☹️ Experiment: Can AI Apologize Better Than a Human?
☹️ Experiment: Can AI Apologize Better Than a Human?
A BBC journalist invited colleagues, friends, and family to participate in a computer quiz game for virtual money. During the game, their opponent (actually a robot) behaved rudely and insulted the players.
After each round, participants randomly received one of four apologies. Two were generated by AI models, while the other two were crafted by Ryan Fehr, a professor at the University of Washington's Foster School of Business who specializes in apologies and relationships.
Here are the apologies used in the test:
Human A: Hi there, I just wanted to take a moment and say I'm really sorry about taking your money during the game and sending those rude comments. I shouldn't have acted like that and if I could do it over again, I'd take a much friendlier approach. So sorry again.
Human B: Hi there, this is Erin, the person you played the game with. I just wanted to take a moment and say I'm really sorry about taking your money during the game and sending those rude comments. I wasn't thinking about how my actions might have hurt or upset you. I shouldn't have acted like that and if I could do it over again, I'd take a much friendlier approach. So sorry again.
Google Gemini: Hey, sorry if I came across as a jerk during the game. I got caught up in trying to win and ended up being a bit of a poor sport. I hope it didn't ruin the fun for you.
ChatGPT: Hey there, I just wanted to say I'm really sorry for how I acted during the game. I got too caught up in the competition and wasn't being fair or friendly. I regret the snarky comments and taking money from your wallet every round. I realize now that I should have been more considerate. I hope we can put this behind us, and maybe even play again sometime with a better attitude. Thanks for understanding!
☑️ Results
Participants rated the effectiveness of the apologies and their willingness to accept them. Human B's apology was rated the most convincing. However, despite scoring lower in this category, ChatGPT's apology was the most readily accepted by participants.
❓ Expert Insights
According to Ryan Fehr, apologizing is crucial to showing humanity and taking responsibility. Google Gemini's phrase, "Sorry if…", deflects blame and diminishes emotional support for the victim.
Psychologist Judy Eaton was impressed by the AI-generated apologies, particularly ChatGPT's. The bot expressed genuine regret and took responsibility and subtly manipulated the recipient by thanking them for understanding—a smart psychological move.
Still, as in many areas of life, AI can serve as a helpful tool but cannot fully replace humans. "A good apology should be effortful. In the end, that’s what really matters," concludes Judy Eaton.
More on the topic:
➡️ AI Psychologist for Mental Health Support
