Some thoughts on AI and education
For a list of current AI tools for students, see AI tools for students.
The goal of education
The goal of your education is to learn skills that help you contribute to society and prosper personally, and to signal these skills through credentials.
As AI capabilities are advancing rapidly and highly unevenly, there is substantial uncertainty about future AI capabilities and therefore about the skills you should focus on. Current AI systems excel at many skills commonly taught in economics or business studies (e.g., mathematics and programming). Nevertheless, AI systems still fall short in unstructured tasks that do not follow directly from the verifiable tasks they were trained on. For example, AI systems are not (yet) good at teaching, consulting, and choosing interesting topics.
Some skills at which AI systems are superhuman (e.g., mathematics) may still be useful to learn, to sharpen your thinking and to use AI insights. This is much like calculation, which is still a useful skill despite the existence of calculators.
State of AI (as of October 2026)
In economics and statistics, AI is very useful.
In theoretical work:
- literature search
- brainstorming
- proofs
- finding counterexamples
- checks
- simulations
In empirical work:
- finding data
- analyzing data
- finding bugs
AI systems still fall short in:
- comparing the merits of different ideas
- judging trade-offs between design choices for modeling
- finding interesting things to work on
- correcting/rewriting large projects (at least compared to one-shotting them)
- good exposition
- generating test questions and explaining insights
The future of AI
Future capabilities of AI are hard to predict. Recent progress has relied on improvements in architecture, increasing compute, human-generated inputs, and automated learning on verifiable tasks. Due to heavily increasing investment, these inputs are bound to grow substantially in the coming years, and capabilities with them.
Experts consider two boundary cases possible:
In the bullish case, AI capabilities increase to the point where human input no longer improves AI output. In this case, the returns to education would be mostly personal.
In the other extreme, improvements are incremental, and AI does not extend its capabilities significantly beyond the targeted abilities. Even in this case, large parts of the current economy would be automated, and AI would significantly change and support research. But human guidance and checks would still improve AI output. Thus, economic education would be valuable, but only insofar as it is complementary to AI.
Things you will no longer do in five years
- coding (but potentially still design choices)
- mathematical proofs (but potentially still applying mathematical results)
- reading original texts and summarizing them for your particular interest
But, as I said above, coding, mathematics, and summarizing may still be skills worth training to improve your performance on other tasks further up the ladder.
AI complements
So which skills do I think will be useful in the years to come?
- understanding mathematical results and translating them to reality
- interpreting empirical results and translating them into decisions
- explaining results to humans
- taste for what output is interesting
- judgment on the feasibility of projects
- organizing
- originality (even if pure AI is better, diversification through human input may still yield benefits)
Studying with AI
AI is useful for learning, and you should learn to use it productively.
Let’s distinguish two extremes in working with AI:
- AI-centered: AI produces the finished output. You try to understand it and provide feedback for iteration.
- Human-centered: You manage every step yourself and use AI for feedback.
In most cases, some point between these two extremes is optimal, where you let the AI do some steps (which you may or may not verify). Arguably, it makes a difference whether the goal is good output or effective learning. Clearly, for an exercise in a lecture, AI-centered solutions will be good but leave you with little of the learning benefit. The educational literature emphasizes the benefit of struggle. So try to fail first before you use AI.
In my experience, the human-centered approach is better and more convenient for most projects, especially if you also want to learn. This strategy broadly includes:
- write down your own thoughts/ideas/plans
- use AI for
- knowledge questions
- explanations
- feedback
- brainstorming
- copy-editing (potentially writing)
- testing you (not always good)
- create space for thinking
- only using AI chats is not ideal
- this can be anything that works for you. Try out:
- writing
- a notebook (or a whiteboard with other students)
- oral discussions
- asking/explaining to other students
- walks