Why Experts Still Underestimate AI, Despite Amazing Results
The flaws are real, but do they blind us to how powerful AI has become?
Imagine a writing professor whose best student still has obvious flaws: too many adverbs one day, none the next, scenes starting too early. Is that student any good? Of course. But if the teacher only lists flaws, no one would guess they're excellent. That's how we often treat advanced AI, according to an experienced software engineer who uses AI to write code daily.
Modern large language models—the technology behind tools like ChatGPT—can seem brilliant one minute and clueless the next. They may fail a simple request and then handle an impressively hard task. Many experts point to those failures to argue AI is overrated. But the engineer argues this is a bias: judging an entire system by its most humiliating mistakes. Human beings are inconsistent too. You wouldn't call someone dumb because they miss one question on a test they mostly aced.
This matters because how we judge AI affects whether we use it wisely. If managers, doctors, teachers, and policymakers write off AI because of occasional blips, they miss practical benefits in speed, creativity, and problem-solving. At the same time, if we pretend AI never fails, we set ourselves up for trouble. The real lesson is not that AI is all-powerful or useless. It's that we should look at the whole track record, not just the worst day.
- Even the best AI makes mistakes—it's still capable overall.
- People often judge AI by a single failure, like judging a student by one weak essay.
- A fairer approach: measure AI's performance across many tasks, not just its low points.
Why It Matters
How managers judge AI's occasional failures will decide whether they harness its real economic power or ignore it.