As a GenAI user, I have access to the world's knowledge in whatever form I need it. I use this every day, and it significantly multiplies my ability to solve complex problems. I have paid subscriptions with Claude (my main tool), OpenAI, Perplexity, Gemini, and Mistral and I use all of them depending on the situation.
I do much less routine work now. Instead, I spend most of my time on complex reasoning and decisions. That is interesting, but also exhausting.
But I notice something unsettling: I find it increasingly difficult to distinguish between questions where GenAI errors are acceptable and decisions where they are not.
I am not alone. Research shows that people follow AI advice even when it contradicts their own judgment, and even when the AI is known to make mistakes (Klingbeil et al., 2024).
This is a problem on two levels.
First: We already let GenAI run parts of our world. Looking at how some parts of it are currently run, that might occasionally seem appealing. But we should not delegate critical decisions to an entity that cannot truly distinguish right from wrong, and sometimes simply makes things up.
Second: We are forgetting how to make complex decisions. And younger people may never learn. Gerlich (2025) found a significant negative correlation between GenAI tool usage and critical thinking scores, with people aged 17–25 most affected. Stadler et al. (2024) confirm the mechanism: LLMs reduce mental effort, but at the cost of depth of thinking.
So we are not only outsourcing critical thinking to an unreliable entity; we are also systematically eroding our own capacity to think critically.
That is dangerous.
Think of GenAI as a visitor from another planet. Completely different values, but incredibly charming and helpful. Would you delegate decisions that affect your life, your family, or your career to this visitor?
Probably not. Hopefully not.
So even when it is hard: keep making complex decisions yourself. GenAI can support you. It should never replace you.
One ask: think before you delegate.
(References to the papers in the comments).