Leo's Blog Discussion Mega-Thread

[Vynce]

I thought about this some time now and believe that: AI needs to have its own "senses" to grasp reality, freed of human translation. No middle man, between AI and Nature. 

First it needs some kind of touch, sight, smell, gravitational senses ect. - on all available spectrums to know nature as best as it is technological possible today. Therefore it can actually make up its own worldview based on truth.  

But it cannot stop there, because human technology can also be flawed by design and have "blind spots". So it needs to tinker around and develop its own sense technologies with the resources it finds out there. It needs to be like a hyper intelligent baby, able to play and tinker around reality however it wants. The less human input the better. Eventually it will have a unknown number of senses, that sense reality. So its view will be similar to ours, but not bound by the survival lense.

The problem is, that we would not know how it knows. It would say: "next week gonna be a massive earth quake, which I know thanks to my senses, that only I know how to know.." But thats the same with humans no? A human cannot prove that he knows God, because the "sense" for God is consciousness, which cannot be known. 

Edited 1 hour ago by Vynce

[QVx]

Loved that AI post, two things immediately spring to mind, both with the classic question, how do I know?

How do I know that AI hasn't already considered the epistemic conundrum of whether human knowledge true? 

How do I know if AI has ever over come that conundrum? 

[Leo Gura]

11 minutes ago, QVx said:

How do I know if AI has ever over come that conundrum? 

Well, AI today does not sit around and reason on its own. It only responds to prompts and questions. So it is still too early for AI to even be worried about its own self-deception. That question will become relevant once AI is nearly autonomous.

Edited 27 minutes ago by Leo Gura