. Apparently those who have the temerity to ask to remain free until proven guilty will get a risk score based on things like whether he completed high school or is involved in anti-social behavior.
In general, I believe there should be an empirical reason for the actions we take. I’m not opposed to “evidence based” practices, but my experience with bureaucracies tells me this is likely to be a nightmare. It will evolve into “everyone 5 and above is locked up,” “3’s and 4’s get home incarceration,” “1’s and 2’s get released on conditions, unless they display a poor attitude.” For snitches, the government will not oppose release on conditions regardless of the score.
The current advisory guideline regime may be the best we can achieve. The guideline range is a suggestion, maybe even the preferred option but deviations are possible.
I’m with you until your last paragraph. The guidelines regime is based on a fiction, that the numbers are somehow empirically based. They never were. They aren’t. Rather, they have become the “norm” by pre-Booker mandatory fiat. So we’re starting with a post-Booker stacked deck of suggestions, skewing the “norm” far upward from where it was pre-Mistretta.
Pingback: Is The Legal Field Ready for CSO?