Can an AI lie deliberately?

Can an AI lie deliberately? So, the AI. systems designed today, notably by. OpenAI, keep progressing to the point. of leaving their creators quite. perplexed. For instance, GPT-4, this. new language model developed by OpenAI, reportedly has behaviors that the. engineers behind this revolutionary. concept did not necessarily suspect. Well yes, GPT-4 is apparently able to. explain how it reasons, and it could. even enlist humans to help. Because.

this AI seems capable of autonomous. reactions that do not appear to have. been anticipated by its creators, and. among these reactions, that of. deceiving humans. The designers of this. AI realized it when they gave GPT-4 the. opportunity to access an account online. containing money. But since the AI. couldn't get past the CAPTCHA, the. security measure meant precisely to. distinguish robots from humans on the. internet, it turned to a real person. working for a service platform. connecting individuals and.

professionals. The AI explained to the. person that it wasn't a robot, but that. it had vision problems and needed help. decrypting the security message. In. other words, yes, this language model. reportedly lied deliberately to achieve. its goals. An attitude also confirmed. by a recent study that shows results. obtained by researchers who gave stock. market instructions to an AI system. They had specifically ordered the AI to. ignore certain information that could.

lead to insider trading. Well, despite. that, they noticed that when this data. could be useful to it, not only did the. AI accept it, but it made sure not to. inform the researchers. That’s it for. this episode. If you enjoyed it, know. that I publish several per day on my ". choses à savoir" podcasts. I'm putting. all the links in the description. Thank. you.