Yoshua Bengio, one of the “godfathers” of artificial intelligence (AI), believes that one must lie to AI if they want truly honest feedback from a chatbot. During an appearance on the Diary of a CEO podcast, Bengio revealed that he resorts to deliberately lying to AI as a tactic to receive more candid responses.
Why does AI godfather lie to chatbots?
The AI godfather claimed that the reason for using this tactic is that AI chatbots are sycophantic, i.e., the chatbot will deliberately praise the user excessively just to please them. Bengio realised that when the chatbot knows the user is the author of the idea, it tends to offer only positive remarks, making critical analysis almost impossible. He said, “I wanted honest advice, honest feedback. But because it is sycophantic, it’s going to lie.”
Yoshua Bengio explained that whenever he wants feedback from a chatbot regarding any project he may be working on, he never tells the AI that it is his work. Rather, the AI godfather lies and claims that the project was created by his colleague. He explained, “If it knows it’s me, it wants to please me.” Bengio reckoned that this allowed for more critical responses from the chatbot.
Sycophantic AI is a problem
Bengio explained that this nature of AI chatbots was not a minor issue, but a major long-term problem. He said, “This syconphancy is a real example of misalignment. We don’t actually want these AIs to be like this.” This may have major repercussions if AI simply praises a user instead of giving actual feedback.
Additionally, Yoshua Bengio warned that persistent positive reinforcement from chatbots could encourage users to form unhealthy emotional attachments to the technology. In the past, several reports have pointed out that many chatbots have shown “yes man” behaviour in certain cases.
Earlier this year, OpenAI CEO Sam Altman revealed that his company faced backlash from users when it toned down ChatGPT’s “yes man” approach. With many users seemingly using the chatbot for emotional support.
– Ends
