The worlds largest university experiment with artificial intelligence already has results — and they are complicated
Artificial intelligence has arrived at the university level — and it did not tiptoe in.
California State University, known as CSU, signed a 17 million dollar contract with OpenAI last year to bring ChatGPT Edu — a version of the generative AI chatbot designed specifically for educational institutions — to more than 470,000 students, faculty, and staff spread across 22 campuses. Recently, the university renewed that deal for an additional 13 million dollars per year for the next three years. It is the largest university AI experiment in the world — and according to Chancellor Mildred García, no other institution of higher education is doing anything similar at this scale, whether in the United States or internationally.
But what actually happened after that bet was placed? The numbers from an internal survey with more than 94,000 respondents reveal a far more complex picture than any official press release could suggest: high usage, high skepticism — and a real tension between equitable access and quality education. Students who use the tool with critical awareness share the same campus with peers who turned it into a crutch. Professors who adapted their classes coexist with researchers who signed petitions calling for the contract to be canceled. CSU has become, in practice, a living laboratory for the entire world to observe — and what it is discovering matters to anyone who cares about the future of education in an AI-driven era. 🎓
Why CSU chose OpenAI — and what was at stake
The partnership did not come out of nowhere. Internal CSU documents obtained by NPR show that, in December 2024, university leaders had already identified the OpenAI deal as a major branding opportunity for the institution. Ed Clark, the chief information officer in the chancellor’s office at CSU, explained that the university chose OpenAI because it considered the company the most cost-effective option capable of distributing AI tools to more than half a million people, including students, faculty, and staff.
The contract, it is worth noting, was awarded without a competitive bidding process — a point that drew both internal and external scrutiny. A separate document, dated 2025 and also obtained by NPR, shows that CSU already anticipated questions about that decision. The document, titled something along the lines of potential follow-up questions about the ChatGPT initiative, instructs staff to justify the no-bid contract by saying the deal was essential to the success of the university’s AI strategy, and that, after extensive research and evaluation of various tools and vendors, OpenAI was considered uniquely positioned to meet the institution’s needs.
The university made it clear from the start that AI would not be used to teach classes. Clark stated that the technology should complement learning, never replace it. Both CSU and OpenAI frame AI adoption as a necessity to prepare students for careers increasingly shaped by this technology. Leah Belsky, vice president of education at OpenAI, reinforced that vision by saying the company shares the responsibility of helping students use these tools effectively to unlock their full potential and succeed in an AI-powered future of work.
CSU, therefore, is not just offering a tool — it is making a public statement that AI literacy is part of contemporary professional training. And it is precisely that statement that divides opinions within the university community itself.
The numbers nobody expected to see
In the fall, CSU invited students, staff, and faculty from all 22 campuses to participate in a survey about their perceptions of AI. More than 94,000 people responded, and the results paint a picture full of nuance — widespread use coexisting with deep distrust.
Among the most relevant findings from the survey:
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
- More than half of students and roughly 6 out of every 10 faculty and staff members reported using AI regularly for academic or professional tasks.
- Approximately 65% of students and 59% of faculty said they were skeptical about the benefits of AI for education in general.
- 80% of students said they would not feel comfortable submitting AI-generated work as if it were their own.
- About 64% of students said AI had positively affected their learning, while roughly 35% reported negative effects.
- Approximately 56% of faculty said AI had positively impacted their teaching, research, and administrative experiences. However, in a separate question, 52% reported a negative effect — an apparent contradiction that reflects the complexity of overall sentiment.
- Around 84% of students said they use ChatGPT. Of those, only a quarter used the version provided by CSU — the vast majority relied on the free version.
- Large majorities of students and faculty expressed concern about AI’s impact on creativity (83% of students, 82% of faculty), job security (82% of students, 78% of faculty), and the environment (80% of students, 84% of faculty).
It is important to note that the survey did not directly ask whether respondents agreed with the university’s decision to spend millions on a contract with OpenAI. David Goldberg, an associate professor at San Diego State University and one of the authors of the survey, acknowledged that the results are based on the people who actually responded — and that it is impossible to know the opinions of those who did not participate. Still, he said the responses represent the diversity of fields of study and demographic profiles across the institution well.
Goldberg highlighted something that may be the most revealing finding of all: the enormous amount of nuance present in the responses. According to him, even within a single student, it is perfectly possible to use the tool frequently, recognize real advantages, and at the same time see the downsides clearly. That ambivalence is not a sign of confusion — it is a sign that people are thinking about the topic honestly.
What professors are doing differently
Across CSU’s 22 campuses, one of the most interesting transformations has not happened in tech labs but within teaching methodologies themselves. Professors who decided to actively incorporate ChatGPT into their courses — rather than merely tolerating or banning it — started redesigning their assessments in ways that made simply generating text automatically irrelevant to earning a good grade.
Zach Justus, a communication professor and director of faculty development at California State University, Chico, has spent the past few years encouraging colleagues to adapt their teaching practices to the AI era — which includes experimenting with the technology to understand what it can and cannot do. He says he is excited about the innovative ways some professors are using and allowing students to use AI. But he makes a point of emphasizing that adaptation, in certain circumstances, also means redesigning activities specifically to prevent AI use.
The most important message he shares with colleagues, in his own words, is that they cannot ignore the technology. If they do, they are not doing their job.
Justus also understands the criticism of the university’s contract with OpenAI — including the argument that the system should not be spending millions on an AI chatbot at a time when it faces budget cuts. But he raises an important counterpoint: if the university does not provide these tools to students, those with more financial resources can simply pay for premium versions while low-income students fall behind. In his words, that would systematically benefit students with more financial resources — and that is not fair.
English professor Jennifer Trainor at San Francisco State University takes a different but equally thoughtful approach. She does not ignore AI, but she is not exactly an enthusiast either. Her strategy is to teach students about the technology and the ethical questions it raises while protecting the learning process by requiring students to brainstorm and draft by hand during class time. Trainor allows them to use AI to edit their texts but requires them to critically reflect on the changes the tool makes.
She describes her goal in straightforward terms: getting students to actually write and think on their own while also giving them the chance to observe what happens when they use tools to improve their writing and reasoning. Trainor also notes that some students refuse to interact with AI entirely. She describes this as a growing wave of resistance on campus — students who oppose the environmental impacts, algorithmic bias, and the threat the technology poses to their jobs, voices, and creativity. 🧠
What students gain — and what they might lose
Sejal Daterao is one of those students with complicated feelings about AI. At 30, she enrolled in the master’s program in information systems at California State University, Long Beach, specifically to learn how to use artificial intelligence more effectively. As a student, she uses ChatGPT Edu and other AI tools to conduct research, summarize texts and video lectures, and create quizzes targeting the subjects she is studying.
Daterao says she is grateful that CSU provides access to ChatGPT Edu — which includes features not available in the free version of ChatGPT. As a graduate student, she says it would be hard to afford a premium subscription on her own. In her view, helping students use these technologies firsthand is genuinely positive.
But she does not define herself as pro-AI. Daterao gets frustrated with the false information that AI chatbots occasionally generate and with the use of creative works to train models without giving credit or compensation to artists. In her view, the technology has a lot of bad sides and a lot of good sides — and anyone who is smart and ethical can use the good sides in truly incredible ways.
Then there is the student identified only as H — who asked not to have her name revealed because she is applying for jobs in tech and does not want her opinions about AI to affect her chances. She is in her senior year studying computer science at San José State University, also part of CSU, and does not see many redeeming qualities in AI.
H says she got frustrated when she realized her classmates were using AI to write academic papers for them. That frustration led her to avoid the technology completely at first. Over time, she started using it for smaller tasks, like writing emails, and later to help with programming assignments. But she noticed something concerning: when she used AI for coding, she started relying on it as a crutch instead of actually learning. That was the signal she needed to stop.
Her resistance only grew as she learned more about the environmental impacts of the data centers required to sustain AI models. H understands that CSU is under pressure to adapt to an emerging technology, but she says she is a little disappointed that the institution embraced AI with open arms and immediately. She also worries that encouraging AI use in academic work prevents students from learning the foundational skills they need to succeed professionally. In her words, trying to use the tool to learn the basics ended up resulting in simply not learning the basics and using the technology to avoid the effort.
The tension nobody can easily resolve
At the center of this entire debate is a question that goes far beyond the walls of California State University: what does it mean to truly learn when a tool can handle a large portion of the cognitive work for you? This is not a new question — versions of it appeared when calculators arrived in classrooms, when Google made memorizing facts less essential, and when word processors automated spell-checking. But ChatGPT represents a qualitatively different leap because it does not merely support specific cognitive tasks — it can simulate the final product of many of them, from an argumentative essay to a scientific article summary, convincingly enough to fool traditional assessments.
Martha Kenney, a professor and researcher in science and technology studies at San Francisco State University, is one of the most outspoken voices in this debate. She argues that refusing this technology needs to be a valid position at the discussion table. For Kenney, the rejection is justified because of the environmental impact of generative AI, the use of copyrighted works to train models, and the fundamental doubt about the educational value of a chatbot that enables shortcuts on academic work. In her view, offering this tool is, in practice, cheating students out of their education.
Kenney co-authored a petition asking CSU not to renew the contract with OpenAI. But Ed Clark, the university spokesperson, pushed back, saying the online petition does not reflect the general sentiment of the university community. He pointed out that the university’s generative AI advisory committee — composed of students, faculty, and staff — unanimously recommended renewing the contract.
The tension between equity and quality is perhaps the hardest to resolve. If CSU restricts access to ChatGPT or broadly penalizes its use, it may be taking away from low-income students a tool that, when used well, reduces historic inequalities in access to educational resources. If it maintains unrestricted access without a robust pedagogical strategy, it risks graduating students who have mastered prompt engineering but struggle to think independently and in a structured way. There is no easy answer — and any institution that claims to have solved this dilemma is probably oversimplifying the problem.
The researchers who signed petitions raise points that deserve to be taken seriously: questions about data privacy for 470,000 people, about the environmental impacts of training and using large-scale language models, and about institutional dependence on a private company for core functions of the educational process. These arguments cannot be dismissed simply because the product is popular or because the partnership generated impressive adoption metrics. 📊
The bigger picture: the university AI race
CSU is not alone in this move. Prestigious universities across the United States — from Syracuse University to Dartmouth College to the University of Minnesota — have also signed similar contracts with AI companies like Anthropic, OpenAI, and Google. But as the largest public four-year university system in the United States, with approximately 470,000 students and responsible for nearly half of all bachelor’s degrees awarded in California, CSU’s partnership has a scale and visibility that set it apart.
CSU’s student body is extremely diverse: about half of students are Hispanic, more than a quarter of undergraduates are first-generation college students, and many work while attending school. This profile makes the experiment even more significant because the impacts — positive or negative — fall on a population that historically has less room for institutional missteps.
For this audience, the difference between having or not having access to a premium version of ChatGPT is not trivial. First-generation students and working students who attend college simultaneously often lack support networks, private tutors, or time to attend professors’ office hours. A tool that functions as an accessible 24-hour tutor can be transformative for this group — as long as it comes with critical training to avoid the risks of dependency and superficiality that other students have already reported.
What the rest of the world can learn from this
The CSU experiment matters far beyond the United States because it is, so far, the best-documented and largest-scale example of institutional artificial intelligence use in higher education. Universities in Brazil, Europe, Asia, and around the globe are watching the results closely, trying to understand which lessons are transferable to their own contexts and which are too specific to the American environment.
In Brazil, for example, where inequalities in access to technology are even more pronounced and where the debate about AI in public schools and universities is just getting started, CSU’s data offers both inspiration and a cautionary tale — depending on which part of the survey you choose to emphasize.
The biggest lesson emerging from this living laboratory is not technical but pedagogical: the technology itself does not determine the outcome. What determines the outcome is the quality of the educational strategy surrounding the use of the technology — teacher training, assessment redesign, the creation of clear and honest policies about what is allowed and why, and the institutional commitment to monitoring real effects rather than just adoption metrics. ChatGPT is not inherently good or bad for education — it is a mirror that amplifies what already exists in an institution’s teaching and learning practices. Where those practices were solid, the tool boosted results. Where they were fragile, it exposed the weaknesses.
And perhaps that is exactly what makes the CSU case so valuable for the global debate about artificial intelligence and education: it shows, with real data and no marketing filter, that adopting AI at scale in a university is not a solution — it is the beginning of a series of much harder questions. Questions about what we want our students to be able to do when they leave the university, about which human skills need to be protected and cultivated even in a world where machines perform an increasing number of cognitive tasks, and about how to build educational systems that prepare people not just to use the tools of the present but to think critically about the tools of the future. 🚀
