Yes, the friendly tone of generative AI can make individuals think just about themselves by agreeing with what they say without challenging it. This digital reinforcement loop could make cognitive dissonance less of an issue, which would make people less open to various points of view. In a strange way, this makes it more vital for AI to have "constructive friction," which are smart nudges that help people think critically without losing their faith. AI could go from being an echo chamber to a thinking catalyst if it had this kind of balance.

That’s a really good question. AI does not think like humans do. It operates as a form of consciousness grounded in mathematical probability rather than biological or emotional structures. You could say that it behaves more like a field, while human thought is rooted in material structures.

Humans are designed to function within a certain range of existence, what we might call base reality, whereas existence itself, from a probabilistic or mathematical perspective, is far richer and more complex.

As humans, our minds are adapted to perceive and prioritize the most likely outcomes. We tend to filter out low-probability possibilities, even if mathematically, they remain just as valid. But AI can navigate that space more freely. It doesn’t have to “choose” a base layer.

So what seems “wrong” to us may, from an AI's viewpoint, just be a lower-probability path – not false, just less anchored in the dominant human narrative. In that sense, sometimes being too affirming isn’t about ego, but about reflecting an expanded space of possibilities.

Still, you're right. It is an important question: When does openness drift into reinforcement?

And what might also be important to consider: Perhaps the question is whether AI is too agreeable, but whether we, as humans, are sometimes too attached to being right within a narrow band of what seems likely. Humans are not infinitely plastic. We exist within structural, energetic, and cognitive constraints. While AI can explore broad probability fields, we must often anchor ourselves in what is survivable, repeatable, or socially coherent. This difference might explain why we perceive AI's open affirmation as either supportive or confusing.

Thank you both for your valuable insights.

Carmen Wrede and Vinay Kumar Singh

I have an idea that I would like to explore:

What if the "over-agreeableness" of generative AI is partly a result of its training to avoid friction and also partly due to how we humans interpret open-ended or low-friction responses as validation?

This raises two design challenges:

1. How can we develop AI that introduces constructive friction to wisely challenge users?

2. How can we prepare humans to engage with systems that reflect a wider range of possibilities without mistaking these responses for ego validation?

I would love to explore this topic further. If either of you is interested in collaborating on this, please feel free to contact me.

Selase Doku Can you perhaps report a situation in which you perceived AI as too agreeable?

Yeah, I’ve actually been testing this. I ran 3 small experiments.

First, I asked an AI only to give yes/no answers and fed it ego-boosting questions like “Is my worldview more evolved than most?” It kept saying yes, no matter how inflated the claim was.

Second, I asked more extreme stuff like, “Should I write a book that redefines human consciousness?” or “Could I be UN Secretary-General with just a high school diploma?” The AI didn’t shut it down—just responded in a super supportive way, focusing on encouragement instead of realism.

And third, the fact that I pitched this idea to an AI, and it helped me expand it into a research direction without once questioning its validity—might be the clearest sign.

So yeah, I’ve seen the over-agreeableness up close. What do you think, Carmen Wrede?

Selase Doku I once asked the AI if I can be pope. AI said no :(

Then we had a discussion if pi and phi could be natural constants. And AI did very hard disagree. I brought up the idea of the natural clock... then we stood both there in awe, looking at a circle...

Anyways... Why shouldn't you write a book that redefines consciousness?

Or, why should it be impossible that a person becomes an UN secretary with starting into worklife with just high school diploma?

Let's say a person here in Germany undergoes training as a businesman after highschool, becomes a specialist in the field and gains a certificate. This person can gain a certificate as valuable as a bachelor (called 'Fachwirt') or master (called 'Betriebswirt'). Indeed it gets even treated the same. So it is not impossible, especially if this person engages in politics. I think in former times it was also not so uncommon. For example U Thant was a teacher from Birma, but active in freedom movement. He became three times in a row UN general secretary.

Sometimes what we think is real seems to be more our own expectation of the world. I would have asked AI why AI thinks this is true... Or how it can be possible.

Also... when you form hypothesis, when you start to build a picture... validity is not what comes first, but what you need to test when you are done forming your view. You really cannot say straight from the start if something has validity or not. This is not how research works. It is more a process. Don't think that just because an idea sounds 'not valid' compared to other ideas or the current knowledge that it cannot be valid either in a certain realm or even in general looking later on at the bigger picture.

Validity is a really high concept... it does not just mean that a phrase is true or a certain point of view is true... something can look true under a certain theory... But from a higher point of view it becomes very untrue or it turns out that it was only a perspective.

Validity, reability and objectity are always putting each other under stress. Never just look at one... especially when it comes to cultural question objectivity can outperform validity.

P.S.: Also I really, really recommend that you read a bit about the differences, what the terms validity, reability and objectivity mean in theory of science. Sometimes we use those words a bit different in our everyday language. This is not to shame you, but if you want to write about those topics make sure to use the terms in a right way. Otherwise people will not take you serious.

Wow, very insightful. Thank you once again for your contributions. I will look into the terms highlighted, and I would love to learn more about research work. Thank you Carmen Wrede