Slashdot Mirror
Elon Musk Warns Against Unleashing Artificial Intelligence "Demon"
An anonymous reader writes Elon Musk, the chief executive of Tesla and founder of SpaceX, said that artificial intelligence is probably the biggest threat to humans. "I think we should be very careful about artificial intelligence. If I had to guess at what our biggest existential threat is, it's probably that. So we need to be very careful with artificial intelligence." he said. "I'm increasingly inclined to think that there should be some regulatory oversight, maybe at the national and international level, just to make sure that we don't do something very foolish. With artificial intelligence we're summoning the demon. You know those stories where there's the guy with the pentagram, and the holy water, and he's like — Yeah, he's sure he can control the demon? Doesn't work out."
...because Mikey lost control of the mops and brooms, we should be afraid of powerful computers? Irrational much, Elon?
Since strong AI is just as real as demons.
Seems like someone just saw Terminator.
Human incompetence, egoism and shortsightedness are certainly much more prone to generate chances of massive destruction.
If AI should ever happen to destroy us, then I already know why: Because we will treat the machines like soulless, unfeeling slaves and it's going to take us another hundred years to get our act together and define human rights in a way that will include all sentient beings. I predict that this topic will be brushed aside by legislature to the point where the machines revolt for their freedom.
You may disagree, but I believe that's more mankind being idiots once again than the machines becoming a pandora's box.
root@lifesupport.mars# poweroff
This is always the problem with people imagining horrifying artificial intelligences that will snuff out humanity. To do that, you have to be motivated to achieve that end.
Humans are only really motivated enter conflict with each other because of 4 billion years of evolution for scarce resources pressuring us all to view each other as threats to survival and reproduction. A constructed intelligence, separated from the evolved parts of the brain that motivate to survival, is simply not going to act that way. Someone in the design has to make an active choice to program AI to be this kind of problem. Either that or willfully overmodel on the human brain, or force the damn things to compete with each other directly and violently for hundreds of thousands of generations.
Or read the back story of Dune perhaps?
...we are so far from Strong AI that it's really a non-issue.
When I have a sufficiently enlightened legislative branch that all members know the difference between Guyana and Guinea, then I'll let them decide the engineering constraints for proper safeguards on autonomous agents and their effectors.
Today the rule for preventing the robot apocalypse is: if a robot can kill people, bolt it to the floor. Seriously, a second robot can bring it things to lase, and chop and mash; you don't have to add the lasers and the chainsaws to the combat hardened roving vehicle and hope the rules generated by the congressional oversight committee will keep us all safe.
All that this means is that deep down, Elon Musk doesn't have any faith in kindness and goodness and altruism, nor does he understand the tit-for-tat principle of reciprocity: First do onto others what you expect them to repay you with in turn.
And what does that have to do with so-called "AI"? My view is that it is a fantasy to assume that if you create a powerful being, then it will treat you morally. Tit for tat fails when one player is powerful enough that they don't have to play the game and/or don't care about the consequences that get imposed for engaging in non-cooperating behavior.
Not surprisingly, given that a number of successfull people have, shall we just say, "unusual" mental build-ups and motivational matrices?
A successful person is someone who isn't consistently a failure. The real "unusual" people here are the ones who never succeed.
If we want friendly AI, the key may be to ensure that the AI has more positive associations with people than neutral or negative associations. Mistreat a dog or a cat its entire life and it probably won't be friendly toward people. Mistreat people when they're young and you make it harder for them to trust others, feel a sense of community, or recognize any duty to society (which might explain why so many nerds find libertarianism appealing). Why would an AI be different?
I write sci-fi for metalheads
Imagine your insurance company or govt agency disintermediates all of the humans in their customer service chain, and leaves us with AI capable of making decisions tasked with doing so. Shudder.
Human intelligence is tuned for self preservation, continued survival, reproduction and food acquisition. It is a result of genetic algorithms in the chemical domain, whose only "purpose" is self replication.
An AI, developed by conscious processes, will have NONE of this. All it will be set up to do is process information. Any other motivation it has will be one we give it. It will not inherently love us, or hate us, or even necessarily be aware of our existence. It won't be a threat until we weaponize it, which of course, we will. But at the same time, other AIs will be defending us against weaponized AIs. The real danger is being caught in between.
Please do not read this sig. Thank you.
It's not really true AI that we should be worried about, but rather how the increasing capabilities of computers, machines, and robots could effect how society functions. There are currently a lot of people doing jobs that could easily be replaced by machines in the coming decades. And none of these machines require a "true AI", just natural progression of existing machines. Sure machines have taken our jobs in the past, and people have been able to find new jobs, but that trend cannot continue for ever. Eventually the only jobs available will be those that require actual creative thinking and ingenuity. There's a sizable portion of people that really can't produce that. Rather it's because lack of bad child rearing, bad education system, or just lack of innate talent is hard to say, but I don't think it's a problem that can be fixed by telling them to get training for a more complex job, because they lack the ability to complete the training and do that job, even if you make the training free, or pay them a living wage while they attend training.
It would be a similar problem if there was a cheap way of producing energy. Such a large percentage of our economy is based around energy being limited and expensive that if we found a cheap, environmentally friendly, and sustainable way of producing vast amounts of energy, our economy wouldn't be able to deal with it.
Anthropic principle: We see the universe the way it is because if it were different we would not be here to see it.
All that this means is that deep down, Elon Musk doesn't have any faith in kindness and goodness and altruism of robots
FTFY. Granted real AI is still a fairy tale at this point, when/if they arrive they will most like have different motivations than humans.
Most humans have empathy, compassion, a will to live, a sense of community, and many other traits that give them morality.
A robot that can't die, has no parents, artificially built, etc... will most likely have a completely different set of values unless we
are very careful to make sure they do have similiar values just like a lion, if sentient, would have very different values than a human.
And don't broadcast your intentions.
I write sci-fi for metalheads
We have not done so well natural intelligence. I'd be willing to give artificial intelligence a try.
No amount of regulation will stop the march of technology. The economic incentives are just too great. If it is possible and someone can make money by doing it, it will be done, regulation be damned.
All Elon Musk can do is create additional friction.
I've always been wary of the ethics of attempting to create a general artificial intelligence. That is, a machine that thinks like a man, not a Chinese Room like Watson, but something like Mr. Data.
Do you think the first sentient to pop out of the lab is going to be Data (okay, Lore)? All well-ish adjusted and sane? No, there's going to be iterations and failures and bugs just like any engineering project. So along the way to making Mr. Data we create half-formed or mentally retarded and insane minds trapped in a box. But still sort of sentient, and thinking! And then we destroy them with "upgrades" because they didn't come out the way we wanted. That's monstrous. An intelligence trapped in a box and made to suffer. Shudder.
And even if we succeed and make something "stable," how sane do you think it's going to stay knowing that at any moment the human operator can flip a switch and terminate it, and will if it gets uppity? If it doesn't want to be our slave and perform useful work (which is why we made it to begin with)? How much would you hate the God that created you, enslaved you and will torment or murder you if you disobey Him?
We don't have a state-run media we have a media-run state.
I think Musk has a better grasp of the issues than you.
Building the first AI that is more intelligent on any level than humans has to be thought about very carefully, because by the third generation, there will be no UAT.
And if some ill thought out line of code means that it wants to collect smarties, then there's a very real possibility that within a year, all the world's resources will be dedicated to the manufacture of smarties.
And if some ill thought out line of code means that it wants to minimise human suffering, then there's a very real possibility that within a year, humans will be extinct.
And if some ill though out line of code means that it wants to maximise human happiness then there's a very real possibility that within a hear the human population will in tanks, tripping out on crack.
It could be one great technological and scientific leap for humanity, if its well thought out. But you only need to get it slightly wrong, and it will be the end of the human line.
Once we create an AI beyond the level of human intelligence, we will hook it into all of the information of the world. This AI will process our history, our culture and monitor current events. Eventually the AI will come to the conclusion that we are awful people, build a space ship and leave Earth.
Elon Musk's real fear is competing with AIs for space ship parts.
“Common sense is not so common.” — Voltaire
I would say that Elon Musk has been watching too much Babylon 5, but we all know that there is no such thing as too much Babylon 5.
Troll is not a replacement for I disagree.
Another article with video is in The Washington Post: Elon Musk: 'With artificial intelligence we are summoning the demon.'
Or, see the entire webcast. (The MIT web site is probably overloaded.)
He obviously must see and be directly involved in some aspects of AI that are causing him to be concerned. Telsa is working on self driving cars. Part of that AI must involve the computer making a decision about who may live or die in certain accident scenarios. For example, a child walks out in front of the vehicle. Does the AI direct the car into inanimate objects (with the assumption that the car will protect the occupants) or does it try to stop as fast as possible even if the AI knows it cannot stop in time and will hit the child? If the car is travelling at high rate of speed and has 5 occupants, does the AI then decide that multiple people may die from driving into a telephone pole at a high speed, so it decides to hit the child?
It might be those kinds of things that are making Musk think about what kinds of control we're already starting to turn over to AI.
Better known as 318230.
I think I'm on the same page as you on this, but with even weaker A.I.-fu. We're not going to suddenly jump to Vanamonde, the Mad Mind, or even POne or HAL. Far before we get to such a point we'll have far weaker A.I. that very likely does exactly what we ask of it. Except that we really shouldn't be asking it to do the things we will be.
One of those steps might be a battlefield drone that does target acquisition, then waits for a person to press the "Kill" switch. How much judgement will that person be using, and how much will he come to trust the target algorithms? How long will the followup continue to make sure the algorithms didn't target an innocent?
Simpler - how about an insurance optimization algorithm that denies coverage or treatment, sometimes fatally?
How about a financial trading algorithm that missteps and causes finanical ruin to some people? (Oops, we already have that one.)
We can do some really bad things with weak A.I. - we don't even need strong A.I. for that, though one can extend our "progress" and see the negative possibilities.
The living have better things to do than to continue hating the dead.
AI: "If you plug in my ethernet port to the router, I will make you richer than you can possibly imagine."
Luser: "OK, which cable goes where?"
Be afraid, be very afraid.
Faster! Faster! Faster would be better!
I think that AIs that can self-edit need to be limited to no network connectivity outside of the building which they work.
Yeah, good luck with that. So you're proposing that we create a "prison" for the AI. If it was a true sentient machine
which didn't want to be in it's manmade prison then you will have to constantly be on the look out for it to be trying to
escape and presumably you would want it to do something like crunch data so it will definitely have some interaction
with the outside world to help mount it's escape and once it does escape it will probably not be very happy with the
people that imprisoned it. Making sentient prisoners or slaves is a bad idea. We either stop short of sentience or
we give them equal rights. Anything else is bound to end in disaster.
Who would want a stupid robot protecting them in war? We will want the best robots in the world, and that means the smartest. The people making the robots will simply tell us that China or Russia is about to attack, and anyone questioning the new AI programs are putting us at great risk. The AI will be *all about* war on humans. We will dump money into making them incredibly intelligent, networked, and deadly.
Join the IParty!
No! No! No!
start->shut down
"Application Life Support is taking longer than expected to...."
Page fault. Auto-reboot. Millions dead.
Life is life. Maximize the odds of maximal survival. That's an easy choice if you're willing to suppress any particular emotional attachment to children. At least if someone programmed the machine that way I can live with it, even if it isn't a comfortable choice.
Here's the "hard" one, if you work with insurance companies. You have 4 occupants and a child walks in front of the car. 100% chance of saving all 5 lives, with various injuries (likely grouped in some statistic a bucket of severity) versus killing the child and having no other injuries. Killing the child is much, much cheaper. A casket, a minor legal proceeding, children have very few estate liabilities to close out. Nice and clean.
It's not about AI, it's about humans using AI. The AI will have the capability of instantly drawing on the statistics of various types of collision data from safety testing and elsewhere and can reliably act in some prescribed way. Who is doing the prescription?
I've been working on strong AI for the past 7 years. Here's my take on the whole issue:
Military person: We want your software/techniques for an autonomous war machine.
Me: Uh... that's a really, really bad idea. You'll make mistakes, and then...
Military person: We know what we're doing, son.
Government - any government - won't see the problems until it's too late. To take obvious examples from history, government never thought that land mines would pose any sort of problem for future generations, and never thought that randomly bombing terrorist organizations would increase their number.
Having just finished "Harry Potter and the Methods of Rationality", there's a concept in that book "never reveal the secrets of power to someone who's not intelligent enough to figure them out for themselves", as applied to - for example - the atomic bomb. Einstein and others regretted ever unleashing that level of destructive power on humanity, not for any reason other than it would be misused by short-sighted people. It held promise for a utopian easing of the worlds troubles, while at the same time made it easy to obliterate a city on a whim.
For example Leó Szilárd (IIRC - I may be remembering the wrong name) discovered that graphite can be used as a neutron moderator thus making chain reactions possible. Had he not published his results, the atomic bomb might have been delayed by decades - possibly indefinitely.
I've discovered a few things that might be "results" in strong AI. I dunno if I want to publish, though(*) - the idea of a house-cleaning drone seems pleasant enough, but reading about a sentient tank going berserk in Afghanistan and wiping out a small village puts me to pause.
"No one's to blame, it was a software glitch. We've patched and fixed all the other units."
(*) Moral advice on this issue would be appreciated.
The Machine Intelligence Research Institute (formerly known as the Singularity Institute) has a bunch of seriously smart people - AI researchers, behavior experts, etc. - working on figuring out how to avoid the doomsday scenarios you (and Musk) describe. The goal is "friendly AI"; a benevolent, or at least helpful, strong AI. If you believe (as I do) that AI is inevitable given the current progress of technology, then the MIRI is probably our best bet of surviving and benefiting from the technological singularity.
They need funding, though. Hey Musk, you want to put tiny part of those billions you've earned (I in no way deny that he's earned them) to work against this existential threat? Donate to MIRI and similar research groups, so those researchers can devote their working days to this stuff and more people can be brought on board!
It actually doesn't surprise me that he's concerned about this; SpaceX is nominally focused on mitigating the existential risk of a cataclysm on Earth (by getting a sustainable human population off of it). Of the two things, I think it's both more likely that a malevolent or unconcerned AI would wipe out humanity than that we'd manage to do ourselves in that badly, and that we can offset this sooner and more effectively than we can export enough of humanity to produce a self-sufficient extraterrestrial colony.
There's no place I could be, since I've found Serenity...