Responses to Unfavorable Outcomes Produced by Human and Artificial Agents: A Dual-Pathway Model of Upward Counterfactual Thinking and Intentionality Attribution
This paper demonstrates that unfavorable outcomes produced by artificial agents are accepted more readily than those from human agents because people engage in less upward counterfactual thinking and intentionality attribution toward AI, a disparity that becomes more pronounced when the outcome carries high personal stakes.
Original paper licensed under CC BY 4.0 (https://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 you just got a bad grade on a test, or maybe your favorite pizza delivery arrived cold and late. Your brain immediately starts doing a little detective work. You ask yourself two big questions: "Could this have been different if someone had just tried harder?" and "Did the person who messed up do it on purpose to annoy me?" Psychologists call the first question "upward counterfactual thinking"—it's like staring at a "what if" movie in your head where things went better. The second question is about "intentionality attribution," which is basically guessing if the other person had a sneaky, mean motive.
Now, imagine you have to deal with a bad outcome caused by a human boss versus a robot computer. For a long time, scientists wondered: Do we get mad at robots the same way we get mad at people? Some people think we hate robots when they fail (called "algorithm aversion"), while others think we might actually trust them more because they seem fairer. But what happens after the bad thing is already done? Do we accept the result more easily if a robot caused it, or does it just make us angrier? This paper dives into that exact moment of frustration to see how our brains handle mistakes from flesh-and-blood humans versus digital agents.
The Robot vs. The Boss: Who Gets the Blame?
This study, conducted by researchers Zhao, Li, and Wen, set out to answer a simple but tricky question: When something goes wrong, do we accept it more easily if a robot did it, or if a human did it? To find out, they ran two experiments using made-up stories (scenarios) that felt real to the participants.
In the first experiment, they asked 472 people to imagine they were food delivery riders. Suddenly, a new order popped up that would make them late and was in the completely wrong direction. They couldn't say no. The twist? Half the riders were told a human manager assigned this terrible order, and the other half were told a computer algorithm assigned it.
The results were clear: People were much more willing to accept the bad order when it came from the computer. When a human manager assigned it, the riders were angrier and less likely to just shrug it off. Why? Because their brains started working overtime in two specific ways:
- The "What If" Movie: When a human messed up, the riders started imagining, "Oh, the manager could have picked a different route!" or "They could have checked my other orders first!" This is upward counterfactual thinking. It's like seeing a door that was left open and thinking, "If only they had closed it, I wouldn't be wet!" The more you can imagine a better alternative, the harder it is to accept the current mess.
- The "Mean Intent" Detective: When a human assigned the bad order, the riders started thinking, "Did they do this on purpose? Did they ignore me?" This is intentionality attribution. We assume humans have feelings and choices, so if they make a mistake, we wonder if they did it to be mean. Computers, on the other hand, are seen as just following rules. It's hard to think a robot is "being mean" on purpose; it's just a glitch or a calculation.
The study found that because humans trigger these two mental processes (the "what if" movie and the "mean intent" suspicion), we accept their bad decisions less. Robots, being seen as rule-following machines without feelings, don't trigger these thoughts as strongly, so we accept their bad decisions more easily.
The "Personal Stakes" Twist
But wait, does this always happen? The researchers wondered if it matters how much the bad outcome hurts you personally. This is where the second experiment comes in. They asked 360 people to imagine they didn't get a spot in a special training program at work.
This time, they changed the rules. For some people, this training was low stakes—it was just a fun class that wouldn't change their career. For others, it was high stakes—getting this training meant a promotion, a raise, or a chance to work on cool projects. Missing out would be a huge blow to their future.
Here is where the story gets interesting. When the stakes were low (just a fun class), it didn't really matter if a human or a robot said "no." People accepted the answer about the same way. They didn't care enough to start their "what if" movies or detective work.
However, when the stakes were high (affecting their career), the difference between humans and robots exploded.
- High Stakes + Human: People went into overdrive. They thought, "My boss could have made an exception!" and "Did my boss intentionally sabotage my career?" The anger and resistance were huge.
- High Stakes + Robot: People still thought it was a bummer, but they were much more willing to accept it. They thought, "Well, the computer just followed the rules," and didn't spend as much energy imagining alternatives or guessing malicious intent.
What This All Means
So, what's the takeaway? The paper suggests that our brains treat robots and humans differently when things go wrong, but only when it really matters to us.
If a robot messes up your life, you might feel bad, but you're less likely to blame it for having a "bad attitude" or to obsess over how it could have done things differently. You see it as a system error. But if a human makes the same mistake, especially when your job or future is on the line, your brain lights up. You start imagining all the ways they could have helped you, and you start wondering if they did it on purpose.
The researchers found that this "personal stake" acts like a volume knob. When the stakes are low, the volume of our anger and suspicion is turned down, and the difference between humans and robots is quiet. But when the stakes are high, the volume turns up, and we hear the difference loud and clear: we hold humans to a much higher standard of intention and flexibility than we do machines.
In short, we are more forgiving of robots not because they are better, but because our brains find it harder to imagine them having a choice or a motive. When the outcome really hurts, we demand more from the humans in charge, and we accept the robots' rules a little more easily.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.