On AI, consciousness and intent


There is a lot of discussion about AI becoming conscious. It's not 100% clear what the various researchers and pundits mean by that; I would say the most important aspect of consciousness in the context of AI is the notion of intent. This means that the AI becomes able to develop its own goals and then optimise itself -- and the world around it, because it is so intelligent and powerful -- to reach these goals. This leads to the challenge of AI alignment (can we align the AI's goals with ours?) and the risks associated with the various falours of the paperclip maximiser narrative.


Personally, I don't think we will get there anytime soon (by which I don't want to imply that research towards AI alignment doesn't make sense). However: I don't think that AI needs to be conscious -- intentful -- for it to potentially become a serious problem fo people and societies. Because in some sense AIs will borrow the intents of humans. And these can be very problematic. This perspective became clear to me over the last couple of months, listening to and reading various experts (of which I am not one). Let me explain by giving a bunch of examples.


The recommender algorithms used by all the well-known social networks basically optimize for engagement. And engagement can be increased by suggesting content that is sensationalising, radicalising, and polarising. This drive towards "bad" content has clearly had (and still has) negative consequences for societal cohesion, democratic deliberation and lots of peoples' psychological health. These algorithms don't have intent. They didn't optimizes themselves towards suggesting these kinds of contents. They "made" their human overlords do that, because the humans wanted to optimise profit. Same thing with the LLMs these days. Google felt pressured by market and ego forces to publish their LLM because OpenAI published theirs despite serious doubts about its readiness, inside Google and outside. The problem also occurs on an intentional level. The US will never decide to slow down AI deployment because "if we slow down, China will do it anyway". Likely true.


So, to summarise: even an AI that does not have its own intent will probably lead to lots of risks for our societies because the intents of humans are likely to drive it in problematic directions. Powerful AI, even unconscious and without intent, can be quite problematic at scale.


Two caveats. While I emphasise the negative consequences of AI, it will also have lots of positive ones. I have not decided whether I think the benefits will outweigh the the risks, or not.


Second, this dynamic is not restricted to AI. There are lots of downward spirals driven by market and ego forces that also have bad consequences for society, for example the (on average) decreasing quality of journalism. However, AI makes everything so much faster and larger in scale that it might become a much bigger problem than low-quality journalism.


And a final note: I do not think that this is capitalism's fault. I think these dynamics are deeply in grained in humans -- which is why I always wrote "market and ego forces".


Anyway. What do you think?