As WashU expands its investment in artificial intelligence, questions about the technology are moving beyond whether students should use ChatGPT for an assignment. The University is building AI into courses and research, providing students with enterprise AI tools, and creating a new senior administrative position devoted specifically to the technology.

Student Life (SL) spoke with Nathan Jacobs, WashU’s new vice provost for artificial intelligence; Betsy Sinclair, vice provost for innovation; and Andrew Butler, vice provost for undergraduate education, about what those changes mean in practice. The conversation has been edited for clarity and brevity. 

SL: WashU already had AI work happening through individual schools, WashU Information Technology, and groups like the Center for Teaching and Learning. What does having a vice provost for artificial intelligence actually change? What decisions are now yours, and what decisions are not?

Jacobs: It’s maybe hard to draw really precise boundaries around the decision-making authority. There’s a lot of work that’s happening across the institution in AI, and there’s also an uneven application of some of that.

The realization was that we needed to have a central place to advocate for the needs of faculty, both for their research and for teaching. Essentially, that’s what my role is: to be an advocate for teaching and research needs across the institution with respect to AI. That involves partnering with schools that are already doing things and partnering with people like Betsy and Andy whose work also touches this space.

SL: How much is WashU actually spending on its AI initiative?

Sinclair: You’ve asked a complicated question. We’re in the early days, and I don’t think we know yet.

We do things in beta, and then we come back and take a look. Last January, we started by asking what we could give students essentially for free, which is how the campus community had access to Gemini. Over the spring, we held listening sessions with students and faculty and the broader research community. Faculty and students came back and said they needed more, but they also cared a lot about responsible and safe use. They didn’t just want a tool turned on; they wanted an educational framework around it.

A lot of the work also involves reallocating the time of people who already work here — people from the library, WashU IT, the Center for Teaching and Learning, and DI2. There was an expense associated with paying faculty for their time; 88 faculty worked over the summer to develop new courses or modules within existing courses. But everybody else already worked [summers] here. So when you ask about cost, part of it is really an allocation question, and that is difficult to put a dollar sign around.

We are trying to measure the impact of these programs, and those measurements will help guide how we allocate resources going forward. 

SL: So the University does not yet have a total figure for what it is spending on the initiative?

Sinclair: Not yet. A lot of these programs are still in pilot phases, and the learning-outcome data are still being collected. 

SL: WashU initially emphasized ChatGPT Edu and now describes Claude as a preferred platform for students. Why the shift, and how do you keep WashU from becoming dependent on whichever company holds the University’s contract?

Jacobs: We actually have multiple options. Students can use Microsoft Copilot, Gemini is available, we still have some existing access to ChatGPT, and now we also have Claude.

We’re not vendor captured at the moment. Part of this is a question about cost. We’re trying to figure out what students want, how we avoid vendor capture because we don’t know what the future looks like, and how we make sure students’ needs are met. 

SL: What can WashU actually see when a student uses a University-provided AI account?

Jacobs: My understanding is that there is the ability to see the interactions, but the policy is generally that people are not regularly looking at those. In extreme cases, there could potentially be a review.

That is really an internal WashU IT policy discussion. It’s not something I have access to. I think it is fairly similar to the way students interact with their University email or Canvas. I don’t think there is anything about the access policy that is likely to be different for these tools than for existing tools that we provide students. 

SL: Could a student’s AI account records ever be requested as part of an academic-integrity investigation?

Jacobs: That’s a question I probably should be able to answer, but because I’m new here, I don’t know for sure.

I would imagine that, much like other records held by the University, it would be governed in the same kind of way. If it is possible to review Canvas data, email data, or other records for those purposes, then the AI interactions would be fair game as well. 

SL: What evidence would convince you that AI is actually making a WashU education better, rather than simply making certain work easier?

Butler: This is a really complicated question. I’m a cognitive psychologist, and I study how people learn. If the question is, “Is this thing good?” the answer is it depends.

There are places where interacting with AI — using it to offload parts of cognition or extend cognition — can create transformative learning experiences for students. But there are certainly other places where that may not be true. It may be the exact opposite and actually be stealing learning from students.

We need to help faculty and students understand what they are gaining and what they are losing when they use AI in these different situations. 

Sinclair: Chancellor Martin’s charge has been that students should be able to lead in an age of AI. If that is the charge, students need to understand the dangers, risks, and threats of these technologies while also having real depth in their own skills.

We’ve already seen students involved in that work. When faculty developed courses through the AI Curriculum Corps, students helped “red team” some of the assignments — essentially battle-testing them to see whether they would actually deepen the educational experience.

In a class I taught with a professor in Romance languages, students built private chatbots using their own previous writing. For each assignment, the chatbot wrote one version, and the student wrote another by hand. Students then compared them — looking at figurative language, grammatical structure, active verbs, and other elements of writing.

The goal was not simply to teach students the difference between human and AI writing. It was to help students understand their own narrative voice. 

SL: WashU gives instructors broad autonomy to decide whether and how students use AI. Where should that autonomy end? Are there any uses the University believes should always be permitted or always be prohibited?

Butler: One of the greatest challenges is that there is such a diversity of perspective among faculty and among students. We need to maintain the agency of both.

I don’t think everybody needs to use AI. But people need to be AI aware. We can’t bury our heads in the sand and pretend that it doesn’t exist.

At the same time, if people have strong religious, environmental, or other ethical concerns and don’t want to use AI in a particular way, we need to respect that. You’re going to see very different approaches across different courses. In some, AI may be infused into everything because that makes sense for the discipline. In others, faculty may believe students should not engage with AI at all because it does not add value to what the course is trying to teach.

That means students are going to encounter different policies as they move between courses. Unfortunately, I think that is part of what we are going to have to navigate. But I would like to make it easier for students so those differences do not become so confusing that students get tripped up. 

Jacobs: If I could enforce one thing, it would be that faculty are aware of these tools and understand how they work so they can think about them in the context of how they teach. AI might change the content they need to cover, or it might change the way they assess student performance. 

SL: As AI becomes capable of doing more of the work students traditionally had to do themselves, what intellectual skills do students still need to develop without AI assistance?

Butler: One useful framework is “competency-gated AI use.” If there are higher-order thinking skills people are acquiring, you probably need to be able to do those things yourself first. Once you start developing those abilities, then offloading certain parts of the work to AI can be reasonable.

Maybe that means that in introductory or first-year courses, students largely do some things without AI. As they develop habits of thinking and competencies over time, they can shift toward using AI to offload simpler parts of more complex work.

One place where I’m a little worried is foundational knowledge. I think some students look at memorizing bodies of knowledge or doing other building-block work as the obvious things to offload. I would say: Be careful there. Those are often the things that enable the higher-order thinking you want to do later. 

Jacobs: Students are ultimately responsible for the work they produce when they go out into the world.

AI creates the ability to offload some work, so students need to understand that they are still responsible for the quality and correctness of whatever they produce. They need the foundational skills to evaluate whether something is actually correct.

It’s about expertise. There really is no shortcut to developing expertise. AI can make it feel like you are an expert if you’re not careful. 

SL: How should faculty decide which parts of the learning process students can reasonably offload to AI and which kinds of “friction” are actually worth preserving?

Jacobs: That is an interesting challenge. Sometimes the thing a student thought was drudgery or something on the side of the official learning objective turns out to be a skill that becomes useful later.

The challenge for an instructor is figuring out which bits of friction were actually helpful for the student and which were simply extra work that did not do anything meaningful. That is part of why we have landed on trusting faculty in individual disciplines and programs to make some of those decisions. Which frictions really matter can be very different across disciplines. 

SL: What role do students themselves have in deciding what WashU’s approach to AI becomes?

Sinclair: We would really love for students who feel called to be involved in this process to participate. The capacity for students to co-create here is enormous, and the stakes are high for all of our constituencies to get this right.

Butler: The student voice is incredibly important. There are going to be opportunities for students to serve on advisory panels and give input on programs and initiatives. We need to be more in conversation about these things — student to student, student to faculty, and faculty to faculty.

That is how we establish norms and values, reaffirm those things, and figure out how to innovate and get better.