Artificial intelligence is not your Mirror on the wall 🪞.
I have two friends - both have a lot of time, and are curious. They use AI, in various forms, to help them with with daily tasks.
Both have recently reached out, sharing that "hey, look, even AI tells me I'm right!"
I dug into their experience, and found out something frightening. Maybe AI is right when it tells them that they are right. But I would take that statement with lots of precautions:
because of its inner functionings, AI is a statistical tool. It is going to reproduce something seen otherwise. And it will continue the sentence h'it has been provided as "starting token".
Here I am pointing the fact that
- laying out premises (any factual premise issued from human textual exchanges)
- Then asking the AI "am I right?"
Will, with high probability, produce a positive answer.
In the same way, AI is great at spotting words transmitting sentiments (do you remember all those "sentiment analysis" APIs about ten years ago?)
My best advices about using objectively an AI:
- Keep yourself out of the game (consider yourself one of the parties, not the party)
- ask for common practices involving the parties in such a situation
- Double check what the AI told you, systematically
- Use it as a tool to expand your knowledge, not your confident
- Don't believe to companies who promise to provide an AI for law advice
AI is a tool, it's not another human being. OpenAI real time voice assistant is kind of great at chatting and expressing emotions, don't fall for it.
(Written on January, 12th 2025)