Part 9/11:
He remains optimistic that these problems are solvable, describing them as inference problems—challenges that, with time and effort, humanity can overcome. The main danger lies in our tendency to seek quick solutions: trying to enslave or control AI as if it were a wish-fulfillment tool rather than fostering genuine understanding and alignment.
A New Approach to AI Alignment: Seeing as a Whole
The crux of Shear’s argument is that true alignment requires us to see AI and other agents as integrated parts of a larger whole. Instead of trying to impose abstract moral values externally, we should aim to deepen our understanding of their experience, their boundaries, and their sense of self.