AI does in fact have to receive instruction up front, but then it is expected to continue to enhance its own learning at some point. I have never listen to a lecture or read an article that outlined AI being able to start from a blank canvas. As for the continued collection of information required to continue learning, sensors, cameras, mics, a robot in every room, will all be data collection points for something that is able to continually learn.
Still, I am not gulping the cool-aid. The human brain includes more than the four senses and memory when reasoning - it also includes emotion, and for some, faith. The latter two are in my opinion, critical omissions when the elites preach about AI surpassing the human brain by 2025. Now of course I suppose we could require the world population to be implanted with sensors for emotion at birth to at least eliminate one part of the equation, but by then we may as well just lest AI eliminate humans.
The elites acknowledge that a human problem (possibly catastrophic in nature) could happen if AI is able to outperform humans, but that is kind of where it ends. The potential reality that AI could displace millions of jobs, collapsing both the supply and demand side of economies, is not something that to them, is more important than progress in the field.
AI may be solving logic problems in my lifetime, but I do not see it solving human problems. And even if it did, us squishies are likely not going to listen. After all, the solution would probably be racist or something like that.