I think the interesting point is that he has a fleet of cars that are all driving around coming across new situations.
These are fed into the big AI that figures out new rules, these rules are sent to the cars to see how they work.
So the AI is learning way faster than any human could, it is learning in parallel. We can only drive on car at a time.
And this is version 1.0 of AI, I do wonder how relevant humans will be when AI is learning everything in parallel . Robots, rovers, the internet, cars, factories, humans, weather stations, satellites, everything hooked up to these AIs who learn and learn and learn.
Just like a parent can watch a child and predict their mood, what they want, what will make them happy, how to control and manipulate them... won't the AIs do the same thing. Sometimes with our blessing. How useful will it be to have an AI keep an eye on paedophiles, monitor them, maybe with tactics to distract them and to engage them in some other way. Society likes that. Now how far will we push that? Will we allow the AI to almost re-program them? To make them better? The AI will understand the human brain, it will understand how to change triggers and behaviours (don't we already have therapists who do this?) ... do we allow this to happen?
Fake news ... wouldn't an AI be able to perform mass experiments on the population to see how to change the popular vote. Say a Trump supporter dislikes muslims, then some gentle persuasion by showing a more pro-muslim news feed .... would the government want to alter this behaviour. We currently try to do this with terrorists, with illegal activity by blocking pro-<insert illegal activity here> websites ... when do we move from blocking a site to actively promoting the other point of view?
An AI that is reading what we read, simulating what we think (are we so unique? Could a population be simulated with a few thousand different personality types?) ... simulate how we respond to certain news stories, what are triggers are.
These are fed into the big AI that figures out new rules, these rules are sent to the cars to see how they work.
So the AI is learning way faster than any human could, it is learning in parallel. We can only drive on car at a time.
And this is version 1.0 of AI, I do wonder how relevant humans will be when AI is learning everything in parallel . Robots, rovers, the internet, cars, factories, humans, weather stations, satellites, everything hooked up to these AIs who learn and learn and learn.
Just like a parent can watch a child and predict their mood, what they want, what will make them happy, how to control and manipulate them... won't the AIs do the same thing. Sometimes with our blessing. How useful will it be to have an AI keep an eye on paedophiles, monitor them, maybe with tactics to distract them and to engage them in some other way. Society likes that. Now how far will we push that? Will we allow the AI to almost re-program them? To make them better? The AI will understand the human brain, it will understand how to change triggers and behaviours (don't we already have therapists who do this?) ... do we allow this to happen?
Fake news ... wouldn't an AI be able to perform mass experiments on the population to see how to change the popular vote. Say a Trump supporter dislikes muslims, then some gentle persuasion by showing a more pro-muslim news feed .... would the government want to alter this behaviour. We currently try to do this with terrorists, with illegal activity by blocking pro-<insert illegal activity here> websites ... when do we move from blocking a site to actively promoting the other point of view?
An AI that is reading what we read, simulating what we think (are we so unique? Could a population be simulated with a few thousand different personality types?) ... simulate how we respond to certain news stories, what are triggers are.