Currently there is a disagreement going on between two of the inhabitants of the upper echelons of the modern tech world, Mark Zuckerberg and Elon Musk about the dangers of artificial intelligence. (I am going to assume everyone reading this knows who both of these men are).
Elon Musk has been strident about the existential dangers of AI to humanity. Zuckerberg is an optimist about AI. As a result there has been a bit of a feud, played out as these things are these days – on twitter.
Elon’s latest salvo is below.
I’ve talked to Mark about this. His understanding of the subject is limited.
— Elon Musk (@elonmusk) July 25, 2017
Of course the question is, who is right? Well actually, I don’t have to think very long about this at all and say straight up Elon Musk is right, no questions asked.
But why can I be that unequivocal like this? It’s simple really. Imagine an AI that got smarter than humans. Now assume that it is a “good” AI. It’s a good point to start from. By good I naturally mean that it actions towards humans are benevolent.
Now the thing is, humans change their minds all the the time. So an AI that is smarter than humans could simply change it’s mind about being nice to humans.
I’m not saying that this would happen, but what I am saying is that whether the AI at that point chooses to be benevolent or malevolent is at that point beyond the control of humans. It is by definition not possible to control something that is smarter than you. It gets to choose how it acts towards you.
The problem I see is that it is almost inevitable that AI will become smarter than us. And that is the point at which we have lost control. Not only that, we have NO CAPACITY in circumstances like this to regain control. Once the genie is out of the bottle, it won’t be going back in.
We have seen situations like this in fiction, as far back as 1968 in the film 2001 A Space Odyssey with the computer HAL, as he says to the main protagonist Dave:
“HAL: I know that you and Frank were planning to disconnect me, and I’m afraid that’s something I cannot allow to happen. “
More recently I have watched the TV series “Person Of Interest” whose central theme is two warring AIs, both of whom are so far beyond the intelligence of humans that humans become pawns in their game.
But that’s science fiction right? Not relevant to reality?
Well the cap is off the genie bottle and that genie smoke is swirling up the bottle’s neck. Recently, Google reported that it’s AI, Deepmind, has learned to become highly aggressive in stressful situations. Let me expand on that. What the researchers showed that the AI resorted to sabotage, greed and aggression to achieve their goals when needed.
When you think about it, shouldn’t any system programmed by humans end up with the same foibles as humans?
But what tempers the darker nature of humans is a little thing called empathy right? Can’t we just program empathy into the machines?
To that question I reply how? Empathy is an understanding of the suffering of others based on knowing what it feels like to feel pain ourselves. By that definition, it is impossible to teach something that can’t feel pain empathy.
So as a result, we end up with an entity that can be imbued with all of the darker aspects of human nature, but can’t be imbued with the key limiter.
In conclusion, I’m pretty much saying that if global warming doesn’t get us, AI will. Anyway, have a nice day!