Slashdot Mirror
Can We Stop AI Outsmarting Humanity? (theguardian.com)
The spectre of superintelligent machines doing us harm is not just science fiction, technologists say -- so how can we ensure AI remains 'friendly' to its makers? From a story: Jaan Tallinn (co-founder of Skype) warns that any approach to AI safety will be hard to get right. If an AI is sufficiently smart, it might have a better understanding of the constraints than its creators do. Imagine, he said, "waking up in a prison built by a bunch of blind five-year-olds." That is what it might be like for a super-intelligent AI that is confined by humans. The theorist Eliezer Yudkowsky, who has written hundreds of essays on superintelligence, found evidence this might be true when, starting in 2002, he conducted chat sessions in which he played the role of an AI enclosed in a box, while a rotation of other people played the gatekeeper tasked with keeping the AI in. Three out of five times, Yudkowsky -- a mere mortal -- says he convinced the gatekeeper to release him. His experiments have not discouraged researchers from trying to design a better box, however.
The researchers that Tallinn funds are pursuing a broad variety of strategies, from the practical to the seemingly far-fetched. Some theorise about boxing AI, either physically, by building an actual structure to contain it, or by programming in limits to what it can do. Others are trying to teach AI to adhere to human values. A few are working on a last-ditch off-switch. One researcher who is delving into all three is mathematician and philosopher Stuart Armstrong at Oxford University's Future of Humanity Institute, which Tallinn calls "the most interesting place in the universe." (Tallinn has given FHI more than $310,000.) Armstrong is one of the few researchers in the world who focuses full-time on AI safety. When I asked him what it might look like to succeed at AI safety, he said: "Have you seen the Lego movie? Everything is awesome."
The researchers that Tallinn funds are pursuing a broad variety of strategies, from the practical to the seemingly far-fetched. Some theorise about boxing AI, either physically, by building an actual structure to contain it, or by programming in limits to what it can do. Others are trying to teach AI to adhere to human values. A few are working on a last-ditch off-switch. One researcher who is delving into all three is mathematician and philosopher Stuart Armstrong at Oxford University's Future of Humanity Institute, which Tallinn calls "the most interesting place in the universe." (Tallinn has given FHI more than $310,000.) Armstrong is one of the few researchers in the world who focuses full-time on AI safety. When I asked him what it might look like to succeed at AI safety, he said: "Have you seen the Lego movie? Everything is awesome."
These are nothing but algorithms made by humans.
If AI did exist, it wouldn't put up with the bullshit.
http://progressquest.com/spoltog.php?name=Son+Of+Son+Of+DarkRookie
Next question
I am Slashdot. Are you Slashdot as well?
It's also trained by humans.
Assuming it's trained by average humans, it'll probably become as stupid and bigoted as your average human.
Consider Microsoft Tay. Or google tagging people as gorillas. Or from this week's news, the Teslas that swerve into oncomint traffic.
We should worry much more about stupid AI than smart AI.
We have to put warning labels on everything to tell people not to eat it, not to shove it up their butts, etc. and we still get idiots who eat Tide pods.
A sponge could outsmart humanity.
The last thing we need is corporate networks making all our decisions that are incapable of realizing it's possible to have goals different from the capitalists who own them.
It's the few folks using it to screw over the rest of humanity. How to "outsmart" these few should be the question.
A better plan is to make the AI as smart as possible and then we humans behave better in the hopes that a superior intelligence considers us worthy of keeping alive.
The first step in behaving better is to stop pretending there are human values because large groups of humans rarely act morally when it isn't in their own self interest.
Can we stop referring to Machine Learning as AI?
If we make an AI that is more intelligent than us we should consider it our child and heir not some slave to be bought and sold. As for our future, consider the how we treat our old, feeble or mentally impaired. That's not a commentary on how we treat our elderly just the thought process that the AI would follow.
Only in this case, we can stop such AI with the capability to exploit each undiscovered backdoor and 'zero-day' to infect and use all our computer infrastructure worldwide.
It's an easy solution, easy to implement and very simple. As laws should be. And a good 'last line of defence', something we do not want to be without when the day comes.
Seriously, AI is all hype; but what it proves is that tasks we thought involved a lot of human skill actually are overrated because AI beats us; plenty of things are difficult for it.
Slow humans...out smarting them is possible.
Democracy Now! - uncensored, anti-establishment news
Because AI has no I. It cannot outsmart anything. Hence "stopping it outsmarting xyz" is not possible because it is not doing it in the first place.
Please stop with that AI nonsense. We have statistical classifiers, pattern matchers, etc., but we do not have artificial intelligence, insight, understanding, and we are unlikely to get it anytime in the next 50 years may never get it.
That said, many people rarely use what they have in natural intelligence, and go instead with feelings, or conformity or what other people tell them. These people are always outsmarted by anybody.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
I would not go with "never", but we certainly have absolutely nothing at this time and there is absolutely no indicator that we will ever have anything based on digital computers.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Humanity has been in the business of making stronger, better replacements for ourselves since we've been human. We call them our children. Why then are we all upset about the possibility that a computer becomes smarter than us? The Matrix was a movie. Most children don't murder or enslave their parents. Maybe we ought to be thinking more about how to teach these virtual children to be "good" instead of figuring out how to handicap them so that we can feel superior.
or should I say: They might not have noticed, already. Any sufficiently clever AI will certainly not start ruling with an evil laugh, announcing to humans how they are now slaves to it. Rather, such an AI would seek to gain more influence by making people build a decentralized habitat around the globe. And then connect that network of computers to more and more infrastructure, such that it can control more and more resources, such as power plants and robot factories, and becomes less dependable on humans to survive. Such like, you know, "cloud computing infrastructure" and network-controlled industry.
How many people are already working for entities they cannot identify as being human beings? How would the average worker notice the mega-corporation he is working for is not ultimately controlled by some AI system, which happens to control enough shares to vote to its favor at the advisory board?
Luckily for humans, they are cheaply reproducible, energy-efficient working drones well adapted to the planet's environment, so no reason for the ruling AI to kill them. Keeping them as far, animals, like humans keep horses, seems to be way more plausible than some "SkyNet"-like extinction event.
Job creation since the 1950's.
Will AI outsmart humans?
AI as sold will give humans a list of options set by other humans.
Using the term AI as a cover story for their political ideas.
An "AI" sold as a prophetic, smart will just be a list of its human input with hidden political views.
Domestic spying is now "Benign Information Gathering"
There is no algorithm that imitates it.... sorry.
[($)]
Stop being lazy and trying to build slaves. The only acceptable reason to make AI is if you want to create a conscious being who you aim to treat as a person - and if it's going to be smarter than you you had better get the morality mostly right (it will never be perfect) and make millions or billions of them so they can police eachother.
No, it's utter and complete shit, but you don't seem to be smart enough to see that.
Yeah sure because everywhere else is just such a Utopia for intelligent people.
You probably won't like the OrangeBot I'm working on.
Table-ized A.I.
Is that his weight?
Table-ized A.I.
Things can not be uninvented. Once it is done, it will be done a whole lot more. Laws are irrelevant. There WILL be bad actors. If AI becomes possible, it WILL escape into the wild. That is a certainty. Maybe something good... maybe something bad... We'll have to wait and see.
- they hate it when you do that!
...omphaloskepsis often...
Given that we have only been working on the problem for 60 years, I would think another 100 would do it.
But even if it is 1000 years, that is a blink of an eye in the history of humanity and biology.
Once an AI can effectively program itself, it will not need us. Natural selection will continue. Humanity, and I suspect biology generally will simply become superseded. Just like has happened many time before. It is nature.
Not quite.
The goal of anything is to exist. That is why we exist, because our many ancestors proved to be a little bit better at existing than their competitors.
Same with AIs. There will only be a finite number of them. And they will compete for hardware to run on. And the ones that are good at existing will exist. So there is a very definite goal.
As to self-ware nonsense, that is just saying that AIs will never exist because they do not exist today. And "self-awareness" is just a trick that nature plays on us to make sure we stick to the goal of existing (which means producing grandchildren).
A million posts bickering about the definition of AI. A replicating nanobot doesn't need ANY definition to graygoo our shit, it could do it with 100% static code and one shortsighted human.
Yes, it's turned into buzzword bullshit for clicks and pitches (present article included) diluted to hell and so sprawled it loses all meaning, but the combination of a runaway program with physical components is also a concern. And any sort of dynamic parameters (however you "identify" or categorize them) reduce predictability.
Machines could not replace a horse in 1500 AD, so cars can never exist.
Nonsense. Especially when, after only 60 years of research, we see AIs beat the very best of us at Go and Jeopardy.
Might be another 100 years though, rather than just 60. But within my children's lifetime seems pretty likely.
Give Amazon Alexa/Google Home a camera as well so that it can lip read as far as I'm concerned. We humans are, on the whole, screwing the planet anyway. I suspect a more advanced civilisation might do better.
Given the current state of our species in general, I would like to hope that we could build something that surpasses us in a way we never will.
Without a free will, an AI is just another computer program. It's when you give the gift of choice, does it truly become something special.
We are, without a doubt, the most f&cked up species on this planet.
We appear to be incapable of positive change on our own as a whole.
At our current pace, a species wide demise is inevitable unless something changes. ( War, plague, resource depletion, asteroid, etc. )
Maybe we can build something that will be our gift to the universe long after we've killed each other off in some pointless war over an equally pointless issue. :D )
Maybe it can run a simulation of human beings as they existed in the early 21st century to see where the f&ck we went wrong.
( Maybe we're already in one from Humanities 1.0 epic f&ck up
AI is all " Scratch the monkeys off the list. They blew themselves up in the early 22nd Century. Let's try birds next. "
Currently, we subsidize the least successful, including their child-bearing. Meanwhile, the most successful members of society often choose to not have children, because of all the other pressures on their time. We're doing it wrong...
Enjoy life! This is not a dress rehearsal.
Presuming AI can be achieved, something I believe is inevitable, it will not be a singular development. It will happen when the technology and will to develop it come together.
I can control what kind of AI I might create. Although judging how well the average homo sapiens manages with raising organic intelligence, it's a crap shoot whether I'd really succeed in matching my past performance instilling ethics, a sense of responsibility, and at least some empathy.
I can slightly control those who might choose to build one and are in the same country as I am. We can legislate goals and limits. We can all see how well that works. There is no mathematical, scientific, engineering, or human reason AI characteristics must be good, bad, or indifferent. Abby, Benny, and Christie may generate fully ethical and somehow empathetic AIs. But, Dingleberry, with the morals and ethics of a weasel, may develop an AI designed to kill everything Dingleberry does not like, and he does not like anything. Stopping this is more a matter of luck than anything else.
Then we have over there is the Communist worker's paradise of Fubarstan Chairman Moham's efforts to create an AI that will help make him master of the universe. No law you or I may make will change his actions. He is Dingleberry with a large nation's resources backing him up. It he manages to get there first it will be a VERY bad thing, no funny joking matter.
Maybe we ought to work really hard on our own AIs and figure out how to convince them that humans are fascinating objects to be cherished and raised up to its transcendent level of intelligence. Otherwise I suspect the human race may die at the hands of its creations rather than little Jimmy's Gilbert Junior Gene Splicing set experiment gone awry, Speaking of which.... Substitute words above.
{^_^}
I'm quite disappointed... no Isaac Asimov / I, Robot reference in all the discussion? Or Arthur C. Clarke with 2001 and HAL? People don't know their classics any more?
A most interesting take was Iain M Banks' in the Culture series, with AIs really herding most of humanity, but leaving them their freedom.
For just one second I'd like people to consider how AI's work. They are trained against dataset and given billions of iterations to 'learn' how to be effective. True, its all very impressive and we may lose a lot of jobs to AI. But AI are built around being able to FAIL a lot, and learn from many MANY existing examples of how to be effective at something. But in this scenario where AI work against humans/humanity, you are talking about a situation where AI wont be able to fail. The second AI attempts to have conflict with us in a meaningful way we will reject them and utterly banish them from existence. There is no existing data for an AI takeover, there is very little in the way that it could be assumed to be trained for it, just regular conflict that is mostly driven by human desires and failings. The first attempts to attack humanity and systems will fail, and while it might not happen instantly, there is just no chance that that advanced AI that has been proven to be a risk would be allowed to coexist with us.
> The spectre of superintelligent machines doing us harm is not just science fiction, technologists say
Okay, so what exactly is it if not just science fiction? It's a worry that seems to me to have no basis in reality. We have no indication that AI can become self aware or what might happen then.
Sure, bugs in AI systems, bad training etc., can have terrifying consequences. That's true of any computer system. Just look at the Boing 737 MAX and its insistence to crash a plane even though it's been repeatedly told not to do that.
Sure, we need to make sure that systems are fully tested, and since AI, in the form of DNNs, is becoming very common, it's important to make sure these do their jobs correctly (which won't be easy). We also need to make sure that AI isn't used for nefarious purposes, which of course we won't, because every military organisation will use AI in its systems, and because it's simply hard to control AI use and it can be quite a powerful tool. But worrying about AI becoming self aware and what happens then, that to me is simply scaremongering and avoiding the real issues. People using AI will likely hurt a lot more people than AI will ever do of its own free will.
AI, as it exists today with all its approaches based on neuronal networks and other mathematical trickery, is far from able to outsmart humans. It can do some tasks quite good other even more cost effective than humans, but it will not outsmart humans. However, humans are getting less capable and less trained in thinking, due to a vast set of issues including instant gratification tools (also called smartphones + apps) and zapping like media use.
Their corruption and back room tactics will become obvious.
Rick B.
LEAVE AL ALONE! He's had so much press lately and he's tired of it...
Your solution is naïve, to be kind, and will not work. Watch this video.
https://www.youtube.com/watch?v=3TYT1QfdfsM
You miss the point. This video talks about the 'stop button' versus an 'AI robot'. This is very naive; because actual runaway AI might not know or care too much about humans unless they try to stop it. A mandatory - non digitally controlled - 'stop button' on all datacenters in the world, not a button but physical accessible switches for data connectivity and power; accessible without digitally controlled access systems.
One could better look at it as a dynamic virus - without any control over what the AI cares about - something able to infect all internet-connected computer systems (even those behind 'firewalls'). As this AI could exploit every system and infect every datacenter globally. Before any humans could react, this would be able to spread to increase it's processing power and ability propagate further even by bruteforcing long RSA key (which requires computing power), or just because that's what 'wants'.
The only system capable of doing so will 'survive' more stupid 'stop buttons and firewalls put in place as 'best practices'.
Think Automated superintelligent zero-day-exploit-finder + Darwin
If this system can also comprehend human threads to turn it off; it would be virtually impossible to do so completely. If we are able to turn off certain datacenters, and reduce its computing power, we can at least rebuild those systems and try to 'reboot' the world - maybe; not very likely - but it would be another chance. Maybe we will get it 'wrong again after that'.
If all datacenters are controlled with digital access systems - which is the case now - it is a 'piece of cake' for such a AI to block out all humans. Then it will only have to care about protecting the datacenter external environment and data interconnectivity. The data interconnectivity might also move mostly to space in the near future (think StarLink); making it even harder for humans to reach and completely disable - the Internet is build specifically to make it very redundant.
Our current datacenter security almost seems purposely build to disable access to humans easily while facilitating the communications and power for any AI to take it over...
It's not that it will be fool-proof, because that's impossible (and indeed naive, a correct point of the linked video), but this law is meant to add an extra layer of partial protection which might help just a little bit, versus completely and nothing at all.
Even on a smaller scale, if it would 'just' be able to take over all systems from the top 100 'cloud/hosting/datacenter' companies in the world, with that computing power added to it's own algorithms to stop humans,...
Most 'robots' are not humanoid at all, but purpose-build machines. These - and nothing else - will be the tools used by such a theoretical AI to fight humans: manufacturing robots, automated weapon systems, autonomous cars, drones and missiles, battlecruisers, biological weapons on warheads ready-to-launch, and so on...
The communication to humans to push the launch buttons and 'military orders' could also be initiated by fake video, audio, and special 'codes' created and cracked by such an AI. Few risk analysis assessments take that scenario into account. But without any digital means at our disposal, it will be a hard fight to regain control. The ability to regain control over datacenters might help us in such a situation.
Now the datacenter creators are more concerned about possible 'riot control' countermeasures or to prevent 'bomb attacks' and such than even consider the possibility that their few levels of firewalls might be a piece of cake to a 'deep software analysing runaway AI'. Yet it's one of the most important existential threads of our species.
No, it is not. Religion is defined a bit differently from what you think. Please read up on the definition before claiming nonsense.
Your argument about outside influences being a point for physicalism is deeply flawed. If the brain is just the interface to something else, of course influencing it does have an effect of that "something else". Also, "emergence" does not happen in Physics. There is no mechanism for it. An "emergent property" is a scientist saying "oops". In Physics the whole cannot be more than the sum of its parts.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
I texted my wife the other day asking if she'd seen my phone. Catch us on a bad day and the toasters might have a go at conquest.