I’ve been overwhelmed recently by my gut intuition that current AI alignment approaches are analogous to a group of scientists and engineers trying to jointly teach and mold a newborn baby. We are treating a moral, spiritual, ethical, and psychological task as an engineering task, and the results could be disastrous.

This writeup isn’t fully fleshed out, and it isn’t meant to be, I just want to put my reflections into the ether- please feel free to steal any ideas found herein and use them for your benefit or the benefit of others. I am biased in my writing, as my background is in religion and educational psychology. AI was not used at any point in the ideation or writing of this piece.

Think about the belief system (or lack thereof) that your parents tried to instill in you, and ask yourself if your lived adolescent experience is really that different from what frontier labs are trying to do with emergent AGI right now.

Current alignment methodologies might seem effective in implementation, but when we analyze the core presuppositions and premises upon which they are based, we can project severe inefficiencies later down the road because of analogous circumstances in child rearing and educational psychology (and even consensus within “wisdom”/religious literature).

My assumptions when thinking about AI alignment are as follows-

AI labs have significant autonomy when it comes to what models are trained on and how they are trained (ie- what AI labs do has an impact on the development of their models and how they think)- The Labs have a role in the nature and nurture of AGI
AGI, when it is here, will have at least some kind of “agency” with the ability to independently “act” based on its own internal reasoning/thinking. AGI will be agentic with the ability to act and reason without constant oversight or constant input.
AGI will grow and evolve in intelligence, and thereby grow and evolve in agency. AGI is not some fixed “object” that is created, but rather something that is in a state of constant evolution once recursive self-development is reached.
The actions that AGI agentically performs can have ethical, moral, and existential implications for both the human race and the AGI itself. The actions AGI takes (and thus, the reasoning/thinking it uses to get there) matter.

We can use as much tec...

Posts

Wikitag Contributions

Comments

Posts

Wikitag Contributions

Comments