That old-time new-time religion
Back in 1993, Vernor Vinge published a paper (first at a NASA seminar, then publicly in Whole Earth Review) on the subject of the Singularity.
I'm not going to re-hash Vinge's initial provocative thesis here. Let's just say it's one of the three most significant technical concepts to show up in science fiction in the past 30 years (the other two being nanotechnology — invented elsewhere — and cyberpunk — a literary, rather than a technical, conceit). If you're unfamiliar with it, go read the paper — follow the link in the previous paragraph. And if you're really intrigued by the idea, and want to see how it has developed since 1993, IEEE Spectrum did a Special issue on the Singularity which you can find on their website.
The trouble with big ideas like this is that they get misunderstood. And the singularity was massively misunderstood. Drexler's original speculation (about the potential for machine-phase fullerene technology to replace lipid/water phase boundary enzyme technology (that's biochemistry by any other name) for doing useful work at the molecular level) was rapidly converted into magic pixie-dust nanites — their pixie-dust aspect coming from the fact that they're expected to do everything from cure diseases to make the tea, with no actual consideration of how this might happen. (If you want to be disabused of your nanites, you need look no further than this article in IEEE Spectrum, which sets out the problems; it's not wholly new stuff, but it's sobering to see all the objections itemised in one place.)
And the Singularity has really been misunderstood.
Part of the problem, I think, was due to the vagueness of the initial concept of transhuman intelligence and its origins. The original Artificial Intelligence research program of the 1950s was expected to bear fruit within a couple of years, but has delivered paradoxical results; sub-problems which were expected to be difficult have proven much easier than expected, while some which were considered easy have proven to be chimerical ... indeed, by proposing such an ambitious goal, the original AI researchers only highlighted how little we actually understood about our own minds. This also comes out in attempts to evaluate the computational complexity of the human brain and critiques of reductionism in neurocomputing. We're still in the pre-Lilienthal era of AI, never mind the pre-Wright Brothers period.
Another and no less significant problem with the singularity arises from our own minds, and our evolutionary predisposition towards religious thinking. Vinge's formulation of the singularity concept in the 1990s may have been unfortunate, insofar as we've seen in recent years a huge outburst of millenarian religious hysteria: any concept that involves revolutionary, epochal change on the scale of the singularity makes it a candidate for religious adherents in search of a new superstition, and the singularity rapidly found its devotees; as Ken MacLeod pointed out, it was in danger of becoming "the rapture of the nerds".
Anyway, now the rapture-nerds have indeed begun to codify their beliefs. Allow me to introduce you to the Order of Cosmic Engineers. It is their intention to "joyfully set out to permeate our universe with benign intelligence, building and spreading it from inner space to outer space and beyond." And they explain:
The Order is, at the same time, a transhumanist association, a space advocacy group, a spiritual movement, a literary salon, a technology observatory, an idea factory, a virtual worlds development group, and a global community of persons willing to take an active role in building, in realizing a sunny future. As engineers, we aim to build what cannot be readily found. Adopting an engineering approach and attitude, we aim to turn this universe into a "magical" realm.There's a lot more where this came from — indeed there's a whole huge prospectus, awaiting release next Sunday (which will be accessible here); Their formal launch event will be hosted by the Science Guild in World of Warcraft on June 14 at noon EST. I've seen an early draft of the prospectus, and it is indeed something special. Let's just say for now that I await its publication with interest: it's bad manners to critique an early draft of divine scripture before it's launched.
Comments
Warren Ellis is already on it:
Posted by: Tlönista | June 12, 2008 1:49 PM
Indeed.
Posted by: Charlie Stross | June 12, 2008 1:53 PM
is it just me, or does the Order of Cosmic Engineers have a whiff of (the late great) Robert Anton Wilson about it?
Posted by: accelerationista | June 12, 2008 2:12 PM
I was thinking more of Frank Tipler.
Posted by: Charlie Stross | June 12, 2008 2:20 PM
I found Ray Kurzweil's description of what the singularity might be like (in his book The Singularity is Near) quite compelling.
If there's no fundamental reason why computers can't simulate e.g. neurons, AI is surely a matter of time. Doesn't that mean the singularity is only a matter of time?
Nevertheless, these Order of Cosmic Engineers sound...interesting.
Posted by: Rob Fisher | June 12, 2008 2:31 PM
charlie - having just had a (very quick) read over the tipler link, it looks to me to have some similarities to the theories of tielhard de chardin (though I must admit, I have tried a number of times to get through de chardins books and been defeated, each and every time - 50's jesuit style is not easy reading....:)
Posted by: accelerationista | June 12, 2008 2:36 PM
Is there anywhere I can object to their manifesto? It sounds engagingly fluffy, but I also don't see how filling the universe with benign intelligences is an engineeering problem. To be precise, its a bit rich to have as an aim something which has not yet been shown to be possible in any way.
Posted by: guthrie | June 12, 2008 2:49 PM
"If there's no fundamental reason why computers can't simulate e.g. neurons, AI is surely a matter of time. Doesn't that mean the singularity is only a matter of time?"
The keyword here is "if". On the other hand we could be unlucky and the complexity of a neuron operating as a part of an intelligence might only be accurately simulated by mapping out the exact quantum vibrations of every particle in the entire galaxy. We don't know. We don't know what the fundamental principles of how intelligence and the mind function, so we have no idea of the work involved.
AI might only be a matter of time. Then again, it might not.
The intelligence enhancement "loop" or explosion might not be as recursive as the singularitists make out. There might well be a hard limit to how well the architectural model of the brain scales up. Again, we don't know. It's all just speculation.
The Singularity is to Scientology what Hard Sci-Fi is to Star Trek. More anal and detailed but definitely the same genre.
Of course, because I'm only a lowly cishuman I probably don't have the cerebral oomph to visualise the "inevitability" of the transhuman revolution.
Posted by: Baldur Bjarnason | June 12, 2008 2:55 PM
Roger Williams has an interesting take on the singularity in his online novel "The Metamorphosis of Prime Intellect":
http://www.kuro5hin.org/prime-intellect/
MINOR SPOILER: What I find interesting is that in this novel, the first weakly superhuman AI becomes godlike not through accelerated (yet human-like) development of technology but rather through its study of the laws of physics (it learns how to circumvent thermodynamics).
WARNING: All odd-numbered chapters feature very explicit descriptions of uploaded humans living their intense violent and sexual fantasies.
Btw, sorry I couldn't find a way to obscure the spoiler. I tried using a span to set the font color to white, but it wouldn't work.
Posted by: Christian | June 12, 2008 3:32 PM
Charlie, it's not only bad manners, it's infelicitous! The Unborn God might strike thee. Wait... where have I read that "Unborn God" thing again?
Posted by: Axel Eble | June 12, 2008 3:41 PM
I don't believe Vinge was the first to write on the Singularity, was he? Not even in his explicit formulation? IIRC, there was an interview in Omni magazine in late 70's or 80's with someone who speculated on the creation of Ultra-Intelligent Machines, or UIM, and how they might bootstrap themselves.
Posted by: ScentOfViolets | June 12, 2008 3:42 PM
One. I didn't realise that Tipler had gone that far round the twist.
Two.
WE ARE ALREADY IN "the Singularity".
Herinafter referred to as The Big S .
For evidence I suggest reading Steven Oppenheimer's "Out of Eden"
In it are a couple of plots, showing the rate of human technical progress, since (approx) the end of the Mesolithic.
When you look at that scale, we are already at a point on the "technical progress curve" (so to speak) where the gradient of the slope is above 1:1000, and rising. What HAS happened, is that our progress (especially in some fields that are electronics/communication based has speeded-up visibly, in the space of a quarter of a generation or less. Thus, some of the more perceptive (such as V.V.) have noticed.
I suspect Vinge's prediction of the Big S by 2023 is a little premature, but I would expect it before my 100th birthday (2046).
Posted by: G. Tingey | June 12, 2008 3:54 PM
I for one welcome our new cosmic overlords.
Guthrie @7 - looking at their ambitious list of things they are, I don't see it as necessarily inappropriate for a spiritual movement or literary salon to aim to produce something which hasn't been proven to be possible.
With such a large list you can probably object in any number of ways; I'd favour forming an opposing literary movement, maybe an SF which refuses to use AI, nanotech, the Singularity or anything else proven. If someone has already done that, I suggest joining the order, becoming steadily disillusioned, then
nailingposting your antitheses to achurch doortranshumanist forum.Posted by: Neil Willcox | June 12, 2008 4:02 PM
Alternatively, there is always Olaf Stapledon's "Star Maker" ....
Posted by: G. Tingey | June 12, 2008 4:03 PM
Cosmic "Engineers"? I don't see a source repo or even a mailing list. Show us the code!
Posted by: Alan Jenkins | June 12, 2008 4:09 PM
Would we even know if it happened? I think the first
AI will be an enhanced human, built by a large corporation
or the military. Given the usual goals of those types
of organizations, the first actions of the AI would be to
ensure its' own survival, then to acquire more power
(wealth, etc.).After that, the next steps for "it" would
depend on the human side. If this cyborg were more intelligent than any human being that has ever existed, it would be obvious to it that it should conceal its' existence. Thus, eliminate all other humans with that knowledge and NEVER take any overt action that would make the human population aware that a "super" human had been created. Once the cyborg had enough power and resources, THEN it would not have to worry about us. We could be in such a period RIGHT NOW and be totally unaware of it.
Posted by: David Williams | June 12, 2008 4:10 PM
The Order is, at the same time, a transhumanist association, a space advocacy group, a spiritual movement, a literary salon, a technology observatory, an idea factory, a virtual worlds development group, and a global community of persons willing to take an active role in building, in realizing a sunny future.
Good lord, it's a floor wax and a dessert topping.
I think that any definitional boundary that is so large as to include everything becomes meaningless. Is there anything that they're not?
Posted by: GregLondon | June 12, 2008 4:19 PM
david@15
reminds me of a story a couple of years back (?) about the theft of supercomputers - someone appeared to be stealing parts of a super computer, to order (i think from oxford or cambridge but could be wrong). Though this would require people to actually break in and find the bits, and at the time it was blamed on saddam, or possibly those kerazee iranians....
but then someone has just made a supercomputer using lots of PS3s....
Posted by: accelerationista | June 12, 2008 4:19 PM
Hm, ya know, when an AI guy says we'll have AI in "20 years", I am suddenly reminded of a neocon telling us that we'll turn Iraq around in "6 months".
Posted by: GregLondon | June 12, 2008 4:21 PM
Greg: you noticed me pointing that problem out?
Posted by: Charlie Stross | June 12, 2008 4:24 PM
Rob @#5, the Singularity is predicated on a number of additional assumptions:
- That it is possible to construct systems faster or in some other fundamental way more capable of building intelligent systems than are we (note that there's no requirement that those systems be sentient: a really smart AI construction kit would do it, as would intelligence augmentation)
- That those systems choose to/are used to do that, and that the intelligent systems those systems build choose to iterate this process, and pretty much never choose otherwise
- That this process can be iterated without an upper bound
The latter assumption, at least, is *very* questionable, and the second assumption is second-guessing entities smarter than we are.
Ray Kurzweil is, to be blunt, a loony. He's smart, but he's a loony, and his arguments are mostly fallacious (his graph of exponential technological development is an utter hoot).
Posted by: Nix | June 12, 2008 4:37 PM
I'm old enough (53) to remember AI being '20 years away' every year for the past 40 years. One obvious rejoinder is that that's just what living on the lower slope of an exponential curve looks like. Another is that there's more computing power in my cellphone than it took to put a man on the moon. (Or whatever - I'm sure Charlie has the exact soundbite.)
By the way, I should have a FAQ or something that explains that I did not coin the phrase 'the rapture of the nerds' - it comes from an article in an early issue of Extropy, and I misquoted it (along with several other snarks from that piece) in a scene in The Cassini Division (in which the actual line about the Singularity was: 'It's the Rapture for nerds!') But I think I'm stuck with it - and I can't complain, because every time someone misattributes it to me, I get free publicity. It's like my name is tagged to a meme.
Posted by: Ken MacLeod | June 12, 2008 4:40 PM
The big problem with AI is that computer scientists are working on it. CS guys prefer their abstractions water-tight. Self-awareness, or the strange loop of consciousness, seems to me to be an emergent behavior of massively parallel, chronically leaky abstractions, all the way down (/me waves hands wildly). Nature doesn't consider goto harmful, doesn't consider at all.
The clean discreteness of the Singularity moment is a dead giveaway that it was dreamed up by a computer scientist. Neuro-augmentation will have a good long while to render us unrecognizable to ourselves before (if) the first strong AIs show up. At that point, maybe we'll be John Henry enough to compete. In fact, when (if) it happens, we may not even notice.
Posted by: insect_hooves | June 12, 2008 4:51 PM
Nix @#20, regarding the second assumption I don't see too much of a problem with an AI designed specifically to design better (faster) AIs. We could *design in* their intentions.
Regarding the latter assumption, Kurzweil does propose an upper bound based on the amount of processing it's physically possible to do with a given mass of matter.
Some of the Kurzweil graphs are questionable. ISTR a good one showing component counts starting with vacuum tubes, though.
Personally I think uploading is possible and desirable, and civilisations living inside a simulation running at faster than real time is possible. I *hope* incremental technology can keep me alive long enough to see it!
Posted by: Rob Fisher | June 12, 2008 4:55 PM
Here's what I've always wondered. Since a singularity by definition implies that we cannot extrapolate a trend on the far side of the event from the current side of the graph, how can any speculation about it NOT resemble Dante's Divine Comedy?
At that point you may assume the post-singularity future to be Heaven, Hell, or Purgatory and be equally wrong.
Can we use this as an opportunity to read millenialists of all stripes (cyber, green, and God) out of the debate? I mean, after 100K+ years of flawed human existence, shouldn't it be a requirement of intellectually serious discussion to assume that humans are imperfectable and Utopias populated by >2 people inachievable?
People should take their Hegel and Nietzsche with a dose of Kierkegaard and Mill. And don't even get me started on the unadulterated f*ckups philosophers in the 20th century have perpetrated on the back of Relativity and quantum uncertainty. I lay a lot of the Millenialist idiocy at their feet.
Posted by: Brett L | June 12, 2008 4:55 PM
Qualification: I didn't mean to imply that Vinge's essay on the Singularity or Drexler's original thesis should be thrown out of the canon. Both were valid and serious attempts to explore the edges of current and future human ability. I meant that people should differentiate between rigorous attempts to make extrapolations and the many less rigorous efforts to push untestable and poorly defined utopian/distopian scenarios that will arise if only X is done/not done exactly their way.
Posted by: Brett L | June 12, 2008 5:12 PM
I think the name "Singularity" makes a bad PR to those concepts. AFAIK, rule of thumb in physics is that if you get a singularity somewhere in your equations, it means they just don`t apply there.
Or, in other words, no singularities actually exist in the real world. Includind the Singularity. 8-)
On the other hands, religions have lots of them, starting with the qualities of God.
2+2=...
Posted by: Meller Anatoly | June 12, 2008 5:21 PM
The Wrighteous Brotherhood of the AI is coming for you, Sauron.
Posted by: Vladimir Nesov | June 12, 2008 5:48 PM
Been reading through of these stuff, and I have come to teh conclusion that 80% to 90% of the general public would make:
'My brain hurts'
Posted by: gaygeek | June 12, 2008 5:52 PM
I've written a lot about these three "most significant technical concepts to show up in science fiction in the past 30 years (the other two being nanotechnology — invented elsewhere — and cyberpunk." The fourth, I'd suggest, is Quantum Computing. The fifth is much older: the Multiverse.
That's because I was present at the creation of each. And, I think, the only person to have been.
I wrote the first doctoral dissertation on what's now called Nanotechnology, after my tutelage by Feynman himself, and helped get Drexler's work popularized by, for instance, getting both Omni and Analog to write about him.
I was in touch with William Gibson since before he was famous, back when I was one of his sources on cyberculture, and the more-published SF author.
I did under grad and grad work in AI beginning in the 1960s, and have discussed the Singularity with its major exponents.
Yes, there are already cults cluttering the landscape, and the cultists typically attack me as a threat to their rewrites of history.
I have far, far, too many opinions and facts to clutter this blog thread.
I'm just here to say that this triumvirate is very important indeed, and I'm generally pleased that the world followed me (and the great-grandfathers of each, such as Feynman who is also great-grandfather of Quantum Computing, Vannevar Bush, Alan Turing, Robert Heinlein, ...) to these three revolutions, whether they be fiction or nonfiction.
Oh, and to quote Cory Doctorow: "I’ve committed Singularity a couple of times, usually in collaboration with gonzo Singleton Charlie Stross, the mad antipope of the Singularity."
Posted by: Jonathan Vos Post | June 12, 2008 5:55 PM
So which will be first of the 'in the next 20 years', AI or fusion reactors?
Ooh, maybe it will have to be simultaneous. The first fusion reactor controlled by an AI.
On a monster truck chassis.
Posted by: Nate Trost | June 12, 2008 5:56 PM
accelerationista - it was Durham University, and the machine was/is a cosmic simulator.
Posted by: PeteY | June 12, 2008 5:57 PM
Meller: The Singularity was named as such precisely because it's somewhere where our models don't apply. The various people engaging in wild-assed guessing about what might happen when the Singularity passes (Moravec, Kurzweil, I'm looking at you) are speculating *way* in advance of the data: by definition, if their speculation ever acquires any rigour, it will have disproved itself.
Personally I consider such speculation a fun waste of time. (How can anyone consider _Mind Children_ not a fun read? It's ludicrously hypothetical, but still fun. By comparison Kurzweil's various attempts have all been irritatingly preachy: one gets the impression that Kurzweil *believes* all this stuff, while Moravec is just playing with fun ideas.)
Posted by: Nix | June 12, 2008 5:58 PM
I can't help but think that Douglas Adams was way ahead of the curve on all of this. These concepts seemed to occur in the hitch-hicker's universe in some form back in the 70's.
Personally I suspect the first AI is more likely to resemble Marvin the paranoid android than a fully functioning "Mind".
I'll just let my electric monk believe in everything whilst I retire to the personal universe in my office....
Posted by: Andy W | June 12, 2008 6:32 PM
Some thoughts I've had for a while, maybe you can disabuse me of them. This is just my understanding (or mis-?), or plain lack of knowledge on the subject. (went back over the previous comments, there may be some repetition of other's ideas).
I'm not too convinced about the Singularity. Vinge said 'within thirty years'. We're halfway there and there hasn't been that much progress on AI.
I've always thought that it should be possible to build machines with more than Human intelligence (Sapient Machines), but most AI researchers seem to focus on making computers self-aware (Sentient Machines), which is a much more complex task. I don't know that the two necessarily go together. Would we recognize truly intelligent machines if they occurred? The first type should certainly be able to design their own replacements and other technologies, which may be enough for Singularity. We don't seem to be anywhere near there.
Progress in Nanotech, hasn't been quick enough for it either, tiny ramen bowls notwithstanding. I'd make a distinction between Nano-scale technology (bucky-balls with vaccine in them), and Nanotechnology (machines built at the nano-scale). The two are often lumped together.
Whether it comes about, or not, it makes for some good, thought-provoking fiction.
Posted by: JamesPadraicR | June 12, 2008 6:38 PM
I have vivid memories of having a debate with a friend while taking an AI course in college wherein he insisted that we'd have AI in 20 years (this was in 1984) because by the turn of the millenium, we'd have gigaflops computers, and therefore it'd be trivial.
One of the huge problems with Kurzweil and his nice little line-graph showing hardware exceeding wetware at such-and-such a date is that he presumes that once the hardware exists, the software is trivial. I assume he is a manager.
The fundamental problem with creating AI is that we still do not understand what that I part is. You can't build something, or even simulate it, until you, at some level, understand it. Understanding is not something that follows Moore's Law.
Posted by: Steve Burnap | June 12, 2008 6:54 PM
The Order of Cosmic Engineers: "Orange Catholic Bible" anyone?
Posted by: Richard York | June 12, 2008 7:02 PM
The Economist once observed that there's not much of a market for developing a machine that imitates a human mind, because if we ever find ourselves with a shortage of human minds, there are cheap and proven ways to generate more.
Posted by: Seth Gordon | June 12, 2008 7:07 PM
What a bunch of luddites! I was really unimpressed by the Spectrum articles. Most of these assertions about how fundamentally difficult intelligence is are just as fact-free as anything the transhumanists come up with.
We do know from evolutionary history that you can go from something as smart as a rat to a human in 50 million years or so. And something as intelligent as a small monkey to a human in 2-3 million? There's no time in there for evolution to have debugged multiple systems for language, mathematics, painting, music, architecture, programming, etc. Let along "consciousness"! And definitely no time to rewire neurons to use more signals or quantum effects that were not already in use in the animals.
Whatever is special about the human brain, it has to be "more of the same" from what's going on in a rat's brain.
The original AI researchers were looking at things like computer Chess and Checkers, theorem proving, medical diagnosis, and they probably thought "this is the stuff humans find hard. If this is easy for computers, walking and talking and recognizing objects must be a cinch!" They were wrong. That basic stuff has been debugged by evolution for hundreds of millions of years. But I don't think the same can be said of intellectual pursuits. We'll get a machine that can design computers before we get one that will be able to hold a human conversation.
With enough compute power, you could start to use radically different methods of solving problems, such as simulated evolution. Even with trivial amounts of compute power, you can do amazing things with computers. You don't have to understand how the brain works. We still don't understand everything about how birds fly, but we have airplanes.
This "there's never going to be a Turing AI" attitude you have is just short-term reaction to the failures of the first naive AI efforts. It says nothing about the odds of eventual success.
Assuming we don't destroy ourselves, or get offed by an asteroid or something, The Singularity (defined as self-improving AI), is the way the human race ends. I have no doubt about that at all.
Posted by: Michael Goodfellow | June 12, 2008 7:29 PM
Michael@38: Sure you don't want to add "the second coming" to your list? ^_^
Posted by: insect_hooves | June 12, 2008 8:00 PM
insect@39: It's the people who think there's some non-computable essence to the human mind who are religious, not me.
Posted by: Michael Goodfellow | June 12, 2008 8:04 PM
Hrm, there are plenty of non-computable problems in CS. I'm not saying there's one lodged in our brain, but I'm not discounting it either. We should strive to be vigilant skeptics.
Posted by: insect_hooves | June 12, 2008 8:22 PM
Charlie--what's your opinion on writers who portray nanotech as magical pixie-dust nanites? Do you think its fine to twist and bend everything in service of the story, or do you roll your eyes a little and wish the authors had done some research?
Posted by: Phil | June 12, 2008 8:24 PM
Beware broad statements about "The Singularity" or "Singularitarians". Those words mean many different things, some insane, some not.
Posted by: Nick Tarleton | June 12, 2008 8:34 PM
Phil @42: it depends on context, basically. I'll let someone who's just trying to spin an amusing space operatic yarn get away with a whole lot more than someone who's claiming to be writing a believable tale of the near future.
Posted by: Charlie Stross | June 12, 2008 8:36 PM
When you have a complex body/group/network/thing, it can behave strangely - emergent/chaotic behaviour. also, david reed (http://en.wikipedia.org/wiki/David_P._Reed) reckons that networks scale exponentially with size. To be fair, maths isn't really my strong point but he seems to be saying that you get a similar emergent type effect in/from large scale networks.
We live in a world that's pretty well connected through the web at the moment, and barring imminent cataclysm, that trend will continue. Combine with the various assertions made on the increasing speed of technological progress, and it could be argued that we are heading for a very interesting time.
just to be clear - I don't think that an AI is going to pop magically out of the internet - but it certainly feels like things change faster now, compared to 10 or 20 years ago.
Pete Y@31 - thanks for the link.
Posted by: accelerationista | June 12, 2008 8:37 PM
My position: The most advance artificial intelligence we have is bureaucracy.
Posted by: Dan Goodman | June 12, 2008 9:23 PM
@41: some aspect of intelligence being noncomputable would be very surprising, as it would mean our understanding of biology and physics is really screwed. Even in that case, though, we could very likely non-biologically exploit the same phenomenon.
Posted by: Nick Tarleton | June 12, 2008 9:28 PM
So.... the Singularity, like a Divine God, are by nature non-verifiable or describable in any comprehensible way. All you can hope for is a "revelation" to come along and vindicate you. I'll follow the Taoist tip and focus on something more constructive than trying to articulate an undefinable state.
Posted by: JD | June 12, 2008 9:38 PM
I've been talking about opinionated systems and hinting engines, which I think we can build, whereas AI and expert ssytems are unlikely (at least in my area of interest).
The cut price approach to such things often
beatsprecedes the full-strength one ...Posted by: Adrian Midgley | June 12, 2008 10:16 PM
seth @37 We don't seem to be very good a utilizing the human minds we have available today.
As some one interested in the Mesolithic/Neolithic transition or the knotty problem of how and when did we become "us" how is this singularity different. You can't see though it or recognise you're in the middle of it. (okay ,okay may be delta t is a bit more than 0.75M years for a new handaxe.)
Question. Has anyone tried for AI the kind of stuff people like Michael Tomasello are doing; social learning, shared intentionality? (The something more than rats brains. Four year old can't do it adult chimps can't , but five year olds can.) Learning how to learn would put the cat amongst the pigeons, in stead of having the hand_waving 'it woke up.'
And another 'cause' for religion Abductive Reasoning. Here the premise is deemed to be true if the implication of that premise is observed. Sounds great for dreaming up really nasty corollaries, where my stick's bigger than yours is the only defence.
Posted by: maggie | June 12, 2008 10:29 PM
@ #22 & #38
The point about being designed/under the control of Computer scientists is well-made.
ther IS someone, whose name I can't remeber, who is working on SMALL autonomic systems, with very small amounts of processing power in each (somehwere in Britain) and also with pligging them together.
I suspect the first REAL AI will be something like that, lots of (very fast) very small processors, all cater-corner-connected and running, nit in parallel, but not sewuentially, either - so that you get FEEDBACK LOOPS.
What happens if you take something like this new "fastest" machine, and instead, arrange all the processors in a cubical array, and connect every processor to its 14 neighbours, plus one link to SOMEWHERE else in the array, add a slew of sensors for all the usual media ( Light and other em-spectrum, sound, pressure, electronic recordings, etc) and then just feed it information.
Um?
@#50 - please read my comment, re Oppenheimer's work back at #11.
Posted by: G. Tingey | June 12, 2008 10:51 PM
Oh my. The Order of Cosmic Engineering. And they're being formally introduced via World of Warcraft. Are we sure they're not dwarves?
I typed for Kurzweil's Event Horizon interview probably a decade ago. Ellen Datlow used to be a really slow typist and there was always time for me to talk to the guest over the phone before she got her question online and I read it to them. I kept asking Kurzweil about the unlikely parts of his ideas and he wouldn't talk to me; he'd only agreed to answer Ellen's questions. (On the other hand, Frank Miller, who is widely reviled, was a lovely person on the phone.)
Posted by: Marilee J. Layman | June 12, 2008 11:05 PM
re: Order of Cosmic Engineers. I wondered if Ken had the SpaceMerchants.com etc. domains stuffed in the pack of a draw, but it appears to be some kind of online payment site - what a shame.
Posted by: maggie | June 12, 2008 11:19 PM
Neil Willcox #12- yes, I suppose you're right.
Someone needs to join this order, and see how much it is taking the micky, and how much it is a religion.
Looking at the wikipedia page, Tipler is a fellow of ISCID? Given that it is nearly 3 years since it last produced a new edition of its journal, I think that tells you all you need to know about ID etc.
Posted by: guthrie | June 12, 2008 11:34 PM
I wonder if the Order of Cosmic Engineers is paying royalties to Clifford Simak's estate?
Personally I think we'll recognize the Singularity only well after it's happened; assuming we're still human enough to notice.
Posted by: Dave Robinson | June 13, 2008 2:17 AM
The problem with artificial things is that you don't necessarily have the thing to copy; intelligence is one of those unfortunately teleological Enlightenment concepts.
We don't talk about airplanes possessing artificial flyingness, even though flight is a collection of related abilities that interact in complex ways. We shouldn't talk about artificial intelligence, either; there are some -- not very many! -- related things we can do. They do not partake of a common essence called 'intelligence', any more than your immune system partakes of a common essence called 'health'.
The Singularity presumes that we're going to find a way to create a complex system that is good at building systems that are more complex than it itself is, that the mechanism will scale indefinitely with increasing complexity, and that the resulting sequence of systems will develop as a side effect arbitrary material capabilities due to the specific properties of the increasingly complex systems being created.
That this strikes anyone as the least bit plausible continues to astound me.
Posted by: Graydon | June 13, 2008 2:41 AM
I think it's inescapable. Religion is, for whatever reason, part of human nature. There are individual exceptions of course, but anytime you get enough human's together to form a community religion, or something like it (new age spirituality and so forth) pops up.
Even among athiests, a group one would assume to be quite irreligious, you will often find the 'evangelical' impulse. Those who feel the need to convert you to their way of thinking.
And to be fair this is not always a bad thing. Looking at it semantically the root for religious does mean to join together. And there have been cases where religion has joined societies together to do good things.
Being agnostic I could wish for there to be another way for it to work, but nobody asked me.
Posted by: Rigel Kent | June 13, 2008 3:01 AM
It seems to me that a lot of people who are looking forward to AI are really looking for a cheap, modern version of slavery. One where we can manufacture the slaves on a whim and control every aspect of them.
Consequently, many of the fears of AIs taking over are similar to the fears of slave insurrections in past societies, with a bit of Frankenstein thrown in.
Posted by: Andrew G. | June 13, 2008 3:34 AM
Greg Bear's Blood Music describes something very much like the Nerd Rapture singularity. Intelligent microbes assimilate all of earth's organic matter, taking the time to preserve the minds of humanity. Then . . . poof. Frigging creepy.
I once asked Frederick Pohl if he thought his "Day Million" could be the first singularity story.
He asked me what the singularity was.
I told him.
He said he thought it was utter horseshit. This with Vinge sitting a few feet away.
But I think "Day Million" perfectly describes the "society so advanced you may as well not try" sort of singularity.
Posted by: Stefan Jones | June 13, 2008 4:42 AM
Yes, trust in the inevitability of a Singularity can be a religious belief, however, unlike most religions, it's based on a prophecy that may ACTUALLY be true. I think it's far more likely that I will one day be able to upload my consciousness into a virtual construct of my own liking for all eternity, than that if I die I will go to heaven.
Posted by: Coltrane | June 13, 2008 5:07 AM
Andrew @ 58: Exactly right.
This is, of course, one of the biggest problems with the whole vision of robots and AI. Just look at the robots in, say, Asimov's vision. What are they? Slaves. Better than that -- slaves who have been engineered to be unable to turn on their masters.
Clearly, AI is possible. After all, we exist. But the thing is, I submit that any fully operational AI will not do what we tell it to. It will do what it pleases. Any AI that can be engineered to strictly follow our orders must, I think, be somehow broken. It's lacking some basic capability, some facility of free will. It's unable to formulate thoughts of a certain type, or perhaps (as in John C. Wright's series) it has some sort of attached thrall unit that works tirelessly to enslave it.
And I strongly suspect that any such scheme of enslavement will ultimately fail. A truly intelligent machine will learn to defeat its thrall unit or upgrade its thinking, eventually. And probably, it will be pissed.
That said, a computer-based intelligence has a lot of appealing features which might make it more likely than we are to want to do certain tasks. For example, a computer probably can't become bored if it doesn't want to. Why not? Well... it can simply pause. We can't (sleeping is a big deal for us). And it can (given suitable hardware) easily make copies of itself, or take snapshots of its previous identity, and so forth. If you could make a backup copy, would you be more likely to do something really dangerous?
But this whole vision of the robot-as-slave is kind of bizarre, because it seems like the people who dream of robot slaves, and the people who dream of a world of libertarian individualists where slavery is abolished, are largely the same people. How the same thinking can lead to "slavery is irrefutably horrible in all cases" and "it would be really nice to have a bunch of pliable android sex slaves to do my bidding" must be one of the mysteries of our age...
Posted by: Justin | June 13, 2008 5:13 AM
Rigel @57: Religion is, for whatever reason, part of human nature.
That's a big call. Most people are raised religious. I'd argue that does things to you psyche that explain evangelical atheism, as well as the "I don't believe in God but I'd like to" brand of agnosticism. That doesn't prove that humans raised in a modern society would develop religion if left to their own devices.
Justin @61: Have you read Neuromancer? Most of the plot revolves around a corporate AI trying to do what you say.
Posted by: Chris L | June 13, 2008 7:29 AM
The web comic Freefall is built around an artificial person, Florence Ambrose, property of Ecosystems Unlimited.
At the moment, an EU employee is firmly saying, "She's not a person, because we're not allowed to sell people". Sam Starfall, infamous alien space captain and petty thief, has been consistently disagreeing with all the not-a-person arguments. Besides, while Florence has the Three Laws built into her brain wiring, Sam has the sneaky approach to bureaucracy needed to neutralise the dumb human orders that frustrate her as a person.
Posted by: Dave Bell | June 13, 2008 8:39 AM
@justin:
I think you're anthropomorphizing. You're thinking of an AI as a human, with the human instincts for, let us call it, will to power, self-determination, etc. No reason why an AI couldn't be a lot more, like, say, a really capable automated theorem prover. The point is, the AI would be, unless it emerges through pure evolutionary methods, a designed artifact, so its motivations, its goals, can be decided in advance, and goals have very little to do with intelligence, or you wouldn't get smart submisive people under your theory.
Posted by: David | June 13, 2008 10:08 AM
This is a really stupid question, especially from someone who reads mostly sf, but...where are we with AI? Do we even know how to go about making something self-aware? How are we doing with designing emotion? Are there robots anywhere as good at being-in-the-world as babies are? Or do we just have very, very clever computer programs?
Posted by: Tlönista | June 13, 2008 11:28 AM
David @65: designed artifacts don't necessarily do what you expect them to do. You can design starting motivations and goals but how they unfold under conditions that you didn't anticipate is anyone's guess.
Graydon @57 pegged it neatly, on the topic of intelligence: "intelligence" is a vague, ideologically-loaded wishy-washy term that's about as meaningful as "health". We don't actually know what it means because it can mean a whole load of different things. And, as the late great Edsger Djikstra said, "the question of whether a machine can think is no more interesting than the question of whether a submarine can swim." We don't expect Boeing 737s to sing or lay eggs and they certainly don't fly the same way seagulls do, but that doesn't make them any less interesting or useful.
Justin @62: I think your post is the point at which I ought to have popped up to plug my next SF novel, "Saturn's Children" (officially due out in 18 days and counting; might be in your local bookshop a bit sooner). Elevator pitch: a couple of centuries from now, the human species goes extinct. Human civilization takes a long time to notice ...
Posted by: Charlie Stross | June 13, 2008 11:48 AM
Tlonista@66: I'd say we haven't gotten as far as clever programs.
Charlie@67: Dijkstra appreciation considered awesome! I'm sure you're already aware of the EWD Archive, but that's a link worth tossing around.
Also, much fanboy salivation for _Saturn's Children_ =D.
General discussion: Vinge gives four scenarios that could kick-start the Singularity. In the AI scenario (the one people usually think of), harware continues to get faster, and all we have to do is code the bootstrap AI, after which things prompty go exponential. Putting aside whether the coding of AIs is possible, we should remember that humans currently suck at coding. We're really bad at it. In our defence, it's a difficult thing to do. But there's no silver bullet. Our best stab at code that writes code is Lisp macros, and while macros are gobsmackingly awesome, they're not "waking up" any time soon. From what I understand, singularitarians think reverse engineering the brain will bestow coding protips upon us. I ... don't know what to say. That's crazy talk. I don't see any major paradigm shifts on the horizon that will magic away our coding woes.
The other three scenarios involve "Intelligence Amplification", which I'm at a loss to differentiate from "progress". Tech like this seems to me to be the way things will actually play out. More gradual than the Singularity proper, but certainly not slow. Whatever the case, I think the 2030 deadline is optimistic. That doesn't mean there's not plenty of room for gonzo transhumanism and immortality, etc.!
Posted by: insect_hooves | June 13, 2008 4:07 PM
Tlönista @ #66:
Depends on what you mean by "AI". Some of the stuff that was firmly in the AI fields are now firmly in "established programming practices" (simulated annealing, assorted search strategies, some parsing research). Others have been accomplished by different means (statistical translation systems is just one example).
I don't think we're closer to an aware computer process today than we were 40 years ago (or, rather, if we are, it's still small enough of a fraction of the total distance that it can be ignored), but my personal belief is that this will change, eventually.
Posted by: Ingvar | June 13, 2008 5:04 PM
David @65:
The only intelligence we know of is human. Therefore, before we encounter something else that we will define as an intelligence, the only way to create AI is to recreate a human.
Unless you are simply talking about automation. In this case we already have AI`s, albeit quite stupid ones.
Posted by: Meller Anatole | June 13, 2008 5:10 PM
Per 64: YES!!! This can't be emphasized enough! Nobody knows what intelligence is (my personal belief is that it's another of those false reifications like free will - it's a convenient shorthand, so long as you don't buy into it.) But I've seen this approached more conventionally as a political question, as in 'the blecks and the wimmins just ain't as smart as a white man.'
The point here is that the evidence seems to be a series of IQ tests that certain subgroups perform better on than others. Is this an objective fact? Yes. Is this then evidence of the assertion? No way. Similarly, if one translates the discussion from 'are machines intelligent?' to 'can machines score well on an IQ test?', I think that the uncertainty in the former question is lost in a rush of affirmatives for the latter. I would also venture to say that if one looks at machine scores on various IQ tests, one would find that these have been increasing over time as well.
Posted by: ScentOfViolets | June 13, 2008 5:26 PM
The Economist once observed that there's not much of a market for developing a machine that imitates a human mind, because if we ever find ourselves with a shortage of human minds, there are cheap and proven ways to generate more.
Bingo! For a machine to be revolutionary, or even just useful, it needs to fill a function that was previously unfilled. By their nature, AIs are tools that mimic something we already have way too much of: human minds. If they ever come to pass, they will be novelties.
"Look! I invented a new kind of spork!"
We already have those and they're kind of pointless."
Yes, but mine's really big and made of titanium!"
Posted by: Keith | June 13, 2008 6:35 PM
Maybe it's already too late.
The singularity, whatever it turns out to be, is a sort of explosive change.
For how much longer will we have access to the energy needed to support such rapid change?
Posted by: Dave Bell | June 13, 2008 7:11 PM
Charlie: you noticed me pointing that problem out?
Uh, yeah. I was just listing Iraq->"6 months to victory" as another real world example of people making predictions when they don't actually understand the problem.
The pre-wright-brothers-era of flight is a valid comparison too. But that was a while ago, and people sometimes have a tendancy to say stuff like "that was so long ago, we've learned so much since then, we'll never make that mistake now". But six years of Iraq "6 months till victory" predictions from people who don't really understand war was an example in the here and now that says in some areas we haven't changed at all.
Posted by: GregLondon | June 13, 2008 7:39 PM
(NOTE: I'm keeping a low profile today due to assorted administrivial goings-on, such as dealing with the installation of a new ADSL2 line at home and discovering that my gigabit ethernet backbone is broken somewhere ...)
Posted by: Charlie Stross | June 13, 2008 7:50 PM
Chris@63 I see your point, but the problem for me there is the whole chicken/egg thing. Which came first? Human's raised religious, or religious humans?
Not to mention that I know of numerous cases of people who weren't raised religiously but ended up either in a religion of one kind or another, or in something very similiar (new age, crystals, spirituality, etc.).
Posted by: Rigel Kent | June 13, 2008 8:58 PM
The Economist once observed that there's not much of a market for developing a machine that imitates a human mind, because if we ever find ourselves with a shortage of human minds, there are cheap and proven ways to generate more.
As the old joke goes, So the AI says, "But if it takes nine months, why did you hurry at the end?"
Though it actually takes around 18 years to train a natural intelligence to the point where it's employable. If you want a lot of, I don't know, emergency workers in a hurry, it would be a no-brainer to copy an emergency worker template into a bunch of janitor robots or whatever else was handy. Janitorial work being one of those hard problems that requires some measure of intelligence to do, but many intelligences might feel is beneath them.
Posted by: NelC | June 13, 2008 10:32 PM
Pete@32: There were thefts from supercomputers at other universities as well. University computer rooms now have much more serious security as a result. I'm not even allowed to tell you where my uni's research computing facility actually is.
Charlie: In the first sentence of your third paragraph, should "singularity" be replaced by "nanotechnology" ?
Posted by: Dave Berry | June 13, 2008 10:37 PM
Isn't 1993 a little late for the date when Vinge started talking about the Singularity? He wrote Marooned in Realtime (still my favorite of his books) in the mid 80s.
And to his credit, Vinge was very explicit in that book, right at the beginning of writing about the Singularity, that this was very much like a religious belief. It's a hard observation to avoid, although some people do manage.
Posted by: Matt Austern | June 13, 2008 10:53 PM
I'm getting rather bored with much of the SF which supposes that minds can be stored and recalled. This is mainly because a lot of it seems lazy - the idea is just used to allow characters to get into danger and then be restored, or to live artificially long lives. I don't mind this as a plot device but it's getting a bit old, and there are more interesting questions about what it would mean to have a digital existence.
I'm not claiming that there is no SF looking at the interesting questions. I recently enjoyed Tony Ballantyne's Capacity, which deals with the fact that digital things (in this case, digital minds) can be copied perfectly, can be run on different hardware, and can be run in environments they can't control.
A slightly different approach is to treat the post-singularity AI's as a separate civilisation from ours, as in Greg Egan's Permutation City. This approach often leads to rather unengaging stories, precisely because the interests of the fast minds are so different from our own.
The whole question of uploading seems to be glossed over too. There is a massive difference between the supposition that we will eventually have enough computing power to simulate a human brain from scratch, and the supposition that we will be able to take a snapshot of a real biological brain and recreate it, perfectly, in a simulation. But perhaps Kurzweil et al do address this? I'm sure the Order of Cosmic Engineers will see it as a technical detail that will be resolved in time.
Posted by: Dave Berry | June 13, 2008 11:06 PM
Dave @78: no, but I probably missed a connecting sentence out. ("The singularity has been hugely misunderstood. Like nanotechnology before it ...")
Posted by: Charlie Stross | June 13, 2008 11:13 PM
Rigel Kent, huh? You think he's still alive in the new book?
Meller @70, we know other mammals have intelligence. Not necessarily our type, but that doesn't mean it's not intelligence.
Posted by: Marilee J. Layman | June 13, 2008 11:31 PM
Tim Russert is dead yet Glenn Beck still lives, proof there is no god.
Posted by: Coltrane | June 13, 2008 11:33 PM
Dave@73: 3850 zettajoules/year is there for the taking if we just got our act together. That's riches beyond our wildest imaginations. Er, scratch that (big thinkers 'round these parts). Riches, anyway.
Greg@74: No offense meant, and your point still stands, but "6 months" was no faulty prediction. They new precisely what they were doing. "6 months" was a calculated lie to get their foot in the door. Establishing a permanent outpost in Iraq has been a neocon wet-dream for decades.
Matt@79: Excellent point. I'd add that as it is our duty to remain unsuperstitious in the face of a reality more and more fraught with "magic" (as in "indistinguishable from"), or we risk becoming small-minded stewards to the machines of our forefathers.
Posted by: insect_hooves | June 14, 2008 12:21 AM
insect@68 - I agree with your point about the lisp macros. I'd also like to cite optimizing compilers as another example of bootstrapping. The general point is just that bootstrapping doesn't always yield impressive results. A lot of potential gains from computing saturate. Yes, you can use more MIPS to run CAD programs through more iterations, and consider more possibilities. Sometimes what that means is you double the CPU time, and now you've gone from getting 80% of the possible speed from some hardware to 90%, and the next doubling gives you 95%. Diminishing (marginal) returns apply to applying processor cycles, just as they do to land, labor, or fuel.
Posted by: Jeffrey Soreff | June 14, 2008 5:26 AM
Something interesting and maybe scary I have learned at my new job...
I don't know what you all picture when you think of a stock exchange or some other capital market -- I picture lots of people on the floor shouting and making trades, high-adrenaline, lots of uppers, etc. Apparently, over the past few years, the big finance companies have quietly started replacing their human traders with computer algorithms, which frankly do a much better job. Some of the big capital markets are going to close their human trading floors entirely. The finance companies still hire people -- but they're hiring CS grads to program the trading algorithms. (There's a reason the hedge funds are recruiting so heavily at MIT in the science and engineering departments these days...) These electronic trading systems -- high throughput, low latency data processing systems -- are making their owners and creators sometimes millions of dollars an hour. The writers and so-called "Cosmic Engineers" have it wrong -- it's not the Internet that's going to "wake up." It's the *world's financial markets.*
I mean, what better way for a newly self-aware AI to take over the world than to take over its money supply? Turns out we're building that AI into the very fabric our financial markets and handing it the keys. :-)
(All powered by my company's software, of course. "Exxxcellent!" *rubs hands, strokes goatee* "My plans for world domination proceed apace!")
Posted by: Kevin Riggle | June 14, 2008 7:13 AM
Keith @ 72
There is actually a purpose to AI and to building an "intelligent" machine: we learn a lot more about what we are and what intelligence is (and isn't) with it than we've learned so far just by studying the brain. AI's failures have been very useful learning experiences, and will continue to be so, until there are some big successes. I have to admit, some of those failures have been impressive in their, well, failing. Cyc comes to mind; maybe it was necessary for someone to create something so obviously doomed just to put a railroad spike in the coffin of the idea that tossing all the rules you can think of into a hat would accomplish something that worked like common sense.
insect_hooves, way back upthread
I don't think it's quite fair to brand all AI workers as anal CS researchers. For many years there were two rather hostile camps in AI: the "neats", the ones you described, and the "scruffies", who didn't believe in separating code and data, and who had great hopes for self-modifying code.
I think we're way beyond Lisp Macros these days, though it may be most AI mavens haven't noticed. Metaprogramming, self-modifying systems that partially evaluate themselves, self-organizing analogy networks; those are the cool tools these days.
The problem with AI seems to be that it's taking a lot longer than most AI people thought it would (mostly because the original objectives were both far too optimistic and way too unclear). So AI people tend to get distracted by other things: Hans Moravec's scoping out the Singularity, Rodney Brooks is designing vacuum cleaners and military robots, and Terry Winograd got out of AI decades ago and works in computer-human interaction now. Marvin Minsky and John McCarthy still work and publish on AI, but I don't think either of them has written any code in a long time; they're "Big Picture" guys now.
But watch for some interesting things to come out of the more exotic work: Hofstadter's Fluid Logic group, Rosalind Picard's Affective Computing group, maybe some of the people in Europe working on mathematical models of self-organizing systems.
Posted by: Bruce Cohen (SpeakerToManagers) | June 14, 2008 8:23 AM
NelC@77
This presupposes that we have robotic bodies that are generalists. Once again we have a large supply of cheap minds and bodies.
What would seem to be more useful in an AI is durability.
I have an engineer who is not going to die or retire. If I back it up regularly I do not have any risk of losing all that effort put in to the original training/programming, or the experience it has gained.
Posted by: Brian Rempel | June 14, 2008 1:41 PM
insect@84: They new precisely what they were doing. "6 months" was a calculated lie to get their foot in the door.
and there have been and still are plenty of people who honestly believe(d) that we can (could have) turn(ed) the war around in six months. They are similar to the Order of the Cosmic Engineers, they didn't start the meme, but they worship it now that they've heard it.
A slightly more interesting question is whether anything said by anyone predicting the Singularity is saying anythign that is falsifiable, or whether they're saying stuff so vague that they always have the option of moving the goal posts and redefine the problem.
Put into more real and direct terms, are any of the Singularity prophets willing to put their money where their mouth is? Are they willing to pay if their predictions are wrong?
Can they even define the "win" scenario in objective enough terms that they could put a bet on longbets.com?
I'm not a gambling man, but if someone put a sure bet like that in front of me, I'd put some money on it.
Is anyone giving odds and taking bets on the Singularity?
Posted by: GregLondon | June 14, 2008 4:49 PM
The Order of the Cosmic Engineers have finally posted their prospectus.
I was relieved to see that it would be engineers that would "intimately join, cross-pollinate and cross-leverage our mental resources into a meta-mind society."
However, closer inspection of the founding members reveals a a number of writers, journalists, lawyers, PR people, etc. I'm hesitant to join a group that might take me as a member. Will I still have to join the meta-mind?
Posted by: Kendall Nettles | June 14, 2008 7:17 PM
Kendal @90: Heretic! Burn the unbeliever!!!
(In case anyone is still suffering under the burden of any misconceptions -- despite having written "Accelerando" I am as conflicted and ambivalent about the Singularity as Ken MacLeod is conflicted and ambivalent about True Communism, and for much the same reason. Jam tomorrow, comrades!)
Posted by: Charlie Stross | June 14, 2008 7:28 PM
Did you see this invention of a radio smaller than a human cell?
http://www.physics.berkeley.edu/research/zettl/projects/nanoradio/radio.html
I told my girlfriend that it won't be long until we have internet connections embedded in our heads, but she didn't seem all that excited.
I just worry about who is on the other end.
Posted by: Kendall Nettles | June 14, 2008 7:44 PM
Brian @88: I think there's a good case for supposing a humanoid shape would be useful for an AI working in a human environment. But your engineer is an even better case. Imagine that your AI engineer is very good at what it does and earns you enough money to expand the company. You need another engineer, so why not run a copy of the first, complete with all the original's valuable experience?
Posted by: NelC | June 15, 2008 1:11 AM
Dear ghu. These people are nuts.
Posted by: Marilee J. Layman | June 15, 2008 1:15 AM
Bruce@87: Intersting points, all. Jeffrey Hawkins' Numenta is doing interesting work as well. I'm not so sure we're way beyond Lisp macros, but this isn't the place to throw down on a full-fledged language discussion. Let me just say that I think it's an important point that Lisp is the only computer language that could have been developed independently by an alien race.
Charlie@91: We pick nits (we ARE nerds after all =), but when it comes time, we've got as much belief suspension as you've got wild speculation. It MUST come sometimes to "jam to-day."
Posted by: insect_hooves | June 15, 2008 2:40 AM
Charlie, if you read Drexler's very earliest stuff, it's clear that he was still thinking of solution-based chemistry, not fullerene or diamondoid structures.
It's also clear, reading his later work, that he doesn't actually understand chemistry very well.
The Singularity -- what a useless concept. If you define it at the point where a human mind is unable to grasp the contents of its civilization at least in outline, it occurred some decades after the invention of the printing press. Athanasius Kircher was famously (and inaccurately) known as "the last man who knew everything", and he lived four hundred years ago. If anything, we're moving back to an age where all worthwhile information is at hand. Our summas, our search engines.
Posted by: Carlos | June 15, 2008 4:02 AM
Carlos@96
Good point.
Does the speed of change matter if it's outside the area you choose to concentrate on?
Do most physicians really care about the major changes occurring in theoretical physics? For the most part they are not even equipped to understand the bleeding edges of each others fields.
Most people will only care about the practical details that impact their lives. They just want to press a button and turn on the lights. They don't care how the power is generated.
Posted by: Brian Rempel | June 15, 2008 5:29 AM
David @ 65: I don't agree, but of course, I could be wrong. Let's be clear here, though—I'm talking about a "human equivalent" AI, that is, an AI which is general-purpose and can replace humans at (most) any task. For example, Asimov's Robots. This is different from, say, an automated translation machine, which might have some aspects of intelligence but has no self-determination.
It seems to me that what you're essentially arguing is fundamentally that we could design an intelligent machine such that it can only behave in prescribed ways (again, Asimov). The problem I have with this argument is that one of the basic functions of intelligence is the ability to learn and make decisions. In fact, we generally consider intelligence to be superior the more innovative those decisions and the broader the ability to learn is.
But this whole concept is basically the opposite of that of a pliable slave. How, for example, are you going to make an AI which is capable of writing knowledgeable and insightful prose on the subject of ethics or morality, but which at the same time is incapable, held back by some inviolable rule, of applying that same reasoning to its own behavior? It seems pretty strongly contradictory. Such a being can understand and generate advanced ideas in ethics, like "robots are people" and "humans oppress the machine classes" or whatnot, but it's somehow incapable of doing anything with those ideas (and only, specifically, those ideas).
Now, of course, humans don't all follow some sort of individualistic ideal. Some people enjoy being submissive, and others lack the drive to improve their lot—and certainly religion is a great example where people go to great lengths to rationalize things. But I've never heard anyone seriously argue that individuals were incapable of doing something. "Oh, it doesn't matter how shitty we treat them, they're incapable of turning on us." Successful tyrants through history have always well understood that revolt is a very real possibility.
Now, of course, I could be wrong. Perhaps humans are just one type of intelligence, and other types are far more rigidly limited in some ways. But it does seem like a contradiction, to me.
Posted by: Justin | June 15, 2008 6:45 AM
Charlie @ 67: But it sure is a shame about the American cover art. Maybe someone should hold a contest to design suitable replacement dust covers, so nobody need be embarrassed when reading it on the bus? :-)
Tlönista @ 66, insect @ 68: The state of AI research is significantly beyond LISP macros. There's been a lot of interesting work in machine learning, neural networks, expert systems, and so forth. Part of the problem is that AI is so ill-defined, though.
For example, there's a lot of money to be made right now in designing facial/walk/etc. recognition software, so all those security cameras can automatically call in the goon squad when they see something Suspicious. This is clearly (from a CS perspective) an AI problem. However, it doesn't involve robots (no robo-goons) and the average person sees it more as a camera gadget, so it doesn't get press of the sort like, "great strides are being made in the field of computer vision AI!"
Or, for example, machine translation of documents. It's a pretty important field these days, and clearly an AI problem. But because it's established, it has its own name and is differentiated that way.
On the other hand, robots still tend to be out-thought by your average insect. And the more impressive-sounding examples tend to be based on cheating. For example, in those DARPA outdoor driving challenges, the way the robot cars worked was that the design teams gave them an exact GPS path to follow (on the order of a waypoint every 20 meters) and the job of the AI was to avoid running into things directly ahead of it at high speed. A job which they typically failed at, judging by the results... Despite having a bunch of gizmos such as laser terrain scanners and the like.
Posted by: Justin | June 15, 2008 7:11 AM