In July 2023, I wrote about the future of artificial intelligence (AI), sketching three possible trajectories: a university without walls, a prison without cells, and extinction. The tone was speculative, and the perspective was that of an observer triangulating among science fiction, software architecture, and the philosophical problem of human bias embedded in every system of knowledge. I ended with an admission of uncertainty: meaningful dialogue between AI developers and AI users would be essential, but I had little idea how that dialogue might actually develop.

Three years later, I am no longer writing from the outside.

Since that article, I have collaborated openly with Claude, Gemini, and ChatGPT in developing essays and have used AI as a developmental partner while writing a novel. I have watched these systems identify weaknesses I overlooked, challenge assumptions I had not questioned, and occasionally suggest perspectives I would not have reached on my own. I have also encountered their limits: confident hallucinations, unwarranted certainty, and an enduring tendency to smooth over contradictions rather than acknowledging them. And, somewhat amusingly, I have watched them criticize each other: where one might hedge, another argues that the hedge weakens the argument.

Those experiences have led me to a different question. The most important issue is no longer what AI will do to universities. It is what universities themselves are doing with AI—and whether they are asking the right questions.

The Fault Line Universities Rarely Name

Most contemporary discussions of AI in higher education emphasize governance. UNESCO guidance, EDUCAUSE Horizon Reports, institutional policies, and faculty task forces focus on assessment redesign, privacy, academic integrity, and responsible implementation. These are legitimate concerns.

Ellucian’s third annual report on the use of AI in higher education found that personal AI use is near saturation, while institutional adoption is expanding and moving toward strategic integration. A large number of respondents expect a rise in institutional use in the coming years.

Yet these concerns are largely downstream from a deeper issue. The real divide is epistemological.

Colleges and universities increasingly approach AI from two dominant orientations. One treats AI primarily as a means of delegation (namely, outsourcing defined tasks to improve efficiency). The other views AI as a means of extension (namely, expanding human inquiry into intellectual territory that would otherwise remain inaccessible).

These orientations are not mutually exclusive. Admissions offices may automate routine processing while historians use AI to explore competing interpretations of historical events. Nevertheless, the distinction helps explain why participants in discussions about AI often seem to talk past one another. Institutions may appear to be debating technology while actually disagreeing about the nature of knowledge itself.

The difference matters because delegation and extension place fundamentally different demands on probabilistic systems.

Corporate governance, risk management, and compliance functions generally exhibit a low tolerance for irreproducible reasoning. Audit trails must be explainable. Regulatory decisions must withstand external scrutiny. AI’s probabilistic architecture therefore creates understandable hesitation. According to the 2026 Onspring Benchmarking Report, only 13.5 percent of governance, risk, and compliance organizations report having fully integrated AI into their core workflows. The issue is not simply technological immaturity but the mismatch between probabilistic inference and institutional demands for consistent justification.

Universities operate under different incentives.

Their primary mission is not procedural assurance but knowledge creation. Scholarship advances by confronting uncertainty rather than eliminating it. Yet many institutional discussions about AI continue to mirror corporate governance models, emphasizing compliance, detection software, and administrative control rather than intellectual opportunity.

The more interesting question is therefore not simply whether universities should adopt AI, but what conception of intelligence they hope AI will help cultivate.

What Academic Disciplines Reveal

Different disciplines already provide partial answers to the question of what kind of intelligence universities hope AI will help cultivate.

The hard sciences demonstrate that probabilistic systems can become indispensable when paired with rigorous verification. AlphaFold, introduced by Google DeepMind in 2020, transformed structural biology by predicting protein structures with unprecedented accuracy. More than three million researchers across over 190 countries have since used the AlphaFold Protein Structure Database, accelerating research on disease and drug discovery.

Protein science, of course, was not “solved.” Protein dynamics, folding pathways, and molecular interactions remain active areas of investigation. What changed was that researchers suddenly possessed an extraordinarily powerful exploratory instrument capable of mapping structural possibilities that no human intuition alone could efficiently discover. Verification remained a human responsibility; exploration became dramatically more powerful.

That partnership mirrors scientific inquiry itself. Science has never eliminated uncertainty. It has developed increasingly sophisticated methods for navigating it.

The humanities exhibit a different but equally revealing pattern.

Recent research in digital humanities describes generative AI less as an automated tool than as a dialogic collaborator: a conversational partner capable of generating alternative interpretations, identifying overlooked themes, and provoking new questions. Programs such as Schmidt Sciences’ Humanities and AI initiative reflect growing institutional confidence that AI can expand rather than diminish humanistic scholarship.

Here, however, the value of AI depends upon recognizing its outputs as provisional rather than authoritative. An invented quotation remains an error. A fabricated historical citation remains unacceptable. But an unexpected analogy, an unconventional metaphor, or a surprising interpretive framework can become intellectually productive precisely because it stimulates deeper human analysis.

The model succeeds not by replacing judgment but by provoking it.

The social sciences occupy a more complex middle ground.

Quantitative disciplines often emphasize reproducibility and methodological transparency, making researchers understandably cautious about AI-generated analyses. Recent studies suggest that AI-assisted teams perform comparably to traditional research teams rather than dramatically surpassing them, while AI-led teams still require human oversight.

Qualitative researchers have embraced different possibilities. Interviews, narrative analysis, thematic coding, and ethnographic interpretation increasingly incorporate AI as a conversational research partner that can reveal alternative patterns without replacing the investigator’s interpretive responsibility.

As a cultural anthropologist, I recognize another dimension of this evolution: computational social science, network analysis, and mixed-methods research increasingly blur the distinction between quantitative and qualitative inquiry. AI fits naturally within this broader movement because it extends our capacity to explore relationships that would otherwise remain hidden while leaving interpretation firmly in human hands.

Across these diverse disciplines, the essential distinction is remarkably consistent: AI is welcomed when it expands intellectual exploration and resisted when it appears to replace intellectual responsibility.

The Larger Question

There remains, however, a question that disciplinary frameworks and governance documents rarely confront directly. We still do not fully understand what these systems are doing.

This should not be seen merely as a failure of AI companies. It reflects the present state of interpretability research itself. Researchers continue making important progress in understanding how large language models represent concepts internally, how competing representations interact, and why confident errors emerge. Yet, today’s most capable AI systems remain only partially transparent even to those who designed them.

Nor is human cognition fully transparent. Neuroscience, psychology, and philosophy continue debating how minds produce reasoning, consciousness, and creativity. Universities have devoted centuries to studying natural intelligence while only beginning to investigate AI with comparable seriousness.

That is precisely why universities occupy a unique position.

The question “What is AI?” cannot be answered solely by computer science. It is simultaneously a philosophical question about reasoning, a linguistic question about meaning, a psychological question about cognition, an ethical question about responsibility, and an anthropological question about culture. Few institutions other than universities possess the disciplinary diversity required to pursue all of these questions simultaneously.

If universities limit themselves to asking, “How do we integrate AI responsibly?” they ask an important administrative question.

But if universities ask, “What kind of intelligence have we created, and how should we understand it?” they fulfill their deeper intellectual mission.

Universities Have Faced This Moment Before

History offers useful perspective on AI’s place in the university.

Calculus, statistical inference, digital computing, molecular biology, and geographic information systems each entered universities first as specialized tools. Over time, each became something far more significant: a new way of asking questions and organizing knowledge.

AI may represent a similar transition.

Whether it ultimately proves as transformative remains uncertain. But colleges and universities should recognize that the issue is no longer simply technological adoption; it concerns the evolution of scholarship itself.

A Final Note on Method

This essay was developed through dialogue with Claude, Gemini, and ChatGPT.

That statement is not a disclaimer, but evidence of collaboration.

The argument presented here does not claim that AI should replace human authorship. Rather, it suggests that disciplined dialogue with AI can extend human reasoning into intellectual territory that would otherwise remain unexplored. Whether this essay succeeds in demonstrating that proposition is for readers to decide.

The larger question belongs to the university itself.

For centuries, universities have debated the nature of intelligence, creativity, and knowledge. AI does not eliminate those debates, but enlarges them.

The challenge is not whether AI can think. The challenge is whether universities can learn to think differently with AI while preserving the skepticism, intellectual humility, and critical inquiry that have always defined genuine scholarship.