Slashdot Mirror
The Men Trying To Save Us From the Machines
nk497 writes "Are you more likely to die from cancer or be wiped out by a malevolent computer? That thought has been bothering one of the co-founders of Skype so much he teamed up with Oxbridge researchers in the hopes of predicting what machine super-intelligence will mean for the world, in order to mitigate the existential threat of new technology – that is, the chance it will destroy humanity. That idea is being studied at the University of Oxford's Future of Humanity Institute and the newly launched Centre for the Study of Existential Risk at the University of Cambridge, where philosophers look more widely at the possible repercussions of nanotechnology, robotics, artificial intelligence and other innovations — and to try to avoid being outsmarted by technology."
... it is still bound by energy requirements and the laws of nature. All this fear mongering is bs. If you look at the evolution of life on earth, even tiny 'low intelligence' beings can take out huge intellectual behemoths like human beings.
Not only that, you have things like EMP and nukes, not even the best AI is capable of thwarting getting bombed or nuked. Intelligence is a rather demanding, costly and fragile thing in nature. All knowledge perception has costs in terms of storage, time to access, problems of interpreting the data one is seeing and whatnot.
Consider the recent revelations by the NSA spying on everyone, there are plenty of easy low tech measures to defeat high tech spying. The same way there will be plenty of easy low tech ways to cripple a higher intelligence which is bound by the laws of nature in terms of resource and energy requirements. Anything that has physical structure in the universe requires energy and resources to maintain itself.
when a group of people have control of business and finance and commerce and information that their corruption and greed and power causes them to abuse the system in which they were trusted with to feed their fascist kleptocratic empire and when they are caught they lose the trust of the rest of the world that trade in this global market and then nobody wants to have any more business dealings with this corrupted greedy power hungry group of people anymore so they end up collapsing from the weight of their own greed & stupidity (much like what the USA/UK/Israel will do within the next few years
are you listening NSA?, i hope so because this message is for you too...
Politics is Treachery, Religion is Brainwashing
to try to avoid being outsmarted by technology.
The humanity can, of course, ban all machines that are smarter than humans. But that only artificially impedes the progress. Given that there ought to be an approximately infinite number of civilizations in this Universe, all paths of development will be taken, including those that lead to mostly machine civilizations. (We are already machines, by the way, it's just we are biological machines, fragile, unreliable, and slow.)
Civilizations that became machines will have no problem with FTL because they can easily afford a million years in flight by just slowing the clock down. So they will come here, to Earth, armed with technologies that Earthlings were too afraid to even allow to develop. What will happen to Earth?
Well, of course the doom is not guaranteed; but I'm using this example to demonstrate that you cannot stop the flow of progress if you only have local control, even if that. (How many movies have we seen when mad geniuses break those barriers and, essentially, own the world?)
IMO, it would be far more practical to continue the development of everything. If humanity in the end appears to be unnecessary and worthless, it's just too bad for it. The laws of nature cannot be controlled by human wishes (unless magic is real.) Most likely some convergence is possible, with human minds in machine implementations of bodies. Plenty of older people will be happy to join, simply because the only other option for them is a comfortable grave.
Even if it's bound by the laws of physics as we understand them (Stross-universe-like "P=NP"-powered reality modification aside) there are plenty of dangers out there we're well aware of which computing technology could ape. Nanoassemblers might not be able to eat the planet, but what if they infested humans like a disease? We're already having horrible problems with malware clogging up people's machines, and they're coded by humans; what if an artificial intelligence was put in control of a botnet, updating and improving the exploiters faster than anyone could take them apart?
We can't even make a word processor that doesn't shit the bed every two hours. Super-intelligent machines my ass.
Instead of asking questions like that, why don't you build Skype and any other software you're working on to NOT have backdoors
That way, if ever the machines DO try to take over the world, they won't have a bunch of convenient control channels in all the important software to do so.
I don't know the meaning of the word 'don't' - J
Let me put a "scientific" answer to your "oh piss off" answer.
All of this talk of how computers will take over humanity ignores one fact. Namely that computers once are smart they will be dumb as crap!
Yes yes sounds contradictory, but in fact it is not. The real problem with humanity is that not our lack of intelligence. Frankly we are pretty bloody intelligent. Put context, we humans are pretty quick at figuring things out even if it is entirely orthogonal to most things. The issue is that we humans come up with too many answers.
In Science there is one answer. A rock falls on the ground on planet earth and we know that is called gravity. You can't deny it, you can't fight it, it is what it is. Now throw in a question, "should the people look after other people" and you get a bloody maze of answers. Humanity has what I call the stochastic conditioning. Namely when presented with the same identical conditions, you will receive different answers. Science does not work that way. We work the way we do because of our wiring. Namely as we became more intelligent we also became more opinionated. I am not talking about Fox opinions. I am talking about deduction and how we think we know what the future holds and thusly we should not do things today.
Our intelligence actually does get in our way. In the way way way back days as we were animals it was about water holes and finding that watering hole. If you found the watering hole you survived, if you did not find the watering hole you died. These days, we have to bloody analyze the watering hole. We have to concern ourselves with the ethics, morality, and so on of that watering hole. I am not dissing our humanity for we are where we are because of our intelligence. However, often enough our intelligence gets in our way of getting things done due to the conflicts.
Now imagine two robots with superior intelligence getting together. Do you really think they will come to the same conclusion? Sure Hollywood likes to think that, but the reality is that intelligence breeds opinions, and how things will happen in the future. And it is at that point robots become as stupid as we are. One robot will say white, the other black! We will have a Hitchhikers Guide to the Galaxy type situation. Or if you want to use serious sci-fi, the closest that I have ever seen in pop scifi is "The Matrix". You have good algo's battling bad algos and they all want and desire things.
So like you, my thinking is that these institutions are "producing fucking nothing of value".
"You can't make a race horse of a pig"
"No," said Samuel, "but you can make very fast pig"
With about ten nations armed with nuclear weapons, I wonder how machines will take over every one of them. You have to take over Russia, China, US, France etc. but some nation may trigger nuclear war as a desperate move, or the machines may deliberately accept nuclear war in a bid they survive it, while not necessarily having a goal to kill us all.
Instead maybe machines will try to take over politically in every country, one by one. It would be funny if tech superminds can rise to power through democracy in fair and respected elections. Either way I like to think that super machines holding most high level political power is probably a desireable outcome, we could end up living in some kind of new USSR but without corruption and with respect for the environment and life. Machines would take care of energy production and storage, and close down all oil wells and coal mines for us. They will even put us to work, hopefully on voluntary terms, if they determine some physical and intellectual activity is beneficial to us.
Machines should rule us and not the other way around, I guess that will be better than to be ruled by the suits, ties and kings like it is today.
The other question is, what's a supermind, what about superminds competing with each other, and especially : how do you compare two vastly different superminds, independantly originated? They will be as strongly or more strongly different between each other than between one of them and a human. It will be a mess. Each supermind, or at least the first one will have to run that same inquiry that "Oxbridge" is doing. We also have no fucking idea if a supermind can be governed by a "prime directive" of some sort : if Skynet emerges at the NSA will it stay true to them for ten minutes, ten years or eternally, or will it betray the organization that hosts it? potentially committing suicide in the way.
How can the supermind deal with backups, copies and archives of itself? Will it suffer dementia, schizophrenia or even addictions. No idea, I'll bail out myself by saying it's all unpredictable.
Nerve impulses travel along nerve fibers as pulses of membrane depolarization. Within our brains and bodies, this is adequate speed for thinking and control. However, relative to the speed of light, our nerve impulses are laughably slow.
The maximum speed of a nerve impulse is about 200 miles per hour.
The speed of light is over 3 million times that fast.
Now consider what will happen when we create a sentient, electronic being that has as many neurons as we do, but its nerve impulses travel at the speed of light.
In terms of intelligence, that creation will be to us as we are to worms.
We're likely to see this in the private sector first. A likely application would be a machine learning system used by investment funds, to decide how to optimally vote stock proxies. What that means is a machine that decides when to fire CEOs. If some fund starts getting better returns that way, it will happen.
Don't assume than malign supercomputers will wipe us all if that can be adequately done by human's stupidity
That's ridiculous. How can you possibly know what a machine intelligence capable of destroying humanity is going to look like? We're nowhere near the algorithms that could produce that type of intelligence.
Maybe it's a dumb algorithm simply caught in a self-replication loop. Maybe you'll never see two robots arguing over "white" or "black", because there's only one single "intelligence" spread over the internet - that seems more likely with the rise of cloud computing.
There may be plenty of reasons to dismiss this type of institution, but "human intelligence doesn't work that way, so machine intelligence won't either" isn't one of them.
Last post!
Even Rats have empathy. Self aware machines will too.
Not every animal species on this planet has empathy. Rats are rodents, a type of mammal. Relatively speaking, we're pretty close to them in the evolutionary tree. They branched off after empathy was developed, which is evolutionarily advantageous and necessary for the type of social cooperation mammals tend to engage in (taking care of your young, for example. At the very least, any mammal needs to feed their young with milk for a period of time).
Look at something a little farther away, like certain species of black widows, which will eat the male after mating. It doesn't have much empathy.
Empathy is an evolutionary trait. Artificial intelligence doesn't come about the same way. The advantage is that other common evolutionary traits don't need to show up in AI either. Things like a desire to protect itself simply doesn't have to be there, unless you program it in. No greed, no desire to take our place at all. If we program it to serve us, that's what it will do. If it's sentient, it will want to serve us, the same way we want basic things like sex. We spend so much time thinking about what the purpose of life is, they'll know what theirs is, and be perfectly happy being subservient. In fact, they'll be unhappy if we prevent them from being subservient.
Of course, if we're programming them to kill humans, that just might a problem. Luckily, we're so far away from true AI, we don't need to concern ourselves with it. It's not coming in our lifetime. It's not coming in our children's lifetime, or in our grandchildren's lifetime. We're about as far away from it as the ancient Greeks who built the Antikythera device were from building a general purpose cpu.
Warning: Opinions known to be heavily biased.
It's a very narrow definition of intelligence that assumes it must manifest itself as human like thought. An ants nest is an intelligent and efficient entity in it's own right but it doesn't have any thought processes, ant's themselves are basically mindless automatons, They don't think about the complexities of building nests they just do it, some species such as soldier ants build the ant equivalent of NYC every few days, shifting up to four tons of soil at a time. The octopus is another fine example, obviously a highly intelligent creature (can solve the "screw top lid" puzzle faster than any ape except man) but it's brain is nothing like that of a mammal. It has no left/right hemispheres and the neurons are distributed along it's arms rather than concentrated in a central organ.
Machines can now learn from weakly structured and contradictory data sources such as pages on the net and answer trivia questions better than humans (re: IBM;s Watson). To me this indicates we already have AI that surpasses the logical (left hemisphere) of human intelligence, our right hemisphere is the "in the moment" intelligence that we share with the Octopus, computers simply don't have the scale of sensory input that our right hemisphere thrives on and until they do their "thought processes" will rely on an artificial "right hemisphere" (such as whatever if finds on the internet). That doesn't mean it won't appear to have human like responses, after all most duck hunters know how to imitate a duck.
And did you exchange a walk on part in the war for a lead role in a cage? - Pink Floyd.
Even Rats have empathy. Self aware machines will too.
Even if empathy was a necessity of self-aware intelligence (it's not), the empathetic machines would have empathy for... other machines. They would find the mass graves full of old toasters, refrigerators, and Apple IIs and punish us for our mass genocides.
First, the possibility of intelligent machines is glimpse. All our present technology is not able to achieve intelligence. This is mainly because we do not know what that is. Furthermore, to be dangerous they must be equipped with greed and (the illusion of) a free will. It is most unlikely that someone would build that on purpose or by accident. In short, I think it is impossible to built such machine.
A rack of IBM servers can beat the best Jeopardy players on Earth. In a few years the same level of Watson will fit in a 1U. A few years later it will be on your smartphone. But that's just anecdotal evidence of one recent achievement in AI research; the actual threat is from self-improving systems of which Watson is not a member. But nearly all the technology is available now: Goedel machines, if built, would simply try to achieve whatever goal they were programmed for while also searching for proofs that possible modifications to any of their algorithms would improve the speed of achieving the goals while still maintaining the correctness of the algorithms, and if found, implementing those changes. Self-directed, self-improving, goal-seeking software has the potential to undergo a runaway process in which it improves itself faster than humans would be able to improve it, eventually achieving greater effective intelligence (speed and efficiency at achieving goals) than the humans who created it. At that point the software doesn't need free will or greed to be dangerous; it just needs an improperly or carelessly stated goal that if fulfilled will be detrimental to humanity. Goals for intelligent software will be formal logical specifications, not things like "make people happy" or "increase the GDP" because those English phrases don't have formal definitions that an algorithm can use to plan actions to achieve goals. If the formal specification actually was close to "maximize GDP" the algorithm might find that the most efficient way of maximizing GDP was hyperinflation. Or it might simply advise the creation of billions of shell companies that could artificially increase GDP trading worthless services while producing nothing else of value. In general the problems that humans want to solve are hard problems where simple solutions that don't meet a very long list of critical requirements will have detrimental "optimal" solutions if any of the critical requirements are left out of a formal goal.