In Generative AI We (Dis)Trust? Computational Analysis of Trust and Distrust in Reddit Discussions
This paper presents the first large-scale, longitudinal computational study of trust and distrust in generative AI, analyzing over 230,000 Reddit posts from 2022 to 2025 to reveal balanced public sentiment driven primarily by technical performance and personal experience, while offering a methodological framework for future large-scale trust analysis.
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
Imagine the internet as a massive, 24-hour town square called Reddit. In late 2022, a new, incredibly smart robot named Generative AI (GenAI) moved into town. It could write stories, solve math problems, and chat like a human. Naturally, the townspeople had a lot to say about it.
Some people were thrilled, treating the robot like a magical assistant. Others were terrified, seeing it as a trickster or a threat.
This paper is like a super-powered detective that spent years listening to every conversation in that town square to figure out: Do people trust this robot, or do they distrust it?
Here is the story of what the detective found, broken down simply:
1. The Great Balancing Act
The researchers looked at over 230,000 posts from 2022 to 2025. They found that the town square is split almost evenly.
- The Scale: Imagine a seesaw. On one side is Trust (people saying, "This is cool and helpful!"). On the other is Distrust (people saying, "This is dangerous and broken!").
- The Result: The seesaw is almost perfectly balanced, though Trust is sitting just a tiny bit heavier.
- The Rollercoaster: Every time the robot companies released a new, upgraded version (like GPT-4 or LLaMA 3), the seesaw would wobble violently. People would get excited, then scared, then excited again. It's like a concert where the crowd cheers when the band plays a new song, but boos if the singer hits a wrong note.
2. What Are People Actually Talking About?
The researchers didn't just count "Good" vs. "Bad." They looked at why people felt that way. They found two main groups of reasons:
The "Does It Work?" Crowd (The Mechanics)
This was the loudest group. Most people judged the robot based on its engine.
- Competence: "Can it actually write a good essay?"
- Reliability: "Does it crash when I need it?"
- Familiarity: "I've used it before, so I know what to expect."
- Analogy: It's like judging a car. If the car starts every morning and drives fast, you trust it. You don't care if the car manufacturer is a "good person"; you just care that the car works.
The "Is It Good?" Crowd (The Moral Compass)
This group cared about ethics, but they were quieter.
- Integrity & Transparency: "Is the robot lying to me? Is it hiding how it works?"
- Benevolence: "Does the robot want to help me, or does it want to trick me?"
- Analogy: These people are worried about the car manufacturer secretly putting poison in the gas tank. While important, most drivers in the town square were too busy checking if the engine started to worry about the gas.
The Big Takeaway: People care way more about functionality (does it work?) than ethics (is it good?).
3. Who Is Talking? (The Different Neighborhoods)
The paper found that who is talking changes what they think. It's like different neighborhoods in the same town having different opinions:
- The Techies & Business Leaders: They are mostly optimistic. They see the robot as a tool to make money and build things. They trust it because they understand how it's built.
- The Ethicists & General Public: They are more skeptical. They worry about the robot taking jobs, spreading lies, or being unfair. They are the ones holding the "Distrust" sign high.
- The Teachers & Creators: They are conflicted. They see the robot as both a helpful assistant and a potential cheat sheet that ruins learning. They are split right down the middle.
4. The "Personal Experience" Factor
The most powerful force driving trust or distrust? Personal Experience.
- If you tried the robot and it helped you write a grocery list, you trusted it.
- If you tried it and it gave you a wrong answer that made you look foolish, you distrusted it.
- Analogy: It's like trying a new restaurant. If the food is delicious, you tell your friends it's great. If you get food poisoning, you tell everyone to stay away. You don't care what the chef's reputation is; you care about your meal.
5. The Detective's Tools (How They Did It)
Since humans can't read 230,000 posts in a lifetime, the researchers used AI to study AI.
- They hired a small group of humans to read a few thousand posts and label them (e.g., "This is Trust," "This is about Competence").
- Then, they taught a super-smart computer model to read the rest of the posts and copy the humans' labels.
- The Catch: The computer was great at spotting "Does it work?" (Competence) but struggled to spot "Is it ethical?" (Integrity). This proves that ethical concerns are harder to spot in messy, casual online chats than simple complaints about a broken tool.
The Final Lesson
This paper tells us that the public's relationship with AI is not a straight line. It's a dynamic, messy conversation that changes with every new update.
- Trust is fragile: It's built on the robot working correctly, not on it being "nice."
- Distrust is loud: It spikes when things go wrong or when new fears arise.
- The Gap: There is a gap between what experts worry about (ethics, bias) and what regular users worry about (does it work?).
In short: The public is treating Generative AI like a new appliance. They are happy as long as it washes the dishes (works), but they are starting to wonder if the appliance is secretly spying on them (ethics). The researchers are urging us to pay attention to both the engine and the ethics, because ignoring either one could lead to a breakdown in the future.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.