In the initial post in this string, I laid out my thoughts like this:
If we ask how we ought to change criminal adjudication once we achieve artificial general intelligence (AGI), might we learn things we ought to change—and could change—right now? What might we learn from a lens of AGI even if such technology never comes to pass? The achievement of AGI would force the questions of in which roles ought humans to remain, and in which roles ought humans to remain exclusively (whereas now we have no serious choice). Those questions of replacement, in turn, require that we confront what humans ought to be doing, now and today. Thus, the questions might ‘out’ faults in our current practices. And the real ‘magic’ would be if solutions to those thereby-outed problems, or at least genuine improvements to those thereby-outed problems, are available without those question-spurring, technologically-uncertain intelligent machines.
I’ve done some thinking about what people ought to be doing in criminal justice, interrogating what, if anything, about human decision-making is of deontic importance in the space. My conclusion, along with Kiel Brennan-Marquez, is one of ‘role-reversibility’—that it matters whether a party exercising judgment is reciprocally, meaningfully vulnerable to that same judgment. That makes me a strong proponent of layperson-controlled criminal justice, which aligns with the American founding desire but is nearly extinct in our day-to-day adjudications. So, I have argued for trial lotteries and jury vetoes (a recent bit of empirical work on the latter is here) that could return some measure of citizen control to our systems of criminal justice, and I’ve thought about whether intelligent machines could immediately serve as prosecutors (no) and defense counsel (yes), and about the proper place of executive pardons. What I aim to do next is to think more about the questions we ask of layperson participants in the rare occasions in which they are invited into our adjudications. I’d like more layperson adjudication… but what is happening there? Might it be a bit funky?
Imagine you are seated on a criminal jury. You will be tasked with returning a binary judgment of guilty or not guilty. “Guilty” is a commonplace word having strong connotations; it seems to reflect a complex (at least to me) moral judgment. Yet none of its commonplace meanings seems to correspond to whether the State has proven each and every element of a crime beyond a reasonable doubt, which is allegedly what the layperson jury is to do. Indeed, those jurors have to swear an oath to do just that. Why does modern criminal law go to great pains to elementally ask this ‘proven’ question—we certainly try to drill it into our students in 1L Criminal Law, and jury instructions do the same—when our juries return something altogether different? If we want proven or not proven, why don’t we ask for it? Yet in Scotland, where for historic happenstance a jury can return either ‘not proven’ or ‘not guilty’ to the same legal effect (acquittal), there has been push to eliminate the ‘not proven’ option, including because it seems to decrease conviction rates in prosecution of crimes like sexual assault.
Clearly, then, none of this is easy. Still, isn’t it quite strange? We order the jury—make it swear an oath—to do X, but then we ask it Y? Must jurors then individually and collectively lie? If a juror’s best sense is that a person is ‘guilty as sin,’ but the prosecution put on a terrible case, we make her report “not guilty”? Why not allow her to tell it like she thinks it is: “not proven”? Does this disconnect skew not-wishing-to-speak-dishonestly jurors towards inappropriate conviction? On the other hand, for the juror not enjoying moral condemnation, why isn’t “proven” just the thing? ‘I’m not morally condemning anyone,’ thinks the juror, ‘I’m merely performing the analytical role asked of me, and I believe the elements of the crime proven beyond a reasonable doubt—I see no plausible story in which this defendant did not do as the State claims. Proven.’ What does that have to do with “guilty”?
There has been some great writing of late on probability and verdicts, trying to figure out whether legal proof is, for example, probabilistic (multivariate?) or explanatory—looking to the Blue Bus’s probability paradox and elemental proof’s conjunction paradox. But my sense is that my questions are ‘lower level’ than that: yes, it would be nice to know how jurors are meant to determine ‘proven’… but is ‘proven’ the thing at all? Or is it ‘guilty’?
So, if we want to know whether to replace criminal jurors with genuinely intelligent machines, we need to consider what it is those criminal jurors are doing… and I’m not sure. Is there ‘method to this madness’ of instructing to do thing-one but requiring an answer to thing-two? An opaque—intentionally opaque?—method behind it all? Is it a ‘mistake of history,’ a jury as seemingly omniscient replacement for the ordeals when the church opted out, that has never been meaningfully reconsidered? Or am I making a proverbial mountain of the molehill: there is some simple and satisfying answer evading me—there is some good reason for having guilty and not guilty, rather than proven beyond a reasonable doubt (proven BARD) and not proven. Or, proven BARD, not proven BARD, and not proven. Or, proven BARD, not proven, and innocent. Or, proven, not proven, and refusal to convict. Or, proven, not proven, refusal to convict, excused, and justified. Etcetera.
And it doesn’t seem like the potentially ‘wrong questions’ stop there. Imagine you are seated on a grand jury. You will be asked to return a binary judgment of true bill or no bill, defined essentially as ‘yes, there is a fair probability this person committed a specified crime’ or the negation thereof. The curiosity here seems somewhat the opposite of that for trial jury: given the desire for layperson control and the reality of immense prosecutorial discretion, why are we asking this legalistic, historic/computational question when there is an obvious alternative? Namely, ‘Would you like to see this prosecution continue?’ In other words, why don’t we invoke layperson (group) prosecutorial discretion?
Sure, the grand jury is merely a preliminary stage, but we have stages for good reason. At the very least, if this alternative is something we wish to know, ought not we ask the question and act in some way accordingly? Even if, say, a probable-cause-found prosecution legalistically ought to move forward, could we not also ask whether it normatively should? In other words, no matter what we do with the intelligence so gained, don’t we wish to know? Or do we prefer to hide in opacity? And, if the latter, must we continue to do so if we achieve truly intelligent machines?
The more I think on citizen-controlled criminal justice, the more I wonder at how we do it. Perhaps it is merely that—but critically that—humans fall into Z when some want X and others want mutually exclusive Y. Z isn’t logically either one… but the illogical alternative is something camp ‘not Y’ and camp ‘not X’ can both abide.
Again, I don’t know. But I think it’s worth thinking more about, including because the questions will have a new salience if we achieve potential machine alternatives. Might we then be content with Z?
Are there other times in criminal justice that we seem to ask the facially ‘wrong question’?
Leave a Reply