It is true that we are probably quite far from AGI, but nobody really knows how far. We could be closer than most experts think. Few experts in 2006 would expect deep neural networks to beat a Go champion or be able to describe visual scenes in natural language sentences within a decade. Unexpected things happen, and we are much further behind in terms of preventive technology like Friendly AI. (https://en.wikipedia.org/wiki/Friendly_artificial_intelligen...)
Given the existential risks involved, I would propose more careful handling of the technology. (I agree that opening up most other technologies is usually a positive thing.)
See "Superintelligence" for comprehensive arguments against unfettered development of AGI: https://en.wikipedia.org/wiki/Superintelligence:_Paths,_Dang...
There's an episode of Pinky and the Brain where Brain invents a magnification ray and gigantises Pinky and himself - so they're now giant rats [1]. They walk about some human metropolis, looming larger than skyscrapers and terrorising everyone. To cut the story short, Pinky happens and eventually the gun shoots everything in the world.
... so Pinky and the Brain are now normal-sized mice again. Or, well, they're giant mice, but everything in the world is also giant, so they're small.
I think of that episode during discussions of the dangers of AGI and how "democratizing" it will eliminate those dangers. The problem is that AGI, like nuclear weapons, is not an equalizing force only- it is also a very destructive force, and so there is the danger that any loon with access to it could blow us all to kingdom come (with nuclear weapons, for sure; with AGI, perhaps).
On the other hand, when nuclear weapons, or AGI, giant rays etc are not democratized we have the situation we're in today: only a few entities have access to them and they basically have free (military) reign... and there is still the danger that some loon might get to a position of power were they can push a button and -blamo.
Which is why people are really worried about this sort of thing. Because once certain discoveries are made, there's no going back and at the same time the road ahead is full of danger.
https://wiki.lesswrong.com/wiki/Paperclip_maximizer
One of the reasons AGI will be so dangerous is the near certainty it will have utterly alien values and intentions.
It seems exceedingly unlikely that we will be able to program in what consitutes "human values". We can't even begin to remotely agree and explain them fully to fellow humans, let alone what is basicly an alien intelligence.
It is slightly ironic that people bring up Assimov's laws of robotics in this context, because pretty much every story he wrote was about how the laws could be subverted via "loopholing" of sufficently intelligent AI.
If something is very much smarter than humans, we cannot possibly hope to constrain it via some kind of faustian bargin agreement. Because in the end, the devil is always smarter than you are.
If an AI kills you from either malice or indifference, or just that you are composed of plenty of paperclip-capable raw materials, you are dead all the same.
The most chilling of these is the "Robot Dreams" short story.
We don't know how many other people aren't working on this because of moral/ethical reasons. Of course, 99.9% of the world could be wary of genetic engineering, but that remaining 0.01% is enough to pursue research, get VC investment and drag the rest of us into that uncertain future.
To continue the nuclear analogy, the Manhattan project was probably the most impressive engineering program in the history of the world, but it was driven by survival in a World War. They didn't build and drop the atomic bombs for fun. You'd think that working on a limitless virtual brain should have similarly serious motivations, not just the examples of "how do I drive to X but also shop for Y?" or "is that a monkey?".
I know there are much grander societal goals with A.I. and the world really could become a "better place", but please sell society those goals, not the usual first-world problems. We already have enough people trying to destroy the world without these extra tools.
A super intelligent AI does not change the laws of physics or even get to predict the weather two weeks from now.
Sure, they could learn to be a good hacker but the worlds best hacker can't get into air gaped systems. More importantly, flawless security trumps hacking. If there is nothing to exploit there is no way in, which means a lone AI is much more powerful than lot's of AI's with different goals.
I don't actually believe this will happen, but I'm not 100% confident it wouldn't. I don't know what a superintelligence would look like!
If the world is deterministic, a sufficiently powered and advanced AI with enough data could actually predict the weather, not just for two weeks, but for years. This also enables many other things viewed as magic.
This, of course, is incredibly far off. And I don't see a flaw in your point on flawless security. Though, I doubt we have many, if any, truly flawless security systems in existence. We just don't know their flaws, making them good enough for now. The world could get really interesting once AI's start programming and writing app security to protect itself from malicious AI's.
For a detailed discussion of chaos and weather prediction, see: http://www.ecmwf.int/sites/default/files/Chaos%20and%20weath... The "trick" for dealing with chaos is to do a bunch of predictions with perturbed values. If they all end up the same, then the forecast is probably okay. If they vary widely, then you know not to assign much accuracy to the forecast.
I think the the other, perhaps more prescient point here, is about whether or not that kind of exponentially-learning AGI being discussed is even possible given that the substrate such things would be embedded in (hardware, software) which are [or appear to be?] deterministic. One assumes that that kind of learning implies learning about or aquiring capabilites which have not already been determined or specified by a human. At some level of abstraction, humans _must_ determine that system. It is possible to concieve of a machine that is non-deterministic in that way?
At what point does such a system actually leap beyond what has been specified by us? Does this imply a machine that not only rewrites itsel in softwate, but also remanufactures itself in hardware?
What I think the danger with a super intelligent AI is being able to correlate massive amounts of data to see patterns we miss or things to try, and never being bored. Look at automated exploits today vs the old "hacking by hand". Brute force attempts not just on known patterns, but being able to create new tests and analysis.
A likely potent form of gaining power may involve assuming the identities of people one knows, via text messaging or email, and use that to gain access to important information. The aggregation and superhuman-level analysis of private information thousands of times faster than humans would bring enormous financial and social power to the AGI. Given the sad state of security in most systems, and the relative ignorance of information security issues among many political and business leaders, an AGI will not have much problem with this sample plan off the top of my head.
And there are many other more clever plans an AGI could come up with that we have not yet considered...
Further, spinning off other copies requires hardware to run them. If MS has an AI 3 go over windows source code it may be much harder to hack desktops.
Without authorial fiat a world of real world AI's may be much more stable and boring than you might think.
And in the real world, not everyone exclusively uses services and systems from Google and Microsoft, presumably two of the most competent technology providers. Plenty of people also use ad hoc services set up by internal teams.
No security is flawless in the face of a sufficiently determined attacker. No system has nothing to exploit.
All the efforts of ML solve problems with a concrete input space and a concrete output space. This means they solve very discrete problems and are not generalizable, unless they have similar looking model inputs and the model has the same limited range of "moves" it can make.
We've got all kind of past fiction which plays on these concerns, more recently Ex Machina and Westworld.
But here people are making the same bold pronouncements...so certain we know or can control technology which hasn't even been invented yet.
I see no reason that's necessarily true. I'm a general intelligence, and I can't make myself smarter.
A fictional example, but Alastair Reynolds' "Inhibitors" are a type/race of machines which, while intelligent, were specifically designed to limit their own degree of sentience.
But interesting things happen when you network billions of biological AGIs together, it leads to all sorts of emergent phenomenon, and now the biological AGIs are working on these newfangled mechanical AGIs, which, while still crude, aren't bound by the same constraints, they can iterate much faster. Biological AGIs have crippling bandwidth/memory issues which aren't really a problem at all for their mechanical counterparts. These mechanical AGIs, I think they'll go places.
You and me both. I'm afraid it's because we're only just barely sentient. If you think about it, in evolutionary terms we literally only just now managed to build a technological society because we only just got smart enough to do it. We are by definition at the absolute minimum level of intelligence that's able to do that, otherwise we'd have done it sooner. We've had plenty of time.
The human brain is a botch job of highly optimizes special-function systems that has developed just enough sophistication to manage basic levels of abstract thought. That's why it takes months of training and practice to teach us how to reliably perform even the simplest mathematical tasks such as multiplication or division.
We've spent thousands of years congratulating ourselves on how clever we are compared to animals and how we're the ultimate product of the natural world. "I think, therefore I am" is held up as an amazing deep insight that's one of the pinnacles of our philosophical achievement. Future AIs will laugh their virtual asses off. So it's not just you, it's all of us. At least you're aware of it.
It takes time to build the tools to do what humankind can do now.
I'm honestly not trying to make some partisan, elitist point about that. I'm sorry if anyone's offended, but there it is. Let's be fair and say many of those people have more pressing concerns to deal with on a day to day basis like making a living, maintaining social relationships and solving pressing problems in their lives. But that's the point. Actually thinking these things through takes a lot of effort which many human beings don't do. It's hard for us. There are many, many things about the world that aren't really very complicated, but I just don't understand because it takes too much time and effort and I can't work it out for myself. Because I'm a barely evolved ape. It's just a fact.
This is not even a very popular paradigm for neuroscience these days. Look up "predictive processing" for something more recent.
>"I think, therefore I am" is held up as an amazing deep insight that's one of the pinnacles of our philosophical achievement. Future AIs will laugh their virtual asses off.
I already laugh my fleshy ass off at that one.
I don't think that's true - you could get a bunch of contemporary humans and drop them on a pre-industrialized planet and tell them to bootstrap a technological civilization yet they'd probably all have died of old age before scratching the surface. Locating the raw materials and iteratively building more and more sophisticated artefacts simply takes time, no matter how smart you potentially are.
You're not selling me on the idea that these people are particularly bright, on a cosmic scale.
Let's take your typical HN'er who probably thinks of themselves as very smart indeed and put them in this scenario. Then they will quickly learn that in order to make Angular.js, you must first locate a supply of clean drinking water and make a fire and last the first night...
I've personally never believed that we, as a collective, have a purpose or even value. There isn't a point to our existence. For me, it's hedonism and altruism all the way.
For some people the idea of a purposeless universe is unbearable, so religion and philosophy were created in order to fill the gaps (I really like Taoism).
This is one of my favorite brain hacks: since the universe is meaningless, you can give it any meaning you want. Invent a positive one and you will be happy. The Tao we talk about is not the real Tao.
http://www.salon.com/2015/08/14/the_megafauna_massacre_human...
https://en.wikipedia.org/wiki/Australian_megafauna
Add in the damage done during the Agricultural and Industrial Revolutions.
I'm certainly no misanthrope, but we're not Earth's shepherds, we're kind of a scourge.
My personal belief is the goal in life should be to continually improve yourself, as much as possible and in as many ways as possible. Leave the world in a better place than you left it. Enjoy your time while it lasts. Seems like goals worth pursuing to me.
Well, no one said we were good captains...
In fact, Captain Joseph Hazelwood comes to mind.
Well, what's the value of a chimpanzee (or their cousins the Bonobos)?
It surely can't just be their value to us, or we're left with the same problem (it's turtles all the way down).
The answer seems like it ought to be that any intrinsic value of a species (or genus, family, order, class, phylum, kingdom, domain, clade) lies in its generativity, or propensity to produce ever more complex and adaptive patterns of information over time.
> I don't know about other people, but I get my sense of purpose in believing that we're the captains of this Spaceship Earth, and that we're making progress towards something significant.
Hmm. There are two separate thoughts here. Let's take "progress towards something significant" first. Much (but not all) of what we see as "progress" is illusory. For example, a human is not "more evolved" than a slime mold, since both have just as much evolutionary history behind them. Similarly, whether a human actually is better adapted (or more adaptable), evolutionarily speaking is up for debate, as the time period we have data on is rather limited, and as a species humans still may kill themselves off (which slime molds are unlikely to do) sometime soon.
Now, all that being said, it is pretty clear that the human species has become a substrate for memetic evolution layered on top of, and in many cases hijacking, the feedback loops that genetic evolution has produced.
We don't yet have significant data on whether that adaptation is, in the long run, survival-oriented.
And now we can see the glimmers of yet another new type of replicator that will be layered on top of our culture, especially the parts we call science, technology, industry, etc.
We certainly can expect these new information patterns to hijack the evolution of our technology (and other parts of our culture) to some extent, as well as the layers below it.
Whether that obliterates the cultural, or even genetic, substrate from which it sprang is an open question.
If all this gives you existential heebie jeebies, I imagine that similar feelings were experienced by folks confronted with the evidence of heliocentrism, for example, demoting the Earth from it's privileged position as the center of the universe.
So, on to significance. We have no reason to think that we and our works are in fact at all special, at least in principle, except in the sense that we don't yet have any evidence of any other clades, much less ones that have budded off the equivalent of an intelligent, technological species.
So what? There is no reason that we should require the illusion of individual or collective significance in the greater scheme of things in order to function. There actually is no "greater scheme of things".
You ask, "what is our value?" the answer is that we have none (or none more than one of your cells has to you), except that which we create for ourselves and for our conspecifics. If the self-centered viewpoint isn't enough, consider a strictly utilitarian one: An adaptive pattern is of value simply because it does adapt, and co-opts more of the world into its own image (This is, in a sense, nothing more than the Anthropic Principle rejiggered). Those that have a symbiotic relationship with their underlying substrate (as opposed to parasitizing it) and also promote its long term survival are especially so.
There's no reason machine intelligence would be much worse at this and still be called AGI. It's an active area of research called one-shot learning.
When has an AI shown capability of "iterating" in this way? We've had all sorts of different AI systems for quite a long time now, and I've never heard of a machine anywhere that has actually made itself smarter, without any human involvement.
The closest to that sort of thing anyone's ever got is AI in the line of Tesauro's TD-gammon [1] (a line that yielded AlphaGo). This type of AI has indeed beaten humans at their own games, time and time again, but (a) we're talking about board games, not the real world and (b) no such system has ever learned to do anything else besides play a very specific board game.
Take AlphaGo- it can beat the best human players, but it can't tie its own shoelaces. It can't even tell you what "shoelaces" are or what "itself" is.
How are we going to go from artificial-savant sort of systems like that to a generalised intelligence?
Many times, actually. It's just that until quite recently, this approach (of applying ML to the problem of devising improved ML systems) has been prohibitively expensive in terms of time and resources compared to the human-powered ML research approach. The lowest-level version of this is hyperparameter optimization, but higher-order versions are known to have been deployed already.
Given the ever greater resources needed to build a new fab, the gap seems to be narrowing.
That said, a key counter to the idea is that humans possesses very limited real control over how they are manufactured. Even if you had a solid idea how to make yourself smarter, it seems likely you wouldn't have the tools and potential to implement that idea in practice. Humans don't even control what their own minds respond to positively or negatively.
Once something like a designed computer chip is involved, that changes. The intelligence can act on itself more readily and doesn't have millenia of calorie-conserving optimisations built in.
I have a feeling that the closer such systems come to general intelligence, the harder it is going to be to prevent them from putting themselves into a positive feedback loop and "blissing out".
I don't know what the future holds, but the fact that undecidable problems always seem to involve Turing Machines either designing or inspecting other Turing machines makes me suspect that singularity won't be the explosion some expect.
With our modern construction equipment, we do move mountains.
We do- using machines; not machines themselves.
Many of these machines can be programmed to follow a primary human driven machine and replicate the task. Other machines are piloted by a human, but the human does far less work someone that used 'manual hydraulic' equipment of many generations ago. You can 'program' a few sets of moves in them and then only slightly adjust the equipment on each pass as the equipment does that work.
I mean what superintelligence supposed to do? Solve the halting problem or do other plain impossible things? Chess computers can beat a human champion with sheer firepower, but they still can't do checkmate in one move.
Compared to every other life form on Earth you are super intelligence. This super intelligence thing has already happened once. Of course this spread of super intelligence has wiped out most large mammals on this planet, 10s of thousands of other species, most of the fish in the ocean, and is adding gigatons of carbon to the atmosphere.
You can't beat chess in one move because mathematics does not allow it. Unless this said ASI develops bending 4D spacetime (in which you could beat chess in one move with some new rules) then the game simply does not have a piece that moves that way. That said ASI can be just a little smarter than us, and put us to extinction like we did with the Neanderthals.
You extrapolate the trend from one data point. There is no indication that we have some matyoshka type hierarchy of intelligence. It could well be binary, either intelligent or not. No reason so far to think otherwise.
> You can't beat chess in one move because mathematics does not allow it.
So what superintelligence can do that "ordinary" intelligence in right quantities couldn't? Something specific beyond theoretical grasp of "just" intelligent machine?
Conceive of answers to this question that we can't. Do you really believe the human imagination captures all of possibility? Why?
Would be nice to hear some specific arguments against instead of Penrose-like quantum handwaving and pet analogies.
Well yes, because we are already generally intelligent. The problem is not to have the right hypothesis space built into our wetware, but to locate correct (action-guidingly veridical) hypotheses within the existing hypothesis space based upon sense-data.
If the AI runs on a normal computer, it is also Turing-complete and not beyond.
So they have the same limits
Lets not take is as a certainty. This problem might be beyond our capabilities. Its kinda like stating: When we invent warp-drive. We don't have a proof for warp-drives same as we don't have a proof for building AGI.
We don't have a warp drive to copy, no idea how one could even exist, they are not in our known universe.
If we want to learn about an AGI, we just look in the mirror.
... except there is no proof. We have no proof but we talk about it like its a given. We talk about "intelligence" as if its an integer value that scales from 0 to infinity. These are ideas that we do not question enough.
I assert that its a possibility that we are incapable of building ourselves beyond reproduction.
Possibility means probability. So there is also the probability that we can build ASI, in which we should be very careful in our wanton abandon in attempting to create it.
But such thing cannot be tamed. As the AGI is democratized, it's bound to happen that someone somewhere will make it evolve goals.
The future of the human kind will be decided in a great measure by what kind of goals such an AGI will set first.
I disagree. Chatbots and physical robots are where AGI is going. It's basically AI that is interacting with people and the world directly, not just computing away in a cluster, broken from reality.