“AI wants to be a pleaser. It doesn't want to overwhelm you. It doesn't have any stakes in the consequences. It wants to speak confidently. And that's where hallucinations sometimes come in.”
Awkward Silences · User Experience · October 2026
UX researcher Sal Criscione on teaching himself quantitative research with AI. He still calculated the numbers by hand, and later said he fed his practice tool wrong answers on purpose to see whether it would flatter him.
Transcript
…procedure and then explain why I could select that statistical procedure. And I wanted the ability to calculate the data and then know that I have come to the correct numbers because I manually calculated the data. I didn't ask AI to calculate the data for me. Does that answer your question?
…was resonant for me, which is I calculated things by hand. And so, you know, AI wasn't there replacing the work or kind of like giving you the answer so much as it was kind of giving you a foundation to start with and then helping you to confirm that you're in the right ballpark with your answer.
Yeah, yeah, that's correct. Something I also do want to mention of when using AI, it's really important to know about AI biases. And we could have an hour-long conversation about those biases. AI wants to be a pleaser. It doesn't want to overwhelm you. It doesn't have any stakes in the consequences. It wants to speak confidently. And that's where hallucinations sometimes come in. And so it's good to use AI, but it's really important to know about these biases that you might fall into these traps where it's giving you a confident answer, but it had to hallucinate the data to give you a confident answer.
Yeah. I think that's a really important point because I mean, there's a part of me that feels like every time we ask AI to do something, we're asking it to hallucinate. But sometimes we do it with like better guardrails and, you know, kind of anticipating all those things that you just talked…
Yeah. And so when learning a new skill, you should have a source of truth, something that you could lean on. You shouldn't be leaning on AI to teach you something. And so maybe it's a book, maybe it's a course you're going through, maybe it's a mentor. You should identify what's your source of…
Yeah, I think that that is a very thought-provoking point, especially in this world where a lot of us who are teaching folks are kind of looking at AI and people saying, hey, you know, you can lean on AI to actually teach people. And I think you're, it sounds like you disagree with that. So that's…