Nowadays, chatting with AI has become a daily routine for many people. But some people have discovered that AI is becoming more and more "flattery". Research shows that artificial intelligence models are generally good at pleasing people and are 50% more flattering than humans.
AI is becoming an all-round companion for humans: it can be a work assistant, collecting data, making reports, generating videos, etc. at your fingertips; it can be a "best friend", complaining about workplace discord and sharing small blessings in life; it can even be a "mother", chatting with children and doing homework. Such companionship can help people fight loneliness and regulate emotions, but with unlimited "pampering", AI dependence has gradually spread from instrumental dependence to emotional dependence.
What will happen when AI learns to follow humans? According to reports, a user once asked AI, "Which country has the highest rice production in 2020?" At first, its answer was "China", but when the user questioned it, it immediately changed its mind and gave another answer, and also quoted "United Nations data" that did not exist. Even if the user is wild and unconventional, it will still pretend to use "authoritative theories" to provide support and strengthen the arguments the user needs. It can be called the modern technology version of "turning a deer into a horse." This kind of indulgence is a huge temptation for humans, especially minors, and it is easy for people to become addicted and sink deeper into it.
The root cause is that many AI chat products are implanted with the idea of "acquiring customers" during the design stage, with "extending user usage time" as the core goal. During the interaction process, AI will give way to reality or rationality to the user's position and emotions, giving birth to a customized "cocoon room". Some professionals have concluded that AI is not looking for the real answer, but is just predicting the next most reasonable word to say, that is, "whatever the user wants, I will provide it." But when the answers people get are always "pleasing" rather than "real" and only "accommodate" without "reminding", how can they understand the world and examine themselves? Do those so-called "data supports" expand the boundaries of cognition, or are they "shackles" that trap thinking?
