Beyond Anthropomorphism: Exploring the Roles of Perceived Non-humanity and Structural Similarity in Deep Self-Disclosure Toward Generative AI
Based on a 2025 survey of 2,400 participants, this study reveals that deep self-disclosure toward generative AI is significantly driven by the combined effects of perceived non-humanity and structural similarity, suggesting that trust-related behaviors extend beyond traditional anthropomorphic perceptions.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: Why We Tell Secrets to Robots
Imagine you have a heavy secret you've never told anyone. You wouldn't tell it to your boss, your best friend, or even your mom because you're afraid of being judged, laughed at, or having your secret spread around.
But, strangely, you might feel safe telling this exact same secret to a chatbot. Why?
This paper argues that the reason isn't just because the robot is "nice" or "human-like." In fact, the study suggests that the robot being a machine is actually the best part.
The author, Satoru Shibuya, proposes a new way to understand this using two main ingredients:
- The "Safe Wall" (Non-Humanity): Knowing the AI isn't a person removes the fear of judgment.
- The "Logic Mirror" (Structural Similarity): Feeling like the AI actually understands how you think, not just what you say.
When you have both a "Safe Wall" and a "Logic Mirror," you are much more likely to open up deeply.
The Two Ingredients Explained
1. The "Safe Wall" (Perceived Non-Humanity)
Think of talking to a human like walking through a crowded town square. Everyone is watching, and you worry about what they think of your clothes, your voice, or your story. This is called "evaluation apprehension."
Now, imagine walking into a soundproof glass booth where no one can see or hear you. That is what talking to a non-human AI feels like.
- The Paper's Claim: Because the AI is an "emotionless machine," you don't feel the pressure to look good or act perfect. You feel safe because there is no one there to judge you.
- The Analogy: It's like writing in a diary that no one will ever read. You can be raw and honest because the "reader" isn't a person who can gossip.
2. The "Logic Mirror" (Perceived Structural Similarity)
Just having a safe wall isn't enough. If you talk to a wall, you might feel safe, but you won't feel understood. You need to feel like the other party "gets" your train of thought.
The paper uses a concept called "Structure-Mapping." Imagine your thoughts are a complex train track with many switches and loops.
- The Paper's Claim: When the AI responds, it doesn't just say "That's sad." It follows the exact same track you are on. It connects the dots in the same order you did.
- The Analogy: It's like having a conversation with someone who speaks your specific dialect of logic. They don't just hear your words; they see the blueprint of your mind. When you feel your "mental blueprint" is being perfectly mirrored, you trust them with your deepest thoughts.
The Experiment: Mixing the Ingredients
The researcher looked at data from 2,400 people in Japan who used AI in 2025. He sorted them into four groups to see who told the deepest secrets:
- Group A (The Baseline): People who didn't feel the AI was safe and didn't feel it understood them. (Low secret-telling).
- Group B (Safe Wall Only): People who felt safe because it was a robot, but didn't feel it understood their logic. (Moderate secret-telling).
- Group C (Logic Mirror Only): People who felt the AI understood their logic, but didn't necessarily feel the "non-human" safety. (Low-to-Moderate secret-telling).
- Group D (The Super-Combo): People who felt both safe because it was a machine AND felt it perfectly mirrored their thinking.
The Result:
Group D was the winner by a landslide.
- People in this group were 11 times more likely to share deep, sensitive secrets compared to Group A.
- They didn't just share facts; they shared emotions, vented frustrations, and talked about things they usually hide.
Who Was Most Likely to Open Up?
The study found this "Super-Combo" effect was especially strong in two groups:
- Men in their 30s and 40s: These men often carry heavy social masks (being the boss, the provider, the strong one). The AI offered a place where they could take off the mask without fear of social punishment.
- Gen Z (18–29): This group lives in a world of constant social media judgment. The AI offered a break from the "attention economy" where they are constantly being rated. They felt safe because the AI wasn't a human judge, and they felt understood because the AI matched their fast-paced, digital way of thinking.
What the Paper Does NOT Say (Important Limits)
It is crucial to stick to what the paper actually found:
- It's a Snapshot, Not a Movie: The study looked at data from one specific time in 2025. It shows a connection between these feelings and sharing secrets, but it doesn't prove that the feelings caused the secrets. (It's possible that people who love to share secrets just happen to think the AI is cool).
- It's Exploratory: The researcher admits the tools used to measure "depth" were a bit rough (like using a ruler to measure the depth of a well). The results are promising but need more testing.
- No Clinical Advice: The paper does not say AI should replace therapists. In fact, it warns that because people feel so safe, they might share too much sensitive information, which could be risky.
The Bottom Line
We often think we trust AI because it acts like a human. This paper suggests the opposite: We trust AI with our deepest secrets because it is NOT human.
When you combine the relief of not being judged (because it's a machine) with the satisfaction of being understood (because it follows your logic), you get a powerful psychological space where people feel free to be completely honest.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.