Adversarial Approach to AI Acceleration — Part 1

Every great accomplishment must have a worthy adversary to defeat.

I strongly believe in the statement I made above and this blog will be based on the same premise. This one’s gonna be super short and introspective mostly because I am whipping up this idea and my thoughts the moment the epiphany(?) occurred to me.

As we look at the insane progress and the speed of development of AI today, even as an AI researcher, I am baffled sometimes. Extrapolating the current progress to the lofty idea of AGI/ASI, the inception of it does not seem impossible but rather, very close.

There’s a high chance I might be wrong but I think there’s a way to approach this idea in a superficial sense (currently).

Bare with me while I dabble in a meta plane of conversation & consider the question — What great ideological accomplishment didn’t have its own adversary to defeat? (I certainly can’t think of one, if you do please comment down below). Was the ideology great not because of the adversary it defeated? Applying this chain-of-thought to accelerating AI makes sense to me. On the road to developing AGI, there must be an adversary which constantly peels it down & challenges the “AI” system beyond it’s perceived capabilities. Defeat that, the adversary becomes better and poses a new pressing challenge & so on until…

The genesis of AGI.

Now what would this adversary look like? How would it be capable to challenge an advanced AI? Is one of the results of a worth adversary better alignment of AGI? Much nuances to be worked out but hey, this is just part 1.

I will keep this chain of blogs open until I find something resembling an answer. Please feel free to contribute in the comments.

Thanks for reading.