Open questions, conflicting findings, and areas where the research is unsettled.
Summary
There is a lot about AI that we don’t yet know. We don’t know how effective they are at actually teaching, we don’t know how safe they are, and we don’t know how they’ll impact the job market. We also don’t know about the differentiation in access in the U.S. – who has it and who doesn’t?
But that uncertainty doesn’t stop people – often very smart, well-informed people – from pretending that they know these answers.
Last updated December 19, 2025
Key Information
- We don’t know how AIs stack up against traditional teaching techniques – or how that changes over time.
- Worldwide, countries with stronger economies (unsurprisingly) have stronger access to AI. But we don’t know about differential access among socioeconomic groups within the U.S. or how that potential divide impacts learning.
- There is a lot we don’t know about student usage vs adult usage. For example, many adults use AI for healthcare advice. Do students do the same? Is the information they receive accurate?
- We don’t know what the long-term impacts of AI on jobs will be yet. AI has led to unemployment – but only for the fairly new employees.
- As yet, there is no bullet-proof method of ensuring AIs are perfectly safe. For example, you can “jailbreak” many of their protections by putting your prompts into poetry verses.
Research Articles
- AI tutors beat live class activities – but only if you’re dealing with students at Harvard
- Developed countries have stronger access to AI tools
- 40+ million American use ChatGPT to answer healthcare questions
- How AI is affecting jobs right now (August 2025)
- Poetry can bypass the safety guardrails of Large Language Models