AI's Biggest Unlock Isn't Productivity. It's Access to Expertise
This article argues that AI's true potential lies in democratizing access to expertise, not just boosting productivity. Studies show AI can narrow educational gaps, but only when designed as a tutor rather than an answer machine.
Divi Vijayakumar
Jul 12, 2026
Talent has never belonged only to elite institutions. Opportunity often has.
For most of history, knowledge has lived behind a gate. The best schools. The best professors. The best mentors. The right network. The right geography.
Even the ordinary things were easier if you knew who to ask. Getting your résumé reviewed. Preparing for an interview. Pressure-testing a business idea before you bet a year of your life on it.
I kept coming back to one question. What would the world look like if everyone had a Stanford professor, or a seasoned mentor in their field, one message away?
The internet was supposed to answer that, and it didn’t, quite. The internet democratized information, and information is not mentorship. A search engine doesn’t notice when you’re confused. It won’t explain an idea five different ways, challenge a bad assumption, rehearse with you, or adjust to how you learn.
AI turns information into interaction. That, to me, is the real unlock.
Not writing emails faster. Not generating more code. Not squeezing another hour out of the workday.
It’s the ability to ask unlimited questions without embarrassment. To learn at your own pace. To rehearse before the stakes are real. To go from “I have an idea” to “I can test it” without waiting for permission.
The early evidence is encouraging
A 2026 randomized experiment with 1,174 adults found that generative AI helped participants at every education level, but helped those with less formal education substantially more. Without AI, the higher-education group outperformed the lower-education group by 0.548 standard deviations. With it, the gap fell to 0.139. That’s roughly three quarters of the gap, gone. (NBER Working Paper 34851)
In Edo State, Nigeria, secondary school students joined a six-week after-school program that paired AI tutoring with teacher guidance. The World Bank measured learning gains of about 0.3 standard deviations, which in that setting is worth something like one and a half to two years of ordinary schooling. Girls, who started behind, gained the most. (World Bank)
That is the future I find exciting. Not AI making already-capable people marginally faster, but AI helping more people become capable in the first place.
But access to answers is not access to learning
A Wharton-led study of nearly a thousand high school math students found that unrestricted GPT-4 access improved performance while students were using it. Take the tool away, and those students did 17% worse than peers who never had it at all. (Bastani et al., SSRN)
The same study ran a second version of the tool, one built to give hints instead of answers, with teacher input baked in. Most of the damage disappeared.
Same underlying model. Different product. Opposite outcome.
That distinction is the whole thing.
The opportunity is not to hand everyone an answer machine. It’s to give everyone a patient teacher, a rigorous sparring partner, and a door into rooms they couldn’t previously enter.
AI shouldn’t remove the struggle that learning requires. It should make that struggle more accessible, more personal, and more productive.
Who gets to begin
We keep asking how AI will change today’s jobs. I’m more interested in who it will give a chance to start.
AI’s greatest legacy may not be that it made the already-capable more productive. It may be that it gave millions of people the chance to find out what they were capable of.
I cannot wait to see the next important company, scientific breakthrough, or cure for an incurable disease emerge from the most unexpected person, place, or journey.
References
Cruces, Fernández Meijide, Galiani, Gálvez & Lombardi. Does Generative AI Narrow Education-Based Productivity Gaps? Evidence from a Randomized Experiment. NBER Working Paper 34851, 2026.
From Chalkboards to Chatbots: Transforming Learning in Nigeria, One Prompt at a Time. World Bank. Full working paper: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria.
People-Centered AI in Education: Five Lessons from the Global South. World Bank.
Bastani, Bastani, Sungu, Ge, Kabakcı & Mariman. Generative AI Can Harm Learning. Wharton School Research Paper, 2024.