Are we willing to have forgetful machines (AI)?
When we think of machines, we immediately bring determinism to mind. Machines are cruel, they’re full of 1s and 0s. There is nothing in between. No emotion, no variability, no grey areas. They are governed by facts, and they can store a vast array of them; many, many times more expansive than what the human mind can deal with. That is, with existing machines.
The thing is, when we’ll try to transition to true AI, this model will have to change drastically, due to a fundamental paradigm shift: True AI is foremost about true learning (and by this I mean autonomous learning, without cheating by injecting predigested human knowledge). And true learning necessitates forgetfulness! The modus operandi of the human mind, “if you don’t use it you lose it”, is quintessential for advancing our learning and comprehension.
You almost never learn something first time right. You need to experience several similar circumstances and have repeated attempts, at different angles, to build solid foundation of how you can tackle with a situation and all similar to it. That’s why you need to forget in the meantime, every time. Parts of the approach that worked every time will remain in your working knowledge, and you’ll reuse them. Parts that worked once but not many will be forgotten and leave room for other, novel approaches that may prove more successful. In the end, when you’ve build a solid foundation and learned this thing, it will be because you kept the best of everytime and laid the groundwork that can handle more or less successfully every upcoming iteration. You allowed through forgetfulness to “prune” the tree of knowledge, keep only the strong branches, and thus give the opportunity for growth to new branches, which only if they are proven equally strong will be maintained. Forgetfulness is essential thus. Else your learning progress would quickly become deadlocked, because in the first “wrong turn” you would reach a dead-end and would be unable to backtrack and start afresh.
The reason I elaborated so much on this is because when we’ll finally be ready to built true AI, we’ll need to implement the same mechanism of forgetfulness (alas, we know no better) and the result will be… robots… that will be unlike everything we’ve been imagining all the past decades. So far from determinism, but also so familiar, because they will be almost human in some ways. And the question is, are we ready to accept such an invention? Is it possible that humanity will be underwhelmed, because it was expecting something almost superhuman? Will the potential usage scenarios that we’re building in our fantasies & science fiction just collapse and have to be rebuilt from scratch?
Probably not, but it’s an interesting question to ponder.