I once said to ChatGPT:
“Did you see? Elon Musk did a Nazi salute.”
ChatGPT’s first reaction: “We have to be careful with defamatory rumors.”
I thought that since the event had just taken place, ChatGPT lacked elements.
I asked the same question again the next day. Time for the press and others to feed the analyses of the gesture.
This time, ChatGPT claimed that the gesture could be interpreted as a Nazi salute.
I was forced to demonstrate to him by a + b that it was indeed a Nazi salute: fingers perfectly aligned and not raised as when we salute + rigid arm and not flexible + gesture very appreciated by the German far-right party and other neo nazi groups.
Which is a shame for “learned” AI and gifted in analysis.
Repeat for Steve Bannon’s Nazi salute: ChatGPT brings me the verse of “interpretation”.
I find this very serious for a technology that is supposed to improve the well-being of humanity.
I think it’s imperative that ChatGPT:
- Systematically applies a factual and rigorous approach to these topics.
- Leaves no room for ambiguity or relativization, by relying on factual criteria.
- Is programmed to avoid the trivialization strategies of dangerous ideologies.
Artificial intelligence has a major role to play in the fight against disinformation and the normalization of dangerous ideologies. If it fails on this point, it becomes an unintentional vector of trivialization.
I would like to know how OpenAI plans to improve this point and if other users have also noticed this weakness.
Thank you in advance for your feedback and your commitment on this crucial subject.