AI Helps Researchers Learn New Fields — But Fails When It Matters Most
The harder the topic, the harder the AI is to fact-check.
Scientists today are increasingly asked to work across fields — a biologist who suddenly needs to understand machine learning, or an engineer who must grasp economics. To keep up, many are turning to AI chatbots and 'deep research' tools (AI that searches and writes reports for you) to get up to speed fast. A new study wanted to know how that actually plays out in real life, so researchers followed 15 interdisciplinary scientists over time and interviewed them in depth.
The good news: AI genuinely helps. These researchers used it to fill gaps in subjects where they weren't experts, saving them hours of reading and letting them ask basic questions without embarrassment. Importantly, they still made the creative leaps themselves — they used AI as a study buddy, not as the person who decides what's new and interesting. That matters, because the whole point of their work is discovering something nobody has found before.
The bad news is what the authors call the 'expertise paradox.' The AI's answers were hardest to check precisely when checking mattered most. If you already know a field, you can spot a wrong answer instantly. If you don't, the AI sounds just as confident when it's right and when it's flatly wrong. So the tool is weakest exactly where you're leaning on it hardest.
The study ends with suggestions for builders: AI should adjust how much it explains based on how much you already know, push you to connect ideas across fields, and write in the style your discipline expects. The bigger takeaway for anyone using AI to learn something new: assume it can be wrong, and find a human expert or a real source to confirm anything you plan to act on.
- Researchers followed 15 scientists who work across multiple fields to see how they actually use AI day to day.
- AI was great for filling knowledge gaps, but scientists still came up with the original ideas themselves.
- The 'expertise paradox': AI answers are hardest to fact-check exactly when you know the least about the topic.
Why It Matters
If you use AI to learn something new, its confident wrong answers are hardest to catch when you know least.