Most people stop at “Mount Stupid” with AI. The real risk is that they never go past it. There is a podcast doing the rounds arguing that AI makes us dumber: it is authoritative, it has near-infinite knowledge, it intimidates the human brain into outsourcing thinking, and in ten years we are all non-thinking zombies. I disagree, and I think the reason is the shape of the learning curve itself.
The first peak is a trap
When you first interact with AI, you genuinely feel like an expert at everything. The model is authoritative, the answers sound right, and you stop questioning what it tells you. That is Mount Stupid: the early, over-confident peak of the Dunning-Kruger curve. It is where the “AI makes us dumber” story is set, and for people who never move past it, the story is half right.
What using it heavily actually teaches you
But here is what the doom take misses. The more you use AI, and I use it extensively, the more you rediscover the power of the human brain. It is far more than reasoning, knowledge retrieval and pattern matching, the three things every IQ test measures and the three things AI genuinely beats us at. What AI cannot yet replace, with anything like the same precision, is discernment. Judgment. The ability to know when something is technically right but feels wrong because it touches your values, your experience, the texture of a lived journey, the instinct that lets you smell a flawed argument three sentences in, or spot an inauthentic pitch before it finishes.
Partner, not oracle
The trick is to engage and converse with AI rather than defer to it. Used that way, it sharpens your mind and compounds your learning; I have found a reasoning capacity I never knew I had. Used the other way, as an oracle you stop questioning, it does exactly what the podcast fears. Great powers carry great risks, and the biggest one here is simply that some people never climb past the first peak. It is the same reason we build AI to support human judgment, not replace it, and why the human belongs in the loop only where the loop is expensive to get wrong.
How are you engaging with AI: as a partner, or as an oracle?