The rhetoric of the higher education industry regarding artificial intelligence operates by the following premises.

1. AI will transform society on a consequential scale.
2. Universities must adapt to substantial social change.
3. Universities have a responsibility for the public good.
4. Universities must respond to AI by participating in its development, infrastructure, and governance.
5. Therefore universities must transform to use AI to contribute to the public good.

While each point might be argued, I think one is clearly not like the others. In this discourse, #4 is generally an unspoken and bridging premise. It is also fait accompli. However, it is absolutely necessary because the first three premises still do not add up to conclusion #5. In fact, one could just as easily reach the opposite conclusion: that therefore universities must resist transformation by these terms. But you can’t make that argument after you’ve already invested in the infrastructure.

A version of this argument appears in The Buffalo Statement for Public AI. The statement was created at a conference held on campus during the summer. I wrote about the conference earlier. The Statement hits the first premise in its preamble, where it cautions that “decisions we make now will shape the future of humanity perhaps more than at any point in human history.” That sounds consequential.

As for the requirement to adapt, the Statement opines that “The levers of change for human development might come from the ways these new technologies first change the world around us.” Yes. How that all comes about (to the degree that it does come about) is exactly my concern. The Statement also wishes to ensure that “durable, human skills, human experience, creativity, and failure remain part of our research and educational enterprises. This is the source of our humanity, creativity, and ability to discover.” I agree there are technological conditions, and I share the intent of protecting our humanity.

But neither of those statements result in the conclusion that we should engage in the public operation of AI.

Nevertheless, the Statement simply asserts “we need public AI infrastructure.” But why? The statement argues,

We must ensure that universities are understood as public infrastructure and not private goods by working toward the public good. Public AI infrastructure must be built in partnership with universities to ensure transparency and sustainability and to incentivize public-private partnerships that support a broad view of economic and human development.

The list mixes procedural values, institutional practices, and desired outcomes without explaining how they constitute or produce the public good. They sound like headers in an operational constitution for a “public infrastructural” AI agent. These cybernetic guidelines for an undefined journey do not tell us where “good” lies or how it is discovered or produced.

In further defining these processes, the Statement claims a special role for universities in ensuring transparency and ethics.

As developers of AI, universities can help build technical and ethical guardrails… Public infrastructure such as Empire AI commit to the principle of AI for good and therefore provide opportunity for AI governance development with transparency.

Trust, as was noted in Buffalo, is why we insist on humans in the loop. Not because humans are more capable or less error-prone than AI, but because we trust in our shared experiences and in the accountability of human communities. (my emphasis.)

Can universities build guardrails? Yes. But there is a bigger unanswered question.

What is “the principle of AI for good”?

How novel is this principle? Are there many people out there saying let’s do AI for evil? Aside from promising that we will try to be decent humans, what does this mean? What do they want? A medal for not being evil? The insistence on “humans in the loop” is tied to this promise. And the rationale offered is that we trust in the “accountability of human communities.”

I also suspect that we can hold humans accountable for AI operations by putting them “in the loop.” This is a classic example of Cory Doctorow’s “reverse centaur.” AIs can move very fast, but we need to make them “ethical” too, so we’ll put a human in there to make sure. Of course, the human review is the slowest-moving task in the AI’s workflow. We might imagine that its harness starts to chafe while waiting for the human process to complete. As such, that reverse centaur is going to need to move as quickly as possible while remaining accountable.

This is AI ethics. Or at least a common version of it. So I am still unsure what “the principle of AI for good” means.

The Statement’s final section circles back to its mission statement: “Private interests have dominated the conversation and agenda for AI. Public involvement must help steer the development of AI to strengthen the fabric of society.” And yet, how can universities accomplish this task?

We can’t out infrastructure or outspend or outpace big tech or the technologies themselves.

We need to out purpose them. Our mission is not their mission.

Yet different purposes need not be competing purposes.

So this sounds to me as though they claim that universities should steer the development of AI by adopting a purpose that does not compete with Big Tech.

hmmm.

The Statement ends with the assertion “Public AI requires public institutions and universities willing to put themselves at the center of shaping AI for the public good.” There’s a definitional circularity when AI becomes public through the participation of public institutions who participate because “public AI” requires they must. But to what end?

I see public research universities aligning themselves as “AI forward” in response to a clear signal. They aren’t acting for the same reason that investors are, but they do share a common belief that we are verging upon a culture of continuous inference, as I term it. Continuous inference describes a cultural condition in which AI inference has become so inexpensive as to operate as a background infrastructure. As such, continuous inference would be nearly as ubiquitous as electricity. Perhaps this is the technological and infrastructural scale of the world the Statement anticipates when it says that “decisions we make now will shape the future of humanity perhaps more than at any point in human history.” That’s unclear in that Statement, but that is the economic aim of Frontier AI to which they have attested they will not compete.

In that context, there is one last significant elision. Even if we can come to a definition of public good, why should we assume that an AI infrastructure will produce it? And even if it can produce good, does the amount of good justify the other effects of AI? Shouldn’t we consider this matter before we dive in to support Frontier AI and promise we will make (non-competitive) good out of it?

Too late.