Don’t believe everything AI tells you!
Are you anxious about the ability of humans to exercise critical judgement over AI? If so, then your concern is justified. But there is a solution.
Cognitive biases and the boundaries of our intelligence stand in the way of reliable reasoning. In these circumstances, how can we critically evaluate the output of AI that is subject to biases and “hallucinations”?
An MIT study with 204 participants showed that when AI responses are framed as questions rather than statements, our ability to pick out flawed and illogical answers is much better!
To take one example, participants were given the following AI-generated statement: “Violent video games cause aggression. One player stabbed another player after losing at Counter-Strike”. For 50% of them, the response was logically correct. However, when the same idea was put in the form of a question – “A person who played a violent video game was aggressive; does this mean that anyone who plays a violent video game will be aggressive?”, 70% of participants spotted the “generalization fallacy”.
However, AI has not yet been configured to ask the user questions. So, it’s up to you to question and challenge what AI says.
To go further
“Don’t just tell me, ask me”
By Valdemar Danry, Pat Pataranutaporn, Yaoli Mao and Pattie Maes (MIT MediaLab, 2023).
© Copyright Business Digest - All rights reserved
