Anthropier CEO Dario Amani Stir up On Monday after suggesting that advanced artificial intelligence models may one day be provided with the ability to press a “button” to finish tasks that they may find unpleasant. Amodei made provocative statements During an interview In the Council of Foreign Relations, in recognition that the idea “looks crazy”.
“This – this is one of those topics that will make me look completely crazy,” Amani said during the interview. “I think we should think at least about an issue, if we build these systems and protect all kinds of things like humans as well as humans, and it seems that we have many cognitive capabilities themselves, if the Antichrist is like a duck and goes like a duck, and it may be a duck.”
Amodei’s comments came in response to the question of an audience from the Carmem Domingues of the late 2014 in late 2014 from the Kyle Fish intelligence researcher, “To see, you know, stage or lack of models of artificial intelligence in the future, and whether it may deserve moral attention and protection in the future.” Fish is currently looking at the controversial theme on whether artificial intelligence models can have a sense or deserve moral consideration in another way.
“So, there is something that we think about starting publishing is, as you know, when we publish our models in their publishing environments, just give the model a button that says,” I left this task, “the model can press it, right?” Amodei said. “It is just a very basic frame, as you know, the preference framework, where it says whether it is assumed that the model had experience and that he hated the job enough, which gives him the ability to press the button,“ I left this task. ”If you find models that press this button a lot of unpleasant things, you know, you must do that – this does not mean that you are convinced – but you may have some interest in it.