In January 2024, Governor Hochul announced the Empire AI project as a consortium of public and private NY universities, state funds, and corporate partners. Fundamentally, Empire AI is a planned 15MW data center that is being built on the University at Buffalo’s North Campus. That data center will be located on a service road on campus and its construction, which is underway, is planned to be completed next year.

In addition to building the data center, Hochul later announced funds for the creation of an “AI and Society” department. As such, we should understand the department and the data center as two parts of a single effort. As I see it, this endeavor was always about building an AI data center for academic research purposes and the addition of AiS provided some social cover for the project. Though as I discuss here, that cover is slim in my view.

Poking around in public reporting, Empire AI is the only AI data center of its size currently that exists or is being planned/built on a public university campus. I should also point out that there are a few other public universities building data centers (UMich, TX, etc.) but nothing on campus at the scope of 15MW. UMich’s date center is planned to be larger but sited in Ypsilanti not Ann Arbor. UTexas’ Horizon project is sited in Round Rock and appears roughly the same size as Empire AI.

By UB’s reporting, Empire AI’s data center will increase electricity consumption across all 3 university campuses by 60%. The report is about their planned mitigation efforts. It addresses heat and electricity but it doesn’t mention noise. That said, I wouldn’t immediately compare it to all the familiar data center noise complaints we know. But the specifics don’t seem to be accessible. Maybe a FOIL request. Interestingly, I can’t find any public reporting of any consideration that Empire AI’s data center could be located off-campus in the Buffalo region. I’m not sure why.

It is also worth noting that the AI and Society department created by Hochul (and UB “decision-makers”) is quite unique in the US. The only possible parallel I have found is the department at SUNY Stony Brook that was created by the same Empire AI initiative, though they are repurposing an existing department of “Technology and Society” by adding AI to their name. UB’s AiS department has created a series of “AI-plus” degrees, which I have not found anywhere else in the AAU (Bowling Green may have something similar.)

In short, Empire AI is quite unique. Unique is neither good nor bad on its own. Many things are done only once or rarely because they are singularly bad ideas. Usually if something is unique and viewed as good, others start to follow the example. That hasn’t happened since Hochul’s January 2024 announcement despite the general high speed rush of higher education toward AI.

In other words, all the AAU schools are trying to be “AI Forward” but no one else thinks this approach is a good idea, apparently. My view is that Empire AI and its departmental offspring are just a bad idea and one that UB has not even begun to feel the cost for. For me, that is a separate matter from the broader notion of the role of AI in higher education. There are many ways to go about this besides the Empire AI/AI & Society approach, as the rest of the AAU (and the nation) demonstrates.

There are currently 8 different AI plus degrees at UB. Each includes a common “Society” track and one of three Technical track options. The Society track includes 3 required courses plus choosing two more from a menu of three.

AI 101LEC – AI & Society Credits: 3
AI 111LEC – AI & Ethics Credits: 3
AI 211LEC – Technological Disruption and Diffusion Credits: 3

Choose two of the following AI & Society Electives:
AI 321LEC – AI & Policy Credits: 3
AI 323LEC – AI and Social Structures Credits: 3
AI 325LEC – AI & The Information Environment Credits: 3

To date, these courses are assigned to faculty outside AiS, who clearly have a say in their own courses, but have not participated in the curriculum overall. I was chair of Media Study during the time AiS was formed and was directly involved in conversations with the principles involved. In addition to being chair, a number of us, including me, are advertised as researching AI but were systemically and intentionally excluded from these conversations. In short, my UB experience has been that the university has systematically narrowed the scope of the curriculum by limiting faculty participation to those who are willing to be on the same page with the Empire AI mission. Humanities faculty have been assigned (or perhaps sought out, idk) courses in the society track, but they have been given lino say in what the larger curriculum should be.

To get a sense of the intellectual quality of these courses, we can consider the course description for AI 211: “This course surveys the history of human technologies and the way that they result in disruptions to pre-existing social patterns, from the rise of farming to the invention of the printing press to the latest developments in Artificial Intelligence. The goal of this course is to help students understand and situate today’s disruptive technologies in their historical context” The neolithic revolution, the printing press, and AI: three terms that appear in a sentence. The description is a critical-thinking failure because it begins by classifying AI, farming, and printing as instances of the same historical mechanism: technologies arrive and disrupt pre-existing society. It thereby assumes the very historical status of AI that a course like this ought to investigate. And even if one wants to make that argument, one would be better off with the agricultural and industrial revolutions and saying AI was a culmination of an information revolution. That at least makes logical, parallel sense, even if it is intellectually indefensible.

Then there are the 3 technical tracks, which is where AiS faculty tend to teach, if they teach. The most technical track consists of the standard first two introductory CS courses, an accessible mathematics-for-AI course, and one applied machine-learning course. Considering a CS degree includes roughly 70 credits of CS courses, I’m not sure what you get from four classes. It is a baseline education without much specific application to what AI is right now in technical terms. Of course an education targeting current AI technical concerns would be worse. Why? Because the technical terms of AI are moving so quickly even experts struggle to stay current. The more durable these courses are, the less specifically they are about contemporary AI; the more specifically they address contemporary AI, the faster their technical content becomes obsolete.

As such, we end up questioning the value of a “technical track” and even how technical vs. general math/CS the curriculum is.

Currently only 1 of the 8 majors has students in the easiest technical track: “AI and Responsible Communication.” To quote them directly, here are some of the things you can do with the degree:

• Teaching the public how to tell what content is real.
• Creating better conversational flows for AI models.
• Designing voice assistants.
• Implementing AI-driven marketing campaigns.
• Developing AI ethics policies for government agencies.
• Working with researchers on natural language processing tools.
• Finding ways to integrate AI into traditional communication platforms.

Setting aside the first outcome (which is quixotic anyway), the rest are about training to be a reverse centaur. Rather than a machine adapting itself to augment human capacities, the human is trained to reorganize communication practices and institutions so that AI can operate successfully within them.

The other seven degrees require more technical preparation. Three use the intermediate track and four use the CSE/Applied Machine Learning track. But none amounts to a technical specialization in AI; these are 16-credit service cores appended to the X disciplines. While I do think that general education math and science curricula might be revised to include technical discussion of AI, I don’t think adding these courses on top of existing Gen Ed does much. I reject the assumption that an undergraduate education about AI must begin from introductory CS and mathematics. That is one legitimate route into AI, particularly for students who intend to build computational systems. It is not a prerequisite for studying the history, politics, aesthetics, rhetoric, economics, ethics, infrastructures, cultures, or social operation of AI. As someone who has been researching AI for more than 25 years, I refute the notion that introductory courses in computer science and math are integral to such an education. That’s one legitimate academic path, but this coursework is not a prerequisite for studying the history, politics, aesthetics, rhetoric, economics, ethics, infrastructures, cultures, or social operation of AI.

Indeed this is the whole Empire AI and Society situation. Do UB faculty need access to AI to do research? Some. Should UB faculty study AI? sure, some. Should UB teach AI as a subject/topic in courses? Of course, in some courses. Those aren’t the questions here. The main question is whether or not this approach is a good idea. I don’t think it is. I’ve proposed other approaches. Other universities are demonstrating other approaches.

AI and Society provides the intellectual, ethical, and public-facing legitimation for the much more material commitment represented by Empire AI. The infrastructure came first. “Society” followed as the explanation/excuse of why that infrastructure will serve the public good.