Seth Stodder’s Weak Defense of ATS-P

Seth M.M. Stodder with the Center for Strategic and International Studies, a former director of policy and planning for U.S. Customs and Border Protection, has a piece in the Federalist Society's Engage magazine defending the Automated Targeting System — specifically ATS-P, which is a system for screening border crossers.

The piece starts with the gripping example of a man named Ra’ed Mansour al-Banna who was turned away from the U.S. border and later blew himself up in Iraq. DHS Assistant Secretary for Policy Stewart Baker (who undoubtedly now lives in fear of my relentless blogging!) told the story at a CSIS event in December 2006:

[In 2003,] he showed up at O’Hare International Airport in Chicago with a valid passport from Jordan, a valid visa to come to the United States to conduct business and he asked to be admitted. There’s no bar to his being admitted other than the fact that he had been selected for a second look by our Automated Targeting System. He was flagged as somebody who just ought to be looked at more closely.

And so one of the CBP officers did exactly that: interviewed him, asked him a bunch of questions about what he intended to do in the United States, and concluded, at the end of the day, he just didn’t like the answers. He wasn’t confident that this guy was going to live up to the obligations that we imposed under the visa and he said, I’m sorry, you’ve got a valid visa, you’ve got a valid passport, you’re not going to come into the United States, and he sent him back to Jordan. Eighteen months later, of course, he was in Hillah, Iraq driving the vehicle-borne IED.

Baker is a smart man and he chooses his words carefully. If al-Banna had been identified by ATS-P as a likely terrorist, Baker would have said so. But he didn't. He talked about visa obligations.

Maybe the system identified him as a potential visa over-stayer — he had lived in California for two years — and when al-Banna couldn't convince his interviewer otherwise, CBP excluded him. Maybe, as some news reports have it, CBP sent al-Banna back to Jordan because he falsified details on his visa application (after which he "became withdrawn, holing up in a makeshift studio apartment, sleeping late, and displaying a new interest in religion"). Others say that "Homeland Security officials had no reason to suspect that Albanna had become a terrorist." Until the full story has been examined, this is anecdotal luck at best, not proof of a successful system.

In his paper, Stoddard claims to take on criticisms I have leveled at the program — and some I haven't. Here's the relevant part of Stoddard's article:

Some have disagreed with the 9/11 Commission's assessment of ATS-P’s effectiveness in assisting CBP, and have asserted that ATS-P is simply ineffective. Jeff Jonas and Jim Harper of the Cato Institute have asserted that, in general, “[t]hough data mining has many valuable uses, it is not well suited to the terrorist discovery problem,” because of the purported absence of “terrorism patterns” which [sic] to draw strategic intelligence. During a panel at the Center for Strategic and International Studies (CSIS), Jim Harper applied this analysis to ATS-P as well. But Jonas and Harper do not appear to understand all of ATS-P’s functions—including its link analysis function, operationalizing specific tactical intelligence by drawing linkages between known facts (e.g., a credit card number used by a known terrorist) and travelers seeking admission to the United States (e.g., if the PNR on a traveler indicates that traveler used that same credit card number to purchase his ticket). To the extent this conclusion also is pointed at ATS-P, Jonas and Harper may be uninformed. Indeed, the ultimate testimony to ATS-P’s effectiveness is not al-Banna, but its continued use by CBP and CBP’s ongoing efforts to improve it.

(link added)

Now let's review what I said at the CSIS event:

The story of the suicide bomber in Iraq was gripping and thrilling, frankly, but I think it was an invitation to us to indulge in what’s known as the post hoc ergo propter hoc fallacy. That’s Latin for “because it followed in time, there must be correlation.” Because ATS existed, he was stopped at the border. It may be true, but [it's] not necessarily true. Had he gotten into the country, he would have done in this country what he was able to do in Iraq. Maybe true, but probably not true. The infrastructure isn’t here and the support isn’t here to be able to pull off that kind of thing. So it’s, again, a gripping story, but not necessarily a good basis for policymaking.

In addressing what ATS is, it’s a check against the no-fly list. I think most people are aware of that. Link analysis, it makes pretty good sense in many cases. [Baker d]idn’t address the question of the risk score, which is the most concerning, I think, to most people, for a variety of reasons. And exactly how that risk score is created isn’t known, and I imagine that Secretary Baker and others would refuse to tell us how that risk score is created because that would create a security breach in the system. But it’s precisely there that the capacity for rank unfairness in the system is created. And it’s a system that doesn’t just apply, as I understand it, to foreigners coming to the country, but to everyone traversing the border, and that’s – I'm sorry to be so parochial, but I’m most interested in the rights of American citizens who are traveling internationally and returning to the country.

(emphases added)

Now, it's true that I am less informed about ATS-P than Stodder and Baker. Homeland Security folks hold inside information and they try to use secrecy as a trump card. My oral recitation of ATS-P's details lacks polish, but I know enough to have specifically approved of link analysis while disapproving predictive data mining. In our paper, Jeff Jonas and I excluded link analysis (referred to as "subject-based analysis") from our criticisms. Stodder refutes an objection I did not make, suggesting that I'm uninformed.

And he does not address the objection I did make, based on the paper Jeff Jonas and I published: predictive data mining won't catch terrorists.

His evidence that ATS-P works?: "[I]ts continued use by CBP and CBP's ongoing efforts to improve it." It takes several logical leaps and generous inferences to make that good evidence.

The only other successes with finding immigrating terrorists he cites beyond al-Banna (if indeed he was a terrorist at the time) are "Millennium Bomber" Ahmed Ressam and Mohammed al-Qahtani. Those two, though, were picked up by alert CBP officers unaided by ATS-P (so far as we know — and one expects we would know).

Two terrorists that perhaps should have been picked up by ATS-P but weren't are Nawaf al-Hazmi and Khalid al-Midhar, 9-11 attackers who entered the country and lived openly in the United States even though they were known to be linked to the bombing of the U.S.S. Cole. Before U.S. authorities failed to look for them, ATS-P failed to pick them out for additional questioning at the border. That's typical of data mining for terrorism: high false positives and high false negatives.