TJ Thorsten Jonas On UX Research Geeks

“What AI is, in the end, is a brute-force calculating probability machine. It calculates what's the most probable answer token by token, so the most obvious answers get reproduced again and again. The strong signals in the whole data set will be repeated all the time. The weak signals will vanish over time. So everything will more and more look the same.”

UX Research Geeks · User Experience · November 2026

“What AI is, in the end, is a brute-force calculating probability machine. It calculates what's the most probable answer token by token, so the most obvious answers get reproduced again and again. The strong signals in the whole data set will be repeated all the time. The weak signals will vanish over time. So everything will more and more look the same.” — Thorsten Jonas, UX Research Geeks

Context

Thorsten is explaining why AI outputs converge, and ties it to brands: using a bit less AI could make a brand more distinctive.

Transcript

UX Research Geeks Around 18:43 into the episode
Thorsten Jonas

…it more responsibly, meaning when do I really need a frontier, huge model and when can I use smaller models and when should I maybe not use it at all? Because otherwise we are running into a huge crisis. That's what things actually show at the moment. Sorry to be so depressing. I didn't want to.

Petra Rajkov

…trying to hide that they are using AI for creating certain output or using it for, I don't know, writing and all these activities. So when you talk about de-skilling, what skills are we're losing right now when the AI does the work for us. Have you seen anything like very specific happening?

Thorsten Jonas

The most important thing, and that's down to the core, what especially we as UX people probably do is what's the problem is one of the most important questions that we usually should ask, right? And the question that AI usually cannot answer. That's something that we must not lose, I think, because before I can even start using AI, I need to understand what is the problem, what is the problem of the stakeholders, what's the desires of the users, what's the needs of the users, etc. So drifting a little bit away from your question, I would like to look on what's the skills that we need to keep and that also help us keep our jobs, even with the most powerful tools. It cannot start from scratch. Someone needs to tell the machine what it should do actually. And this thing is much bigger than we think because it's not only one problem, it's understanding. If you are in a bigger organization, it's bringing people from different professions together, from different silos in the organization together. That's the biggest struggle. AI cannot do that and will not do that. We should be very careful not to shift over to the AI. I recently held a…

Petra Rajkov

Yeah, I was part of a conversation where one product team was saying that they cut out the brainstorming at the beginning of the product development cycle, the ideation, and they're like, we ideate with AI now. And to me, it's crazy because the power that I feel when I'm in the room with the people…

Thorsten Jonas

there is a nice quote, and that's already two or three years old by, I think it's John Herman, and he said that in Europe magazine. Don't know. He said something like, ChatGPT is now better than ever in pretending it's something that it's not. And that's what AI is very good at, right? It pretends…

Petra Rajkov

Okay, so let's summarize a little bit what we were discussing or what is the takeaway that we can share with the listeners. What can a researcher or a designer do tomorrow when someone says let's add AI to the process, to the product? You already mentioned The canvas, right? What else?

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from UX Research Geeks