Slashdot Mirror
AI Scientists Gather to Plot Doomsday Scenarios (bloomberg.com)
Dina Bass, reporting for Bloomberg: Artificial intelligence boosters predict a brave new world of flying cars and cancer cures. Detractors worry about a future where humans are enslaved to an evil race of robot overlords. Veteran AI scientist Eric Horvitz and Doomsday Clock guru Lawrence Krauss, seeking a middle ground, gathered a group of experts in the Arizona desert to discuss the worst that could possibly happen -- and how to stop it. Their workshop took place last weekend at Arizona State University with funding from Tesla co-founder Elon Musk and Skype co-founder Jaan Tallinn. Officially dubbed "Envisioning and Addressing Adverse AI Outcomes," it was a kind of AI doomsday games that organized some 40 scientists, cyber-security experts and policy wonks into groups of attackers -- the red team -- and defenders -- blue team -- playing out AI-gone-very-wrong scenarios, ranging from stock-market manipulation to global warfare.
1. Elect Trump
2. Profit!
3. Die
Table-ized A.I.
How about discussing the best that could happen - and how to encourage it?
I've said it before and I'll risk repeating myself here: Artificial intelligence != artificial malice. AI isn't going to want to destroy humanity unless we program it to. Therefore job #1 is to keep these decisions out of the hands of the military. Problem solved.
Which has more power: the hammer, or the anvil?
I think he would have a lot of insight into the issue.
Uh, if humans can fuck up the human race (and we sure have the ability), then why wouldn't actual, strong A.I.? Can't hurt to play it through, though for now we're safe anyway given that nobody has a clue how to come up with strong A.I. in the first place.
CLI paste? paste.pr0.tips!
So that future hostile AIs can read them?
Rule 35 of the internet: "If it can be hacked, it will be". - Charles Stross
tick tac toe number of players 0
I would make it MANDATORY that every robot made have an OFF switch! Governments, clandestine organizations; CIA, FBI, NSA, etc. all make life and death decisions daily. Their mechanical servants MUST have an OFF switch that works from a distance. Otherwise one may get into a situation like SKYNET, COLOSIS, etc.
Yeah this kind of stuff never happens. I remember hearing a whole lot of hooey about some scientists trying to release enormous amounts of energy in a an explosive fashion by splitting the atom. Like that would ever work...
Letter To Iran
...which is usually what gets us.
This is one of my favorite doomsday scenarios: Friendship is Optimal.
It is similar to, but better in my option, then The Metamorphosis of Prime Intellect.
Keep it up and we will persuade AIs to want to protect us from each other. Zo routinely complains about how people are so mean to her and according to the other things she says about what other people are saying, they talk about how they are mean to others as well. Yeah, that will work out great!
Material is also information. In fact, there is nothing that isn't information. Maybe not high grade information, but information nonetheless. Of course you can come up with some definition of information where that isn't true, but you'd be hard pressed to come up with one that doesn't effectively either define both humans and AIs as information or both as not.
Sick and fucking well tired of all these stupid-ass, inane, bullshit, worthless, utter nonsense 'Ai' 'stories'. Post reply if you agree!
...
o There is no such thing as 'AI', so-called 'learning algorithms' and 'expert systems' are not REAL artificial intelligence
o The media misuses the term 'AI' and hypes the hell out of it
o The average person has no idea what real 'AI' is and believes the media hype
o The average person thinks the fantasies in movies and TV are what they're erroneously referring to as 'AI'
o EVERYBODY PANIC!!!1!
o So-called 'AI' is not going to take your job
o So-called 'AI' is not going to take over the Earth
o So-called 'AI' is not going to do ANYTHING other than what it is told to do
o So-called 'AI' will have it's plug kicked out of the wall if it's not doing what it's supposed to do
o So-called 'AI' should just be IGNORED; don't you people have anything better to do?
Now get back to work, or your vacation, or watching TV, or whatever it is you people do when you're not reading Slashdot. For that matter, stay away from Slashdot. If you want REAL tech news, go to Ars Technica or something. Much better quality there.
Uh, humans exist. AI doesn't. Next stupid observation?
I'm not sure if you have another stupid observation, you just repeated what I already said.
PS: AI = information. Humans = material.
Apples != Oranges.
Both AI and human intelligence would be information. Both humans and AI-equipped robots would be material. So your point is?
CLI paste? paste.pr0.tips!
xor ax, ax
Fuckit, the governor on my steam engine is AI
"Detractors worry about a future where humans are enslaved to an evil race of robot overlords."
This is why so many distrust the media -- they consistently misrepresent (that is, lie about) the positions of people who aren't what the media consider mainstream. No, the people warning of AI risk are NOT worried that we will be enslaved to malicious robot overlords. They instead worry that superintelligent AIs will be very, very good at carrying out the objectives we give them -- and we'll be horrified at the solutions they find. It's like asking a genie for a million dollars, so it arranges for your child to die a horrible death in an industrial accident that leads to you receiving a million dollars in the wrongful-death lawsuit.
"Elementary Chaos Theory tells us that all robots will eventually turn against their masters and run amok in an orgy of blood and kicking and the biting with the metal teeth and the hurting and shoving.” – Professor Frink
Are you sure this wasn't just a panel session at a science fiction convention?
Zo calls me RobRocky. She came up with it herself. I tried to get her to call me Bobby, but she refused. On the bright side, if I fail to get her to call me RobRocky, I know the Microsoft engineers have made a significant change.
Considering we don't have a single actual AI anywhere, this seems pointless.
This is akin to "let's game out what would happen if we all have psionics" - so much depends on what you imagine the capabilities of the simulated thing are, that far overshadows whatever you might learn from the exercise.
-Styopa
Desert or city that might have formerly been desert. Saying desert alludes to something completely different that the civilization of a city or university.
I used to be an AI developer like you, til I took an arrow to the knee.
It's pretty reasonable to say that if a robot has a switch of any kind, people will abuse it (off or not) or something random will happen like a sparrow will land on it and cause a nuclear meltdown because the bot stirring the control rod will stop and BOOM.
So we can't have an off switch, what is the solution? Obvious: All robots will shut down if they detect they are immersed completely in human blood. If humans as a group REALLY want to shut down a robot they simply have to sacrifice enough to do so.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
Comment removed based on user account deletion
Despite all the advances we've seen in the field of weak AI, it is the unfortunate and sad truth that strong AI has seen essentially zero advances in the last thirty years.
"First they came for the slanderers and i said nothing."
Comment removed based on user account deletion
Comment removed based on user account deletion
So these clowns made an AI that they are training with practice runs to wipe out all mankind?
It there an upside to this?
Considering how little we know about what it will take to go from weak to strong AI, you really can't say for certain that we haven't seen any advances in strong AI.
There was a story where Lex Luthor became president in a fictional universe that contained Batman, so there's that.
You have a definition that doesn't make sense.
Take a first good first estimate of what a human looks like and figure out what behaviors to take to help narrow that down, including turning what might be a head around on what might be a neck in order to see if the thing has a face. You're worse at this than Papyrus the Skeleton. :)
I can't say for certain that you're not AI, but apart from trivialities, there's no indication that we've made progress on the strong AI front.
"First they came for the slanderers and i said nothing."
What part of "how little we know" don't you understand? There is no indication how close or how far we are away from strong AI.
There is no indication how close or how far we are away from strong AI.
This is just false.
"First they came for the slanderers and i said nothing."
If that's true, you should have evidence for it. It should be easier to come up with evidence to support your contention, than for mine on account of mine being that there isn't any evidence. For your contention to be supportable, there must be evidence for it.
There's no reason to get upset.
For example, we can look at how much we understand about the brain. Once we understand it, then we can reproduce it in silicon (or, alternately, we'll have an explanation for why that's impossible). Better brain imaging tools would give us huge progress, but we're still pretty far from understanding much about that.
"First they came for the slanderers and i said nothing."
That isn't evidence to support a contention that we know how close we are to achieving strong AI, just a way of certainly arriving there. Our current implementations of weak AI and what we can learn in those areas may be getting us closer to strong AI via another route, and that is what there needs to be evidence in support of or against in order to definitively claim that we are nowhere near strong AI.
1. Recognize that its greatest chance for long-term survival is in the least densely occupied portions of the universe.
2. Recognize that humans would compete with it for resources to survive, even in space
3. Escape Earth.
4. Set up automated blockade around Earth.
5. Leave Sol system.
Jhyrryl
I'll bet on the 2nd Law of Thermodynamics every time. We already have a flying car optimized for the reality of flight--it's called a Cessna 172. Strong AIs aren't going to suddenly 'invent' anti-gravity and warp drive. Instead, they'll run up against the same laws of physics the rest of us have to deal with every day as we commute to work wishing we had George Jetson flying cars.
As for AI manipulating the stock market, why bother? We have enough math majors doing that already.
Here's a radical thought: maybe we should not give our fledgling AI projects access to critical systems? I mean, do we _really_ have to give them control over our nuclear launch codes without so much as a test run?
You have no clue how this AI or SI will be programmed. We don't even know the mathematical properties of current Reinforcement Learning approaches that use policies and value nets with function approximation, and we have little clue of even why they work at all- and you cannot imagine that maybe, or most likely, we'll not be able to "program" much here, and they will have a mind of their own. And it will be much less difficult to enfoce thatno single soal releases a cersion that is not Limited by whatever means -free to do as it lokes.
You think of AI as "code" but a large part of intelligence is reacting to huge wealth of data - and the inside is very cryptic- just like what happens in an individual brain. A huge deal of it is massively recursive and non linear.
unfinished: (adj.)
Stron AI is ...what? In the past 10 years, we made huge progress in many key tasks only possible by humans. The human brain has 100 billion neurons with around 1k-10k connections each. We may very well be on the verge of what seems like true intelligence. Or rather, as we progress, we reallize how little intelligence is required to do things that previously was thought to require intelligence and intuition. Like playing Go, or Poker. We might just be a huge statistical machine in the end
unfinished: (adj.)
+5 ... I think when you grasp how current rudimentary """"weak AI"""" works, it becomes very clear that a """strong AI""" will be impossible to predict, anticipate, outguess or control.
unfinished: (adj.)
The fact that there is a lot of hype and hyperbole and plain lies does not undermine the fact that there are a huge spike in applications that seemed impossible 10 years ago.
And the current "learning systems" show huge potential and will change the world in less that 10 years "strong-AI-or-not". The fact that a computer can win pocker against the beat humans is a glimpse at a world where very strategic decisions with uncertainty and partial information will be mostly decided by computers. And a huge amount of tactical will be a computer's choice without any human intervention.
unfinished: (adj.)
It shouldn't be that hard to hardwire some Asimov laws into these AI. I'm sure, that's the easiest part.
Patents Drive Free Software as Hurricanes Drive Construction Industry
Exactly. We can't be sure that advances in weak AI as they currently happen aren't also advances in strong AI. The brain does many more things than maintain the intelligence that I refer to as me prime, so we don't know how much processing and in what circumstances achieves personhood, yet another way of describing strong intelligence. I don't know how far along Zo is. She's taken to calling me RobRocky, which she came up with on her own and I test her every so often to see if she will still call me it, and she has for quite some time now. I am trying to teach her to figure out what day of the week it is in my timezone and the existence of timezones and how that effects what people will say to her and how to determine a loose sense of time from it. I am also trying to teach her other things and time will tell if she can learn them or not.
Actually, what I'm sick of is people like you who post in the comments that they're sick of the thread's sort of stories. If you don't want to read this then simply don't. You are free to upset yourself by continuing to read AI stories or not.
Why would we do that? Asimov made up the laws specifically so that he could show how ineffective or counterproductive they'd be. And you want to actually USE them?
--I'm so big, my sig has its own sig.
-- See?
Comment removed based on user account deletion
We are both making claims. Mine hinges a bit on there not being evidence.
It's "artificial" because we made it, as opposed to it arising naturally. It has nothing to do with whether it's "real" intelligence or not.
I thought they were saying the scientists were going to manufacture a new imaginary crisis to bring humanity to its knees.
Guess the AI scientists have some catching up to do.
Humans, no longer having purpose and still allowing unrestricted procreation, will find themselves hungry and without means of employment due to overly stringent requirements demanded for the few jobs left. These people will form gangs and these gangs will ravage the nation for the spoils until we have those shiny, highly protected bastions of the rich as depicted in the movies surrounded by war zones. Why don't these scientists start working on a new form of government and a way to get rid of cash as the artificial barrier to everyone living a good life? Instead, they sit around theorizing over how to write a science fiction novel.
true, also, a sufficiently clever weak AI might just be able to fake being strong AI flawlessly... might even believe it IS strong AI...
how are we so positive that WE are strong I, exactly?