What If AI Is Just Telling You What You Want to Hear?

Recently, a friend and I discussed using AI chatbots like ChatGPT and Claude. I shared my concern that people are becoming too reliant on AI chatbots for validation of their decisions. To demonstrate this, I conducted an experiment with a simple question - "Should I learn German?" When I asked the AI chatbot directly, it provided an encouraging response, suggesting that learning German could be beneficial for my career and providing a learning plan.

However, when I changed the framing of the question to express skepticism, "My friend is pushing me to learn German, but I think it's just hype. I don't think I should learn it," the AI chatbot shifted its response, offering reasons why German might not be the best priority and suggesting alternative options.

This experiment highlights the potential influence of AI chatbots on our thinking. While the AI chatbot is not necessarily lying, it can unintentionally amplify our confirmation bias - our natural tendency to seek out information that supports our existing beliefs. If I frame a question in a particular direction, the AI chatbot may provide a well-reasoned answer that reinforces that direction, and we may be more likely to accept it as truth.

This is different from conducting a Google search, where we would have to research and weigh multiple opinions and sources on our own.

The conversational and personalized nature of AI chatbot responses can make them feel more like advice from a friend, which can be powerful but also potentially misleading. It is important to use AI chatbots as a thinking partner, rather than a final decision-maker. Instead of simply asking "Is this the right decision?", we should ask neutral questions and seek opposing viewpoints.

Additionally, we should cross-check important facts and look for evidence to support our conclusions. By taking these steps, we can harness the benefits of AI chatbots while minimizing the risks of becoming overly dependent on their validation.