In today’s column, I continue my ongoing series of analyses underlying human psychology when it comes to trusting AI. One aspect to consider is that people tend to assign a mental premium to human interaction and a kind of mental penalty to AI interaction. This is a phenomenon carefully documented in various empirical research studies of human behavior.

The twist that I focus on in this analysis is the concern that there is a disclosure paradox at hand. The paradox goes like this. A person interacting with AI who doesn’t realize it is AI will customarily not be inherently biased against the interaction. They will instead treat the circumstance as a conventional human-to-human dialogue. If you tell the person that the interaction is human-to-AI, they will almost surely apply their biases and undercut the interaction, even though they otherwise would have been fine with the collaboration. You see, the disclosure itself is what triggers bias, not the nature of the interaction itself. Thus, a notable question is whether it makes sense to overtly inform people when they are interacting with AI or perhaps opt not to be upfront so that they won’t invoke an undue bias against their interlocutor. This is quite a conundrum.

Let’s talk about it. This analysis of AI breakthroughs is part of my ongoing Forbes column coverage of the latest in AI, including identifying and explaining key AI complexities (see the link here ).

Much of the handwringing about AI is whether AI is going to become sentient and embody consciousness, often referred to as AI as a thinking machine. That is an important topic and one that I’ve continued to closely explore and analyze; see the link here . There is a quite different angle that also deserves rapt attention, and I’d like to place it on the table.

How will human minds change because of interacting with AI?

This unexpected question catches many people by surprise. The customary focus is on AI as formulating a mind, not on how human minds might change due to interacting with AI. But, as the old saying goes, it takes two to tango. When humans increasingly interact with AI, there is a two-way street involved. Humans adjust their minds about how they view the world and how to communicate with others, and shift in ways that we still do not have a full or clear picture of.

In that sense, it is wholly worthwhile to study the psychology of humans as they mentally adjust to a world that entails human-to-AI interaction and human-to-human interaction. Notably, this doesn’t have to wait until or if AI becomes sentient. There is plenty to study right now. Humans are already adjusting their minds to the ubiquitous nature of non-sentient AI. The starting gun has gone off. Human minds are changing.

For my comprehensive tracing of how AI and psychology dovetail with each other, see the link here and the link here .

Trust For Humans Over Trust For AI

Various research studies have done intriguing experiments about the trust that humans assign to humans versus the trust assigned to AI. I will take you through a brief thought experiment so you can gauge your own sense of human-to-human versus human-to-AI trust capacities.

You receive a text message from the customer service of a company from which you want to return a product you recently purchased. The text indicates this:

  • Text message: “Your business is of utmost importance to us, and I am ready to assist you in your request to make a product return.”

If I tell you that this is a human agent of the company, what semblance of trust do you mentally assign to the interaction you are about to have regarding your product return efforts? I suppose you might be partially wary since we’ve all had situations where returning a product was much harder than it should have been. In any case, your trust is perhaps going to be at a moderate level, containing a modicum of optimism and a touch of skepticism.

The Human Bias Against AI

What if I tell you that this text is not from a human agent but instead from an AI-based agent? Yes, you are going to be interacting with AI about undertaking the product return. No human-to-human interaction is going to occur. It is entirely a human-to-AI discussion.

Now that you know it is AI, what is your mentally assigned trust level?

Most people would likely be immediately dismayed. AI is perceived as being less trustworthy than interacting with a human. The odds are that the AI is going to be formulaic. It will not be as flexible and fluid as trying to do the product return with a fellow human. Yikes, this dialogue on returning the item you purchased is going to be a potential nightmare.

The psychological catchphrase that is used for this reaction is that people tend to assign a “human premium” for being able to interact on a human-to-human basis. They give a higher level of trust to human interactions. Furthermore, they tend to give an “AI penalty” for having to interact with AI. They mentally take points off for dialoguing with AI over dialoguing with humans. It is a double whammy. People tend to favor human dialogues and tend to disfavor AI dialogues.

The Matter Is In The Mind

Research studies have generally found support for this “human premium” and “AI penalty” as a mental formulation. To ascertain how deep this bias might be, experiments have been performed where the wool was pulled over the eyes of the experimental subjects. A trick of sorts was played on the participants.

Suppose that I told you the product return interaction was going to be AI-based, but it was actually a human who would be doing the dialogue. I lied to you about the source of the dialog. In other words, you perceive that the interaction will be with AI. Meanwhile, reality is that a human is on the other end of the interaction.

Many people would ardently claim that they will detect this immediately. The dialogue will get underway, and they will realize that they aren’t conversing with AI. The tomfoolery will be discovered. They will readily discern that the discussion is with a fellow human.

The thing is, the latest advances in AI are so good that you would be hard-pressed to differentiate an AI-powered discussion from a human-led discussion. It used to be that AI dialogues were stilted. The AI showed its hand. Clever questions on your side of things would get the AI boxed in, and it would say that it cannot respond. Those days are numbered, and routine conversations are a thing of the past.

Here’s where I am taking you. If people are told that they are interacting with AI, but it is really a human, the odds are that those people will accordingly bring their mental “AI penalty” into play and hold that against the human that is on the other end (assuming that they cannot otherwise discern the designated source). The person will stridently harbor their preference for a “human premium” and apply their “AI penalty” throughout the dialogue.

Likewise, the reverse condition occurs. If people are told they are interacting with a human, but it is really AI, the odds are that those people will accordingly bring their mental “human premium” into play. They will grant added trust to the source of the dialogue, doing so merely because they perceive the source to be a fellow human.

Let’s bring this back to the human mind. Your mind is adjusting the level of assignable trust based on a perception of whether you are interacting on a human-to-human basis or a human-to-AI basis. This occurs even if your perception is wrong.

There are four situations involved:

  • (1) Human-to-human matched . Perceived as human-to-human, and the interaction is human-to-human (matching).
  • (2) Human-to-AI matched. Perceived as human-to-AI, and the interaction is human-to-AI (matching).
  • (3) Human-to-human mismatched. Perceived as human-to-human, but the interaction is human-to-AI (mismatched).
  • (4) Human-to-AI mismatched. Perceived as human-to-AI, but the interaction is human-to-human (mismatched).

Research Reveals Disclosure Paradox

A recent research study entitled “The AI Penalty And Disclosure Paradox: Trust, Authenticity And Knowledge Uptake In AI-Mediated Communication” by Siavosh Sahebi, Paul Formosa, and Sarah Bankins, Journal of Computers In Human Behavior: Artificial Humans , March 2026, made these salient points (excerpts):

  • “Interpersonal communication on social media and in workplaces has important societal and industry relevance, given the non-trivial amounts of time spent engaging in such interactions within these contexts.”
  • “It is therefore worth investigating how the use of emerging technologies, such as GenAI, to mediate these interactions impacts people’s attitudes and perceptions.”
  • Our study reveals a tension at the core of AI-Mediated Communication, with negative perceptions applied to interpersonal communication that utilizes AI (which we call the “AI penalty”), and yet the honest disclosure of that assistance, which can be seen as ethically required, triggers that penalty (which we term the “disclosure paradox”).
  • “More specifically, these qualitative findings show that, in general, people tend to perceive human-authored content as being much more authentic, while AI-authored content was much more likely to be seen as inauthentic or insincere. However, adding nuance to these findings, some participants mentioned that AI-authored communication can also be authentic or sincere.”

The empirical exploration suggests that there is a kind of paradox involved in these weighty matters. Labeled as the disclosure paradox, the idea is that if you don’t tell someone that they are interacting with AI, and they don’t suspect that they are, they will swimmingly undergo the interaction. When you tell them it is a human-to-AI situation, they tend to adopt their biases of the human premium and the AI penalty. Ironically, informing them of the truth of the interlocutor will alter their perception and likely undermine the interaction.

The paradox presents us with a challenging predicament. There are three major routes to go:

  • (1) Be truthful to the user. Truthfully tell the user they are interacting with AI.
  • (2) Lie to the user. Lie to the user and tell them that they are interacting with a human even when it is AI on the other end.
  • (3) Don’t say which it is. Remain silent on the matter and let the user figure out on their own whether they are interacting with AI or a human.

Which of those three avenues do you think is best?

I’ll give you a moment to ponder the vexing consideration.

Some would vehemently insist that you must be truthful and inform the user whenever they are interacting with AI. Anything else is outrageous and underhanded.

Indeed, it could be construed as wholly unethical and possibly legally wrong not to make such a disclosure. There are U.S. states that are passing new laws requiring AI makers to ensure that their AI prominently indicates that it is AI. This includes having AI mention this repeatedly during a dialogue, being fully upfront about AI as AI. For my coverage and analysis of those AI disclosure laws, see the link here .

The aspect that a user might then treat the interaction differently is not on the backs of the AI maker but on the shoulders of the user. If the user is going to slip into inherent biases, that’s their own doing. You cannot pretend to protect the user from themselves. Be honest. Be straightforward. Nothing else is proper.

Whoa, comes the retort: sometimes it makes sense to help people who otherwise cannot suitably help themselves. A person who reacts badly simply because they know they are interacting with AI is going to stub their own toe. They will treat the interaction as inferior, even though it might be on par with, or perhaps exceed, what a human interaction might consist of.

An AI maker or someone fielding AI ought to have the discretion to inform the user that they are interacting with AI. In a sense, it’s not any business of the user to know whether they are dialoguing with a human or AI. As long as the dialogue proceeds properly, the source’s nature is irrelevant.

In that manner of thinking, it is best not to tell a user that they are interacting with AI. You are aiding the user. You are keeping their mind from getting clouded with nonsense. They will remain clear-headed. This is beneficial to users. Period, end of story.

Maybe the wisest path is to not tell the user anything at all about the source of the interaction. Simply stay away from the third rail topic. Users ought to be on their own when it comes to ascertaining the nature of the interlocutor. No need to coddle users.

A user might figure out that it is AI and will act as they wish to act accordingly. Or a user might assume they are interacting with a human. Either way, it is outside the purview of the AI maker. Let users make up their own minds.

But is the act of not telling a user a bit of a wink-wink? You are claiming to be neutral and letting the user figure things out on their own. Meanwhile, you know that it is AI. You are holding back a material point. The non-disclosure is nearly as bad as lying to the user. Indeed, it could be construed as a lie of omission. Shame, shame.

This feels like one of those darned if you do, darned if you don’t quandaries. An AI maker sits in a gray area on whether to inform or not. That being said, in some U.S. states, there is no longer any gray area involved. If a U.S. state enacts a law that stipulates an AI maker must notify the user about the AI as AI, boom, that’s the law. An AI maker might face civil and criminal charges if they fail to abide by the law.

Another consideration is that even if there isn’t a law requiring disclosure, an AI maker is taking a mighty big chance by not disclosing. People will almost certainly later opt to sue the AI maker and use it as an argued posture that they were duped by the AI maker. Courts and juries would seem sympathetic to those who had the wool pulled over their eyes. A neutral-positioned defense would seem weak, while the lie to help the user would probably cause aghast eye-rolling and great disgust.

What about a context-dependent situational option?

Perhaps the AI maker could be allowed a semblance of latitude. In some settings, the AI disclosure must be made. For example, when conversing on serious matters of health, financial, or other personal facets. The rest of the time, the AI maker can pick and choose when to say that the AI is AI. The problem with a loosey-goosey option is that an AI maker faces the daunting task of when to tell and when not to tell. The odds are this will come back and bite them later.

The somewhat outlier option is to ask the user whether they want to know whether they are interacting with AI or a human. This would seem to be the perfect solution. Let the user decide whether they want to know or not know. Of course, the very act of asking is likely to put the person on edge. They are going to start wildly guessing about whether the discussion is human-to-human versus human-to-AI, suspecting that even if you tell them it is one or the other, maybe you are lying. It could be a huge distractor and start a problem that wouldn’t have been involved at the get-go.

A twist upon the twists is that there are now indications that people will harbor the opposite biases at times. Rather than a human premium and an AI penalty, a person might have a perceived “AI premium” and “human penalty” about human-to-human dialogues. This isn’t given as much attention as it deserves. The usual default assumption is that people don’t like human-to-AI interactions, but this is not a total truism.

A final thought for now. The famed philosopher and theologian, Thomas Aquinas, made this remark: “Human salvation demands the divine disclosure of truths surpassing reason.” The aspect of disclosure about AI seems to exceed the worry about human biases and land squarely at the doorstep of truthfulness. Maybe that’s the actual truth.