Smarter-than-Human Intelligence & The Singularity Summit

runamock writes "Brilliant technologists like Ray Kurzweil and Rodney Brooks are gathering in San Francisco for The Singularity Summit. The Singularity refers to the creation of smarter-than-human intelligence beyond which the future becomes unpredictable. The concept of the Singularity sounds more daunting in the form described by statistician I.J Good in 1965: 'Let an ultra-intelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultra-intelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind. Thus the first ultra-intelligent machine is the last invention that man need ever make.'"

Comments

Re:Actually, no.

[SoVeryTired]

Interestingly enough, man himself fits that description pretty neatly

Not necessarily

[Anonymous Coward]

What if the intelligence of the smartest thing you can design doesn't grow as fast as your own intelligence (i.e. the slope of the graph {x=designer's intelligence, y=intelligence of its best possible design} is less than 1)? Then it would never be possible to be smarter than a robot that's exactly smart enough to design a robot as smart as itself.

Of course...

[julesh]

'Let an ultra-intelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultra-intelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind. Thus the first ultra-intelligent machine is the last invention that man need ever make.'

Of course an ultra-intelligent machine might be smart enough to realise that designing and building a machine that's even smarter than it is a somewhat limiting career move.

I disagree . . .

[DodgeRules]

with the statement:

"Thus the first ultra-intelligent machine is the last invention that man need ever make."
since we will have to invent a way to stop the ultra-intelligent machines from destroying the inferior human race.

Yea right

[suv4x4]

I truly love how people see intelligence as some linear scale where right is "better" (genius) and left is "worse" (retard). But that's exactly why it'll be long before we manage to replicate true intelligence in a machine.

In fact things are far far more complicated, as far as inteligence goes and its utility in real world.

I'll quote Darwin roughly: "The strongest one won't survive, the most intelligent one won't survive. The one who survives, is the most adaptable".

In fact there's such a thing as "too intelligent". It's all about a careful balance of features an organism needs to possess to survive in a given environment.

In fact, if some AI threatens humanity since it considers itself far too intelligent, this may have quite unintended consequences even for this far superior mind, such as humanity get the hand of and nuking half the planet in attempt to lead "war against the machines", killing in the process any complex organism on the planet, ranging from biological to artificial.

And who remains in the end? Certain single-cell organisms which can thrive in a nuclear winter. Screw intelligence.

In fact any intelligent machine would realize it's again all about the careful ballance, and would cooperate with humanity and explore and learn from nature's development versus try to destroy it..

And since we have so shitty idea of what intelligence is, it's quite likely this AI will never be a true superset of the human brain but take on its own development, with potentially hilarious consequences.

I can't wait.

So easy a human could do it

[Ilan Volow]

Even when the ultra-intelligent machines take over, they will still need humans for Geico commercials.

I wouldn't worry about that just yet

[bloody_liberal]

With all due respect to those brilliant thinkers, I think we can learn a lesson from the first 50 years of AI - while it is clear that great things can be achieved with our new and magnificent computational tools (read: computers), I honestly think we are looking for the wrong goals, and as such there is no prospect (risk?) that machines will become truly intelligent any time soon.

Usually people consider cognition as essentially information processing. But here is a different definition (inspired by people like JJ Gibson and Varela):
cognition is the ongoing, open ended interaction with an unpredictable, dynamic environment. This capture, I believe, the essence of the human (and any other living creature) experience in the world, and excludes the computational experience.

We will have to build machines that are capable of open-ended interaction with an unpredictable world in order to hope and see any true sign of intelligence. Since very few are even trying to look in that direction (while most researchers are just looking for the awesome, and often lucrative, applications of our current computational capacity), I don't see any change coming soon.

The singularity has aleady happened

[A Pressbutton]

I have a slight problem with 'singulariries' as Kurzweil describes.

Assuming the ultraintelligent computer cannot do magic, it will be bound by the same physical and logical laws we live by.

An unltraintelligent computer may think 10x faster than us, but not qualitatively 10x better.
It will use the same basic logical steps to solve a problem, just faster and / or in parallel - and this may appear magical looking at the solution but if you sat down and examined the 'recipe', assuming it will tell you, it will be possible to follow the reasoning.

In some ways it could be argued that we have already passed some singularities, try properly understanding all the technology that goes into a modern car, the reasoning behind a mobile phone contract, the code behind ms-windows paperclip thing... well maybe not the last.

The operation of lots of well co-ordinated people working on a problem can act as a simulation for a 'more intelligent' intelligence. It seems a pity one of the achievements is a really good worm used for spam delivery.

Good's bad logic

[Flying pig]

Unfortunately, and much as I appreciate the work of I J Good, his statement about artificial intelligence is not valid. There are several things wrong with it

• It assumes that intelligence is well defined, which it is not
• It assumes that intelligence is the same thing as creativity, which it is not.
• It ignores resource limitations.

Dealing with these points in turn:

Intelligence is not well defined. It is very hard to say how much of what we call "intelligence" is in fact the ability to make many connections between facts stored in a very sophisticated memory architecture. Simply building a machine able to process information very quickly achieves nothing because, without learning and a social context, it does not know what information to acquire and process. In human experience, academically brilliant people often fail because they work on the wrong problems, or without access to necessary knowledge.

Nothing is actually achieved without creativity. We do not know what that is, or to what extent it is a social construct (i.e. it takes a developed society to have the necessary systems in place to translate an idea into a concrete reality.) And this leads onto the third point. It is no good having a highly intelligent, creative machine if its use of resources is such that it cannot replicate in large numbers. It may be that machine intelligence will ultimately replace human intelligence, but it may be that it will simply be too resource hungry. In effect, there may be a threshold of capability needed to solve some problems, and it may be that machine intelligence will run out of energy before it scales sufficiently to solve those problems. A machine society might, in effect, get stuck in the machine 19th century because coal or oil became a limiting resource. (In the same way, the energy and resources needed to be consumed to achieve a first independent space colony may exceed the total energy and resources available on Earth. It may be that a billion years or so of eukaryotic evolution has actually resulted in the optimum balance of intelligence, creativity and resource consumption, and that any attempt to exceed the present capability will tip us into declining resources faster than we can improve matters.

In many ways I hope this is wrong. But the argument that only one superior machine is necessary is, in fact, an inductive step too far. It is assuming that "intelligence" on its own can solve a class of problems which may involve a number of constraints which cannot be avoided - like the Laws of Thermodynamics, or the need for excessive amounts of energy.

We still have no clue how to do strong AI

[Animats]

OK. here's where we are:

• Logic-based AI AI looked so close in the 1960s, once it was realized that you could get a computer to do mathematical logic. All that was necessary was to express the real world in predicate calculus and prove theorems. After all, that's how logicians and philosophers all the way back to Aristotle said thinking worked. Well, no. We understand now that setting up the problem in a formal way is the hard part. That's the part that takes intelligence. Crunching out a solution by theorem proving is easily mechanized, but not too helpful. That formalism is too brittle, because it deals in absolutes.
• Expert systems Today, it's clear that they're no smarter than the rules somebody puts in. But back in the 1980s, when I went through Stanford, people like Prof. Ed Feigenbaum were promising Strong AI Real Soon Now from rule based systems. The claims were embarrassing; at least some of that crowd knew better. All their AI startups went bust, the "AI Winter" of low funding followed, and the whole field was stuck until that crowd was pushed aside.
• Neural nets / genetic algorithms / learning systems These all belong to the family of hill-climbing optimizers. These approaches work on problems where continuous improvement via tweaking is helpful, but usually max out after a while. We still don't really understand how evolution makes favorable jumps. I once said to Koza's crowd that there's a Nobel Prize waiting for whomever figures that out. Nobody has won it yet.
• Bayesian statistics Now used to do many of the things that used to be done with neural nets, but with a better understanding of what's going on inside. Lots of practical problems in AI, from spam filtering to robot navigation, are yielding to modern statistical approaches. Compute power helps here; these approaches take much floating point math. These methods also play well with data mining. Progress continues.

AI is one of those fields, like fusion power, where the delivery date keeps getting further away. For this conference, the claim is "some time in the next century". Back in the 1980s, people in the field were saying 10-15 years.

We're probably there on raw compute power, even though we don't know how to use it. Any medium-sized server farm has more storage capacity that the human brain. If we had a clue how to build a brain, the hardware wouldn't be the problem.

Re:Not quite ...

[Original Replica]

The more than human intelligence will inevitably entail compassion, love, and all the other emotions we have.

But look at how often we write off those emotions as a luxury. When "it's time to get tough" or time "to do what needs to be done" compassion and love go right out the window. Why would it be any different when we are no longer the apex of Earth lifeforms? Need to kill a few million humans to make way for solar farms, oh well, maybe we can keep a few alive on a special reserve somewhere. We humans with our compassion and love killed off how many species? We have enslaved and murdered other humans for how many thousands of years? These more-than-human machines had best be a hella lot better at compassion and love than we are, or humanity is going to hold the same relative place in the world order that Chimpanzees do today. I do not welcome our Machine Overlords.

Re:Not quite ...

[lekikui]

Intelligence is inextricably linked with creativity. I'd highly recommend Hofstadter's writings on the subject, in which he presents ideas of AI, not as a massive calculator, but as a collection of 'symbols', bashing into each other, with parts of the pattern modified by external state.

Think of a hyper-intelligent ant colony - any one ant can't really do much, but running about and interacting with the other nearby ants, they can organize themselves to achieve much harder tasks. Indeed, one of the sample dialogs in Godel, Escher, Bach is on that very subject.

Intelligence and creativity are high-level actions, you're still thinking of an AI as a massive collection of very fast low-level actions. That would be incredibly good at refining ideas, but a machine which can think would be different. It would run on a much higher level, making associations and fuzzy reasoning. You can't implement intelligence in formal rules, but you might be able to do it by specifying some formal rules by which certain objects interact, and then affecting a few of them based on 'external' state.

Read Metamagical Themas and Godel Escher Bach for some ideas of where I'm coming from (actually, read them anyway, they're both really good)

Re:Not quite ...

[kennygraham]

Which really makes a lot of sense. Humans show compassion. Lions, tigers and other less intelligent animals do not.

Correlation != causality. We're not compassionate because of our intelligence, we're compassionate because societies with compassionate members were better at having offspring that survived. That likely wouldn't be the case with these ultra-smart robots.

Sure, intelligence is a prerequisite to compassion, because it requires the complex ability to empathize. But it doesn't necessarily result from intelligence.

Re:Not quite ...

[Space cowboy]

"Compassion is the inevitable result of empathy "

I Disagree. Compassion is not inevitable. You're working from your own tenets and philosophies, a machine need not have those same ideals. Compassion is at least partially born of self-interest. The cynical (or non-empathic, if you prefer) view is that compassionate societies aid those who need it, because later the person previously aided may be able to render aid... "There, but for the grace of God, go I", "Do unto others as you would be done unto", etc., etc.

Are we suggesting that these hyper-intelligent machines would have any self-interest in keeping around the competition for resources that humanity represents ? I'm not trying to be trollish, here - I'm asking a genuine question. Humanity is ruthless in exterminating competing lower lifeforms. Why would we expect superior machines to be any different ?

And even should there be some self-interest in the first generations of such machines, what about the 5th generation, the 10th, the 1000th ? All I'm suggesting is that some thought be put into providing good answers for questions like this *before* we create competition. I'm as much of a technophile as the rest of you, but the phrase goes "look *before* you leap". Later may be, well, too late.

Simon