Anti-Gun Academics Claim AI Can Identify ‘High Risk’ Gun Dealers

It remains to be seen if artificial intelligence poses an existential threat to humanity or if it will be a boon to human existence, but the technology stands to impact our lives in countless smaller ways as well. If anti-gun researchers in California have their way, for instance, artificial intelligence may soon be used to target federally licensed gun dealers.

Researchers at UC-Davis in California claim in a new study that AI can predict which dealers pose an “elevated risk for selling the largest number and fraction of firearms recovered within a year of sale.” Hannah Laqueur, an associate professor of emergency medicine at UC Davis who headed up the study, used two different prediction models to come up with her list of bad apple dealers. Both models essentially used a short time-to-trace to identify dealers that are supposedly at a high-risk of selling “crime guns.” 

The models generally outperformed simpler regression and rule-based approaches, underscoring the value of data-adaptive models for prediction. Key predictors included prior-year crime gun sales, the average age of purchasers, the proportion of “cheap” handgun sales, and the local gun robbery and assault rate.

Many of the dealers with the highest predicted probabilities not only sold a large volume of guns with very short “time-to-crime” but also consistently sold crime guns over multiple years. This suggests that a relatively small group of dealers could be targeted for enforcement, offering the potential for outsized impact. More consistent and targeted inspections of high-risk dealers, along with citations or license revocations, could strengthen deterrence and promote compliance, helping reduce the supply of guns to offenders.

“Our findings show how machine learning techniques, combined with California’s comprehensive firearm transaction and crime gun recovery data, could help identify potentially high-risk retailers,” says Laqueur. “This type of identification can improve the efficiency and effectiveness of inspections and enforcement efforts aimed at interdicting negligent or corrupt dealers.”

Despite Laquer’s claims, there are some major limitations with the study, including this little gem: 

the authors note that dealers selling many short time-to-crime guns may not have violated the law and conversely, non-compliant dealers may not be reflected in short time-to-crime statistics.

In other words, the data may not tell us anything at all about a particular dealer. 

Not every gun that’s traced is used in a crime, and not every gun used in a crime is traced by law enforcement. FFL’s with a high volume of sales are also more likely to sell guns that are later traced, but isn’t evidence of any wrongdoing on the part of the gun dealer. 

In the 2022 ATF trace data for California, the largest category of traced firearms was for “possession of weapon”, which accounted for more than 22,000 traces that year. “Weapons offense” was the second most common reason for a trace, with 8,813 traces performed in California that year. “Found firearm,” “carrying a concealed weapon,” and “firearm under investigation” were the next most common categories of traces, and it’s entirely possible that the majority of those traces were ultimately not connected to any crime at all. 

Given the buzz around AI (and blue state hostility towards our Second Amendment rights) there’s reason to be concerned that law enforcement in states like California will use artificial intelligence to suss out bad apple dealers, even though there are some fundamental flaws in the UC-Davis research. There’s a real risk of falsely identifying a “high risk” dealer, but I suspect that doesn’t matter much to anti-gun advocates. After all, they’d be happy if every FFL was shut down, no matter their practices.