hese days, founders of big AI companies are talking about limiting AI's growth rate. Not necessarily by suspending the training process in its entirety, but rather slowing down the process a bit. Slowing down means slowing down the rate at which frontier AI labs are making models more capable.
Anyway, we are slowing down something that is, sooner or later, inevitable: progress. By definition progress is: Development, advancement, or improvement, as toward a goal. But since we don't know what the goal is, we are just optimizing things for progress as if it were a goal in itself.
We grope in the dark, knowing that we seek light, even if we don't know what light actually is. Could be everything but darkness.
So if we are just seeking progress, why slow down?
Well, obviously we want to avoid falling into a hole, or stumbling over any obstacle hidden by the darkness, and avoiding any dangers could be a smart move in such an uncertain environment. And it's a good move for someone unable to move properly in the dark, but I don't think it is the actual solution to the darkness problem. Let's take opportunity cost as a framework to understand why moving slowly in this phase of humanity isn't a good choice for me.
What are we giving up? What is the stake of going fast, and running as fast as possible?
What we get in exchange for pacing the race is only the illusion of controlling the present, by putting limitations in place, without knowing where this will eventually bring us. We tend to overestimate our abilities to predict the future, but this one is made by people, and as you know, people are the most unpredictable thing in the universe, and as Kenneth Stanley and Joel Lehman point out in "Why Greatness Cannot Be Planned", objectives are deceptive to the initial goal and, citing: "[...] objectives mean sailing to a distant destination with an unknown path while novelty requires only steering away from where we've been already."
So I suggest we embrace the novelty, and adapt to it. Let's surf on the entropy wave.
Of course, no one wants technology to be a menace, but if we want AI to not be a menace for the world, the best option is to make it as strong as possible, and open source it at a level where everybody can use it. As a second Amendment, but for AI. Otherwise we could end up leaving the technological sovereignty too much centralized, and so unstable.
I want to close this meditation with a question.
What if we will always be in the dark, as a species, and light is just an illusion? I just hope we won't go gentle into that good night.