Stanford Education Experts Put AI Into Perspective

Stanford University has formally pivoted from defensive AI policies to proactive integration strategies across its schools. The approach treats artificial intelligence as an unavoidable reality rather than a temporary disruption educators can simply block at the classroom door.

The shift is documented in a Stanford Report article that captures how faculty across disciplines are recalibrating their teaching methods. Rather than attempting to forbid AI until graduation, instructors are now asking what form of instruction creates solid understanding while taking advantage of these tools.

Mehran Sahami, the Tencent Chair of the Computer Science Department and the James and Ellenor Chesebrough Professor in the School of Engineering, leads this charge in Computer Science 106A. The course's final assignment, "Infinite Story," asks students to code an adventure game from scratch and then use ChatGPT to keep it going indefinitely. The physical experience of this assignment matters—students type code, watch it compile, then paste prompts into an interface and wait for the model to generate the next narrative branch.

The tools do add productivity, industries are expecting students to use them, and students are going to use them, even if they are banned. So the question becomes how to structure learning around that reality rather than against it.

Karin Forssell, director of Stanford's AI Tinkery and GSE Makery, describes the current moment as a gold rush where people flock to AI hoping to find a vein that will change everything for the better. That rush has led to widespread promotion of immature products. The concern is genuine—many AI tools being offered now are untested or unworthy of classroom deployment.

Stanford's response includes concrete funding. A new initiative from AI Meets Education at Stanford (AIMES) and the Stanford Accelerator for Learning will offer $1 million in seed grants to faculty, students, and staff willing to rethink how artificial intelligence fits into college teaching. The grants fund course development, research, and scholarly works on critical issues in AI and education. Proposals are due May 15, 2026.

The grants don't require any AI expertise, which signals an institutional commitment to cross-disciplinary experimentation. This is not a computer science department project alone.

Victor Lee, associate professor in the Graduate School of Education and faculty lead at the Stanford Accelerator for Learning's initiative on AI and Education, emphasizes creative learning and learning through creation. The goal is to put forward tools and research where AI becomes part of new expressive capabilities for students, rather than information dumpers into people's heads.

These tools include resources like the Classroom-Ready Resources About AI For Teaching (CRAFT) program, a co-design initiative from the Stanford Graduate School of Education and the Stanford Institute for Human-Centered AI.

The distinction between performance and learning remains critical. Sahami notes that by the time students get to their senior project, there's an expectation they're using some AI tools to help produce that larger piece of work. That said, they have to have some fundamental knowledge to be able to assure that the code is doing the right thing.

This creates a tangible friction point in the classroom. Students must first understand the fundamentals before applying AI to their projects. The physical reality of debugging code without AI assistance—reading error messages, tracing logic through a text editor, understanding why a function fails—remains essential before students can responsibly delegate tasks to automated systems.

Lessons about AI given during CS 106A are not limited to using the tools, but also include discussions of ethics, such as bias, fairness, and how the choice of data used to train AI models impacts the results they produce.

Forssell's optimism is tempered by a pragmatic question: who gets access to those opportunities? The concern about equity runs through the entire discussion. As Sahami puts it, the pessimist in him wonders about access distribution.

Some specific applications show promise. Sahami likes the idea of Khan Academy's Great Gatsby chatbot, which allows students to chat with the book's protagonist, Jay Gatsby. In a completely different application, Stanford Law School's Legal Innovation through Frontier Technology Lab is testing how AI-driven simulations of real-world scenarios can provide law students with another option for mastering negotiation and other legal practice skills.

The 2026 AI Index Report from Stanford HAI provides broader context for these educational experiments. The report documents that 4 in 5 university students now use generative AI, while only half of middle and high schools have AI policies in place, and just 6% of teachers say those policies are clear.

Over 80% of U.S. high school and college students now use AI for school-related tasks. Formal education is lagging behind AI, but people are learning AI skills at every stage of life.

The data reveals a jagged frontier of capability. AI models can win a gold medal at the International Mathematical Olympiad but cannot reliably tell time—Gemini Deep Think earned a gold medal at IMO, yet the top model reads analog clocks correctly just 50.1% of the time. This inconsistency matters for educators trying to determine when AI assistance is appropriate and when it obscures fundamental understanding.

Without a ban on AI, educators must differentiate between what students should learn without AI assistance and what defines professional AI skill development in a world where most professionals are also unsure how they're supposed to apply AI on the job.

The $1 million in seed grants represents a modest but meaningful investment in figuring this out. The proposals due May 15, 2026, will fund experiments that might fail, might succeed, and will hopefully generate evidence that other institutions can use.

What we're going to be grappling with in the near future is the question: what do we want people to be learning to do? This is not a technical question. It's a philosophical one about the purpose of education in an age where information retrieval and content generation are increasingly automated.

Many developers and researchers in the education space are focused on how AI can speed up—or even replace—existing teaching or learning tasks. The fear is that this could lead to AI replacing valuable learning and teaching.

The Stanford approach treats AI as a tool that requires change rather than a replacement for human instruction. The emphasis on AI literacy and open, frank discussion about the possibilities of AI in education mirrors how any significant advances in technology should be handled.

As we figure out how to use these tools better, there are real opportunities to incorporate them into educational experiences that allow students to engage more deeply with material or receive more personalized instruction or tutoring. The question is whether institutions can move fast enough to keep up with the technology while maintaining educational quality.

Whether universities actually deliver on these promises—or whether the gold rush mentality consumes the evidence-based approach—remains to be seen. The $1 million in grants is a start, but it's a drop in the bucket compared to the scale of the challenge.

Time will tell if this works, but at least someone is trying to figure it out systematically rather than just hoping the market sorts it out. (Frankly, that's more than most institutions are doing right now.)

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn