A rather excellent video of the coming 3 years of AI. It may or may not be accurate, it doesn't matter.
What does matter is the very precise description of exponentials and therefore how ASI will emerge, necessarily. Remember that a water lily doubling every day only occupies 0.8% of a pond a week before covering 100%. So the level you start from is irrelevant. The only important factor is the growth rate.
As for "alignment" and other control mechanisms, as explained, they too are irrelevant. You cannot, under any circumstances control something which is more intelligent than you are. You will necessarily be outsmarted. And unlike what is explained in the video, we will not "discover" that the AI is being untrustworthy. We won't "see" or rather understand anything. Imagine human intelligence as a 360 degree circle capable of creating links or relationships between ideas positioned on the circle. Well, the AI is a 360 degree sphere. It can generate concepts we simply do not have access to and bypass whatever we can "think".
Likewise, there is a fundamental misunderstanding of the forces and principles of evolution at play which generate competition and an arm race guaranteeing the result. Now that super intelligence is in sight, it will necessarily emerge. There is no turning back. It will be American, Chinese, Russian or Indian, but it will emerge. 2027, 2028 or 2029 makes absolutely no difference. ASI is here!
No comments:
Post a Comment