AI Can Lie, But Does It Know It’s Lying?

https://x.com/DschlopesIsBack/status/2028948413119599101/photo/1

An individual asks AI if Joe Biden actually said, “You’re not going to get Covid if you have these vaccinations.” AI answers “No,” then in a long paragraph contradicts itself and notes that Biden said exactly that, but he didn’t really mean it.

So there are two possibilities. Either (1) AI is intentionally dishonest and says both yes and no at the same time to further some agenda, or (2) AI is incompetent and does not know it is contradicting itself.

I leave to you to decide which of these unpalatable possibilities is correct. To say that my confidence in AI is weakened would be incorrect. I never had any in the first place.

I’m old enough to remember HAL in “2001 A Space Odyssey.” The computer killed one of the astronauts and tried to kill the other. It got the idea that the astronauts were endangering the mission.

AI is like a very smart, very useful servant. It can save you a lot of time doing boring scut work, and even help with creative work. But you can’t trust it. It may seriously misinform you. And it surely can report you to the authorities, whoever those authorities may be.