Why We Aren’t Getting Any Better At AI Alignment?

AI alignment is a topic of great interest to me. We’ve made a lot of progress in AI over the last few years, but not a lot in AI safety and alignment. It is not like there has been no work in this domain, but it is minuscule compared to how fast the rest of the field is growing. To understand what exactly is happening in AI, we need to zoom out a little bit and look at the field from afar to understand the underpinnings of the current developments. So, without further ado, let’s begin from the very start.

Specific to General AI

AI is not a new field and has been around since the 1960s, but it looked completely different back then. It is only with the release of ChatGPT it reached the laymen. Before it was a cool thing only for a bunch of researchers and a few companies. But now it is everywhere, even in the places we don’t even think about. It has truly reached the entire globe and has become much more interconnected just like the internet.

Early AI was operating in silos, now we are slowly and surely moving towards more of an interconnected global system of intelligence. And to be honest, this is not very different than how human intelligence evolved, in a community.

Early attempts at AI tried to create knowledge systems that might be able to ingest the entire world’s knowledge with sufficient memory. We worked a lot on symbolic systems to create world knowledge. But this experiment failed quickly and we started working towards…