AI Is Not Our Destiny. You Are.

When the graduating Class of 2026 arrived on campus in August 2022, the most advanced publicly available artificial intelligence language model was OpenAI’s GPT-3. Grads, you probably had no idea it existed and didn’t use any AI — recreationally or educationally — your first semester in college.

That November, OpenAI upgraded the model and added a friendlier chat interface, renaming it ChatGPT. Chances are you were among the estimated 100 million people who tried it by the start of your second semester in January 2023.

Now that you’re graduating, AI models are capable of completing almost any assignment you’ve been given — from writing poetry to reproducing complex mathematical proofs.

Some of you used them as a substitute for thinking, cheating yourselves. I know firsthand that the Economics 10: “Principles of Economics” teaching team got problem set submissions with concepts more advanced than what we had taught — concepts a large language model would have recommended. And there was likely much more academic dishonesty we did not know about.

Still, some of you used AI to genuinely understand new concepts. One student told me she was only able to succeed in a graduate-level economics seminar because AI helped her break through the barrier of advanced math to understand deeper ideas.

And some of you even wrote astounding theses that employed AI to make discoveries impossible even a year or two earlier — like tracking illegal fishing vessels, or studying the colonial origins of unequal economic mobility in New Zealand.

In education, as in work and life, AI multiplies your options. Some of those options are great. Some are terrible. The challenge for your graduating class is finding more of the great and fewer of the terrible ones.

It would be convenient if the technology companies could solve this problem for us — if they could build models that refused to enable cheating, that wouldn’t spoon-feed you the illusion of knowledge but instead helped you struggle through the difficult and necessary process of actually acquiring it. That is not going to happen. There are too many companies, too many models, and too many different use cases for any single guardrail to hold.

Harvard has a significant role to play in our evolving relationship with this new technology. While you were here, the pace of the University’s adaptation fell well behind the pace of new and improved AI models. Some faculty developed creative assignments and teaching methods that incorporated AI thoughtfully. Others responded defensively — reducing the weight of take-home work or effectively eliminating it. Still others declined to notice the sea change at all, allowing students to short-circuit their own learning. I myself am painfully aware that I have been behind the curve.

I fear that during your college experience, on balance, the bad options created by AI outweighed the good ones. I’m cautiously optimistic that Harvard, and higher education more broadly, will eventually shift that ratio — perhaps through more in-class time, AI-free monitored spaces for independent work, and structures that help students resist short-term temptation and do the hard work of learning. But we’re not there yet.

The AI companies cannot prevent these downsides. Institutions like Harvard are slow and imperfect. That leaves the final line of defense: you.

The most important thing you can do is be honest with yourself. You know the difference between using AI as a lever — to think more rigorously, to test your ideas, to go further than you could alone — and using it as a shortcut that quietly hollows out your own capabilities. The first builds something. The second borrows against a future you may not be able to cash.

I have spent more than a decade studying the future of work. The more effort I’ve put into understanding these nuances, the more convinced I am of the answer: No one knows. That is why David Laibson and I end Economics 10 with a discussion of the many possible futures ahead — we do not want to give you false confidence that the field of economics, or any of us academics, truly knows the answer.

You are about to enter a world where AI will be woven into almost every part of life. Historically, one’s first years after college meant assisting others to build your own capacity. Now, you will have access to incredibly capable, tireless digital assistants working for you from day one. Learning to use them in ways that complement and strengthen your own judgment — rather than replace it — may be the most valuable skill you can develop in the years ahead.

True learning is about struggling, making mistakes, and eventually figuring out how to correct yourself. If everything is fed to you, that critical feedback loop simply doesn’t exist. Perhaps our slowness in adapting to AI while you were in college will actually work to your benefit: you had to struggle to figure it out for yourselves, and eventually you will be in a position to help reshape it for the rest of us. Hopefully, your class will help us all harness the promise of the future while minimizing the downsides.

The diploma you are receiving today is a beginning. As you already know, there are numerous challenges ahead. AI companies, government regulations, and your future employers will not solve the problems AI creates. Only you can. Good luck.

Jason Furman ’92 is Aetna Professor of the Practice of Economic Policy at the Harvard Kennedy School and the Department of Economics.