Last week, I blogged about law enforcement’s use of automated predictions. There, the “Super Cruncher” system mines data to highlight high-crime areas so that police departments can best allocate resources. What if those predictions provided the sole basis of an officer’s stop and frisk of a particular location? Suppose the computer suggested that a particular corner was a red-hot zone. When the officer saw someone standing at that particular corner at midnight, he took the computer as its prediction and stopped and frisked the person, revealing an illegal firearm. Would the computer’s prediction form the basis of reasonable suspicion supposing that the person standing on the corner did nothing else to raise any concerns about illegality? Last week, I suggested that the retail question would likely be straightforward. The computer’s prediction about a location could not be said to infer anything revealing about a particular person in that location, right?
Professor Orin Kerr brought a recent case to my attention that while not exactly on point is nonetheless illuminating about the value of automated judgments in evaluating a stop for Fourth Amendment purposes. In United States v. Antonio Esquivel-Rios, a trooper pulled over a defendant driving a car with temporary Colorado tags. When the trooper initially called in the tag, the dispatcher told him that the automated system found that the tags were not registered (as the dispatcher explained, the system did not “return the tag”). The dispatcher also cautioned the trooper that Colorado tags “usually do not return.” Said another way, the dispatcher qualified the system’s finding that the tags were not officially on file (and thus could be fraudulent) with the warning that Colorado tags usually did not show up in the system. Why that was the case for Colorado tags was not explained to the trooper. Nonetheless, the trooper pulled over the defendant and got consent to search the car. It turns out the defendant had a pound of meth in his secret glove compartment. In challenging the constitutionality of the stop, the defendant argued that the trooper relied on an unreliable automated finding that could not support a finding of reasonable suspicion. Said another way, the computer’s “no tags” determination did not amount to particularized suspicion because the system’s findings as to Colorado tags was not reliably revealing of criminality.
The opinion began by noting that a “maniacally all-knowing, all-seeing” HAL 9000 computer in government’s hands would raise Fourth Amendment concerns. The Tenth Circuit did not say more about that point, but I take the court to be saying that computers making “pre-crime” Minority Report-ish adjudications about individuals implicates constitutional concerns–procedural due process is certainly at issue. After making that threshold point, the court then got down to business to explore whether the trooper had reasonable suspicion to stop the defendant based on the computer’s “no return” finding and the dispatcher’s qualification of that finding. As the court explored, reasonable suspicion is far less than probable cause, there needs to be some particularized suspicion of criminality. Concerns about the quality of evidence can be offset with quantity, that is, something more suggesting criminality. If there are questions about the system’s reliability, worries about its reliability can diminish if there are other independent indicia of criminality. The trooper, however, only relied on the database report to justify his stop. The computer “no return” hit, the court suggested, could have been enough for reasonable suspicion if the system was reliable. There, such a computer finding would concern the specific individual, not a particular location as I suggested in my initial post. The court’s point is well-taken. In that case, it would have been permissible to rely on computer finding to support a stop because the computer’s finding would relate to evidence about the specific defendant (or his car). In this case, the court explains, the trooper had reason to doubt that the computer hit meant something suspicious about the car’s tags. That Colorado usually does not return hits could mean that Colorado is having bureaucratic problems inputting temporary tags into the system; it could mean that some, most, or vanishingly small number of “no return” findings say something about the tags’ verifiability. What goes into the database impacts the reasonableness of the seizure relying upon it, garbage in, garbage out. The court notes, relying on Professor Kerr’s work, that reasonable suspicion is not a statistical determination, much as probable cause isn’t. But in this case, the database had reliability problems and as the sole reason for the stop, it had to be assessed with a eye to its statistical value. With its concern about the computer finding’s reliability made clear, the court remanded the case to the district court to reconsider the constitutionality of the stop and the evidence found as a result of the stop. The Tenth Circuit’s finding makes a lot of sense, indeed. It also suggests that computer adjudications have to have an indicia of reliability and must relate to a specific individual (rather than location) to support reasonable suspicion.