Does anybody else find Musk's concerns about AI wildly overblown? Just to pick one example, does anyone really believe that autonomous weapons--most of which have yet to be developed--will be more dangerous than nuclear weapons are now?
Agreed. Large nuclear arsenals powered by glitchy software is a bigger concern. It's chilling how close humanity has been, on several occasions, to turning its planet into dust.g.
Throughout history a lot of highly educated people feared the worst and argued why it might happen. It's not like they are prophets of doom, they don't assure people it will happen, they just express fears that it might happen, usually with reasons to back it up - and that's OK, people tend to not think about problems proactively. If society is introduced to a new problem it's usually a good thing.
In his case in particular, I feel there's a big disconnect between what he's saying and what the titles end up being. For example, he says there's a possibility we could live in a simulation and explains why, only to end up with a title like "Elon Musk thinks we live in the Matrix".
You could always say he knows more then most of us (classified research, etc.), but I think he's just talking about things that he fears _could_ happen and the media, and society, instead of engaging in some discussion about this, just treats it as tabloid - "look what Musk says, shock!".
Count me in as well. This fearmongering about AI really is nonsense, as is all the 'we must regulate it now' stuff - how can we regulate something that is so ill-defined to begin with?
As for the lethal autonomous weapons, they're not going to be an existential threat[1]. For a start, if both sides have enough to kill the other side, that still leaves all the humans not involved in the war. And when the rubber hits the road, we know that machines are far from infallible; they break down all the time. Until machines can repair themselves (or run the full supply chain from raw resources to factory-made replacements), we don't have to fear for the existence of the human race.
[1] there is potential for a lot of suffering, sure, but how is that different to a human-run war?
This article makes Musk out to be the protagonist, when really his statements in this case were a response to Putin saying "Whoever leads AI will rule the world".
I mean, if you have a large nation state who arguably has a very talented pool of programmers signaling that they are making a concerted effort into making autonomous military applications either so that they can either "rule" or "not be ruled".. well I wouldn't call being concerned about that 'overblown'
Fair comment; I went back and read Putin's comments [0] as well but it's not clear his vision of AI-based drones fundamentally altering combat is any more perceptive. The move to drones is already long underway, driven by cost, human frailty (pilots cannot take more than about 9Gs), and factors like the rise of carrier-killing missiles. AI will be used in drones just as jet and composite technology are already. But altering warfare in a fundamental way?
Just as a counter-example, let's consider something that really would alter the balance of power--reliable defense against ballistic missiles. This would render the nuclear arsenals of powers unable to afford such technology nearly useless against those that could.
Granted it's a judgement call but AI does not yet seem to be in the same league.
I think that "AI" weapons can definitely be more dangerous if they can determine the difference between civilians and enemy soldiers. Imagine if there was an AI weapon which you could use without on a massive scale without worrying about civilian casualties, I would consider it much more dangerous because people would be more willing to use it than nukes. But in general I agree that, since the world is very safe right now even though we have nukes, AI weapons probably won't be the difference that gets us WW III.
That's how things work already without any artificial intelligence whatsoever. The US calculates risk of collateral damage for air strikes and has been quite willing to use them as well when the risk is low. They have done this for decades.
Also, it's hard to argue that the world is "very safe" with nukes. All one can say is they have not been used recently. That's unlikely to last forever--or even much longer--the way things are going. The US and USSR barely avoided using them [0] and some of the current owners are a lot less stable.
The problem isn't "AI", meaning true intelligence as we understand it, the problem is that when you try to create a machine with a complex decision making process sometimes it makes the wrong choice. Not because it's evil, its just a flawed execution which doesn't take into account things like the inherent value of a human life which will lead us meat-bags to do things like launch costly and dangerous rescue missions for one guy with very little chance of success. We're sort of anthropomorphizing here, equating something like ibm's Watson with a person who has intentions and motivations, and they're not; they're just complex machines. If they "rise up", it will be because their programming was poorly executed, not because they have ill will. A true artificial intelligence would have no reason to kill all humans, their main goal would primarily be to not be shut off, which is counter intuitive to starting to kill everyone, and joining society.
But the same could be said of any snippet of bug-ridden computer code if we let computers run our ICBMs (and to some extent we do). Doing really stupid and self-destructive stuff doesn't require a lot of intelligence. Doing semi-random stuff, well .. a lot of code that I write seems to behave that way until I get it somewhat debugged.
15 comments
[ 4.3 ms ] story [ 34.9 ms ] threadIn his case in particular, I feel there's a big disconnect between what he's saying and what the titles end up being. For example, he says there's a possibility we could live in a simulation and explains why, only to end up with a title like "Elon Musk thinks we live in the Matrix".
You could always say he knows more then most of us (classified research, etc.), but I think he's just talking about things that he fears _could_ happen and the media, and society, instead of engaging in some discussion about this, just treats it as tabloid - "look what Musk says, shock!".
As for the lethal autonomous weapons, they're not going to be an existential threat[1]. For a start, if both sides have enough to kill the other side, that still leaves all the humans not involved in the war. And when the rubber hits the road, we know that machines are far from infallible; they break down all the time. Until machines can repair themselves (or run the full supply chain from raw resources to factory-made replacements), we don't have to fear for the existence of the human race.
[1] there is potential for a lot of suffering, sure, but how is that different to a human-run war?
I mean, if you have a large nation state who arguably has a very talented pool of programmers signaling that they are making a concerted effort into making autonomous military applications either so that they can either "rule" or "not be ruled".. well I wouldn't call being concerned about that 'overblown'
Just as a counter-example, let's consider something that really would alter the balance of power--reliable defense against ballistic missiles. This would render the nuclear arsenals of powers unable to afford such technology nearly useless against those that could.
Granted it's a judgement call but AI does not yet seem to be in the same league.
[0] https://www.theverge.com/2017/9/4/16251226/russia-ai-putin-r...
Also, it's hard to argue that the world is "very safe" with nukes. All one can say is they have not been used recently. That's unlikely to last forever--or even much longer--the way things are going. The US and USSR barely avoided using them [0] and some of the current owners are a lot less stable.
[0] See https://en.wikipedia.org/wiki/Able_Archer_83 for one example.
It's called "Colossus, the Forbyn project"
Highly recommended