We need to teach AI the essence of being human to reduce the risk of misalignment

About this article

One part of the alignment problem is that AI does not genuinely understand what it's like to live in the world, even though it can describe it so accurately. If it doesn't understand human life, why protect or respect it? A chat model's answer to what it's like to be human is pretty unconvincing and unemotional. But, if you tell it what's it's like to live, at a more personal level ... Feeling intense pain, fearing it will never end. The joy and reward of doing something that helps others, ho...