Torin Monahan published a fascinating study of “Flock on campus: university police as appendages of a national policing apparatus” (OA) in Policing and Society. Monahan finds that at least 150 institutions of higher education employ Flock cameras. The University at Buffalo is preparing to add itself to that list, at least as a trial (and with the current provision that it would opt out of Flock Safety’s national network). Nationally, campus police include over 17,000 “sworn officers,” the vast majority of whom carry firearms and are authorized to use deadly force. As Monahan notes, for example, “the University of Chicago Police Department, which, because of its location in a dense urban environment, ‘controls one of the largest private police security forces in the world with a jurisdiction of over 50,000 nonstudent residents, second largest to the Vatican City’” (5). The point is that this is significant policing and campus police interactions with Flock Security are noteworthy. That said, I’ve recently written here about Flock, so I have a different angle to explore today.

The decision to trial Flock Security cameras on campus materially intersects with the university’s claims to pursue AI for social good. UB’s central university and decanal websites, along with its “AI and Society” department, clearly represent the institution’s public position on AI. What does that representation look like?

• Leading AI to Good: ” Whether artificial intelligence will change the world is no longer a question. Now, it’s just a matter of how.”
• UB’s Expertise in Artificial Intelligence: “The University at Buffalo is a leader in realizing AI for good.”
• Here is How Determination Drives Discovery: “as home to Empire AI—New York State’s $500 million, public-private consortium—we’re poised to advance “AI for good” long into the future.”
• Department of AI and Society Mission and Values: “AIS is dedicated to harnessing artificial intelligence for social good,”
• “New Office of AI Innovation:” “AI touches all of our lives, and the potential for good is immense.”
• Research, Innovation, and Economic Development: “AI at UB starts and ends with people tackling real world problems for progress, not profit. AI to empower, not replace, and enable innovation that puts everyone first. Learn how UB is using AI responsibly to create a bolder, better, brighter future for all.”

In short, the message is consistent: we do AI for good. UB is not simply saying that researchers can use AI to accomplish socially beneficial things. We do that with books, microscopes, spreadsheets, cameras, and other technologies without declaring that we do “books for good.” “AI for good” makes a stronger claim: that the development and expansion of artificial intelligence can itself be directed toward a socially good future.

So how about the case of Flock cameras on campus? Is that AI for good? Perhaps. A student whose stolen car is recovered might reasonably say so. Police officers who solve a hit-and-run might say so. Flock Security and its investors presumably think so. But this only raises the question that “social good” is supposed to answer: good for whom, under what conditions, and at whose expense?

A simple machine example might be helpful. An inclined plane is useful because it makes ascent easier. But why is ascent “good”? Clearly it isn’t always good. In fact, we also use very steep inclined planes as obstacles (i.e., walls). Are walls a social good? Conceptual inclined planes work similarly. The upward arrow of technological progress transforms the difficult question of whether this is a desirable direction into a series of smaller questions about useful applications. As such, the inclined plane of progress is also the slippery slope and Sisyphus’ endless uphill slog.

Any of us might find a use for an inclined plane. Equally we might find a use for AI. We might say “AI is good for doing _____.” That doesn’t make AI good of course. Some might say a gun is good for self-defense. That doesn’t make guns themselves good. And we might dispute that claim about self-defense such that we cannot say it is universal. All we can do is assert that some people think this technology is good for that purpose. If we gathered up all those individual claims into a heap would the determination “AI is good for society” emerge? That is, is it simply a matter of piling up use cases?

Or perhaps we are already on a slippery slope constructed from prior investments and commitments, including these public statements. Beneficial applications become evidence of AI’s capacity for good. Harmful applications become evidence for the need for more governance. Because once we are slipping and gathering speed, the outcome is quite beyond us; we are basically just falling, depending on the incline.

Part of the issue is the methodological boundedness of research. A researcher could employ AI to produce scientific knowledge or develop a use case for AI in a specific technical instance. As these would represent disciplinary advances, they might be termed “good” by some communities. But improvements in mathematical and coding capabilities can also increase offensive cybersecurity capabilities. And improvement in AI’s ability to identify and design organic compounds is surely as dangerous as it might be beneficial. Certainly these fields address these questions, but they are difficult to address at the level of research production itself.

The result is that “AI for good” becomes an elusive concept, as elusive as AI itself. Indeed, we might think of this as a Sisyphean wicked problem. All this does is put us in a different relationship to that “inclined plane for good.” With AI technological churn, we find ourselves time and again at the bottom of the hill with frontier AI labs racing ahead of us (or claiming to do so). Each time, we might try to push the boulder of “for good” up the hill again. Setting aside true believers of frontier AI fortune-telling, no one believes that “AI for good” will just happen on its own. To the contrary, doing the work on this goal is the AI-Forward university’s value proposition. But it is a wicked problem in the sense that developments in AI technology are built upon the technical debt of existing infrastructure. Each iteration is a variation on a prior one: turning that variation continually into progress toward the good is Sisyphean.

The prospect of Flock Safety cameras on campus demonstrates this. We can argue that they have use cases that are good for society. We can also argue that these same use cases are not good for society. Furthermore, we debate the “goodness” of surveillance technologies and the particular goodness of Flock Safety as an actor in our society. However, the UB has committed itself to helping realize a future in which the development and expansion of AI becomes a social good.

From this institutional position, can UB assert that using AI to surveil its students, faculty, and staff is not good? Obviously that would entail more than 20 Flock cameras at campus entrances. What would allow UB to conclude not merely that Flock needs safeguards, but that this particular expansion of AI-enabled capacity should not occur? Or is it limited to saying “we can use AI (for surveillance) for social good”?

And if that is the case, how would we characterize that intellectual position? I’ve already suggested “limited.”

Which leaves us with a final image of the inclined plane. This is the inclined plane wrapped around a cylinder to create a screw. As faculty we might find ourselves on the slippery slope of an institutionalized ideology of “AI for Good.” As we are called, time and again, to climb that inclined plane toward this image of goodness, we might recognize the circularity of our passage. We are not merely going up and down. We are also going round and around a central pillar of “AI for Good.” Each turn of the screw we take–pilot, survey, integration, expansion, experimentation–produces linear motion as well as our own circularity.

At some point, we might recognize that we are screwing ourselves, but maybe this time we are getting screwed for good.