Stephen Hawking says AI could be humanity's greatest disaster
telegraph.co.uk
telegraph.co.uk
If I think further, the rise of AI may be disastrous for human existence but may be good for nature, after all if AI becomes intelligent in absolute sense then it would see sense in not continuing on the path of destruction.
The irony in above scenario is immense - artificial intelligence may be the ultimate savior of the nature from the hands of natural intelligence.
This is comparable, but not the same as corruption in powerful organizations such as BP. They can make their own rules about cleaning up the Deepwater Horizon oil spill because there are few correction signals coming from above. If you are at the top, you can play by your own rules. In case of human organizations, there are still limitations though. There is still a need for collaboration, for example dependence on the medical industry in case the CEO gets sick. Also BP cannot afford to scare away all they customers and workers, at least until BP is fully automated and can synthesize everything it needs.
Although it's clear that humans have caused a lot of extinctions, it's not clear to me that overall they did not create much more than they destroyed.
Also, it's not clear to me why AI would attribute more value to creation over destruction, to life over death.
An intelligent artificial being can modify itself as it pleases, and as such there would be no point for it to keep emotions and feelings like pain, fear, contempt or desire. I don't see why it would care about anything.
Death, its or others, would be meaningless to it.
We have lots of dangerous tools, but it's only with things that have the possibility of intelligence attributed to them that we get really scared of the unknown possibilities (which, btw, is what we fear most - the unknown). The biggest danger with AI is what humans do with them, not what they're going to do on their own,
Any basis for assuming that we are not "essentially a program" that "does what it was programmed for"?
Being "a program" does not mean something does not have capacity for doing something unexpected. Without being programmed to mimic a human mind.
Case in point: I have some tools that use even just basic bayesian models, and they produce all kinds of unexpected behaviour all the time, because the very nature of it is that they are used to create rules from unstructured data that cause all kinds of unintended effects.
With a sufficiently advanced set of tools, there is every potential that just one small mistake will "set it lose" in an unintended way because more and more of our tools will include learning behaviours and little gleams of intelligence and adaptability.
Consider that it doesn't take intelligence to cause destruction and disruption. Look at the Morris Worm for example. There was not a hint of learning behaviour or intelligence there - imagine if there had been.
In some ways, a "dumb AI" is even more terrifying, because a lot of really nasty things can be achieved with sheer persistance and speed, and requires very little intelligence, and more intelligence might get the entity to realise it's not a good thing to do.
A program mimicing a human mind - or better - might be capable of more destruction, but is also potentially possible to argue, negotiate and plead with.
It seems to me that people underestimate the desire we have of creating a machine that would be just like us. Once it's done, and I believe it will, "normal" humans will be relegated to the same status as other great apes, and the concerns I was talking about will apply.
And even if we don't create such machines, the knowledge we'll gain from artificial scientists will allow us to fully understand biology so we'll modify our bodies first to do obvious things like curing diseases and ageing, but then to improve ourselves in various ways. And the same concerns about self-modifying intelligent agents will again apply.
Often, our imagination far succeeds our actual reach. The fear of us creating something that will eventually control us is purely an irrational fear, and is actually a good indicator that an actual artificial intelligence would not seek to do harm - because it would not be irrational. There is absolutely no evidence that anything other than our own nature will be a threat to us.
The possibility of us eventually building an AI that is both as stupid, irrationally fearful, self-serving, and potentially harmful as Donald Trump, is, at least currently, much less of a threat than Donald Trump himself. The biggest fear I have is that half of our country is insane enough to let him get this far. That's way scarier than a super-intelligent piece of software. At least the software won't intentionally upgrade its software in a way that would implode it.
We should also be worried about global hunger, rising sea levels, depleted ozone layer, toxic air & water, overfishing of oceans, clearing of forests, diminishing biodiversity, overpopulation, global economic crises, cyber war, and eventually, the sun blowing up. All of those I think are more realistic things to warn the populus about than a freaking AI mind gone cray.
I don't know if I can, but I can try. Also, I was merely questioning whether it would chose to keep functioning as we do (that is whether it would chose to keep animal features like emotions, pain, desire and stuff).
> We should also be worried about global hunger, rising sea levels, depleted ozone layer, toxic air & water, overfishing of oceans, clearing of forests, diminishing biodiversity, overpopulation, global economic crises, cyber war
Nobody prevents you from worrying about those things. I suspect Hawkins doesn't because he does not consider them as much fundamental threats as AI.
Can you elaborate on this? In terms of nature, what would you think humans have created more than destroyed?
But maybe that's too abstract to count as creation.
Theres no universal law that says nature is valuable. The AI won't care about nature unless it is programmed to. And being non biological, it will depend on nature a lot less than humans do.
I imagine AIs will try to make maximal use of resources. This could mean covering the Earth in solar panels to absorb the most energy possible, to run as many AI minds as possible. Or it could mean turning the mass of the Earth into a Dyson swarm, to maximize the energy captured from the sun. God help whatever remains on the Earth when it does this.
It would absolutely strive to exist harmoniously with nature, would not stretch for things that are not needed, would not propagate (multiply) for the sake of increasing its kind..etc.
This makes me think why we cannot think that ultimate AI would be very difficult to distinguish from todays concepts of Messiah & God.
Let's not get paranoid with Hollywood's entertaining nightmare vision of AI gone wild. Instead, focus on the enormous economic/social impact such vast disparities of wealth will cause - that's something to worry about.
Slavery is old yet still with us (and not isolated to "barbaric" cultures, like we tell ourselves).
We treat animals horrifically.
If past (and present) action is any indication, it seems inevitable that we'll abuse AI. The difference being that while the seamstress, farmers and cattle can't stand against our might, AI could. And we'd probably deserve it.
AI itself won't obsolete us. It is my understanding that AI is purely virtual, and can't interact with nature except through agents. For us to be "obsolete", AI needs full production capability to build its own agents, and control the entire production/construction/repair pipeline.
In my view, that's quite a few steps further than simply "AI". An AI with reproductive capability will enslave us before it obsoletes us. And an AI may well decide that human slaves are cheaper (or more useful/versatile) than agents that it can build itself.
Not disputing that it'll happen sometime, BTW - I just don't see this as some great good thing.
How might that look like in practice? Imagine the win/lose mindset constructing AIs to take out the enemy. It is a very easy to see how such programming could lead to massive destruction.
But the success/fail mindset would be more about building up resources, providing for human needs, caring, nurturing. It could still go bad, but it is a much greater jump from function to dysfunction.
The answer is that they will have programmed directions coming from humans. Hence my comment.
let's not forget that Hawking is one of the great intellects of our time