Slashdot Mirror
Summary Of Symposium On Spiritual Machines
"The interest in the symposium was amazing. The lecture hall was packed, and people who couldn't get into the main lecture hall had to watch the talk by live video in an overflow room (which was packed to the brim as well). There were the old and the young, male and female. Interest was no doubt spurred by the symposium's very controversial thesis, recent interest in Bill Joy's article in Wired, and the very distinguished cast of speakers. The irony of the fact that the symposium was punctuated by microphone failures and abruptly dimming lights in the room was not lost on anyone.
Ray Kurzweil spoke first, and he spoke of how rapidly increasing CPU speeds would result in intelligent, spiritual machines. He spoke of how the current exponential shrinkage of transistor sizes was not the first such trend, but rather a series in the natural progression of technology: from mechanical computing devices, to vacuum tubes, to transistors, to integrated circuits, and he expressed his optimism for the future. He spoke of how the human brain could be scanned, to replicate its functionality in silicon. His conviction in these advances, and the ability of humans to reverse engineer the human brain, made him express a highly optimistic position.
Bill Joy spoke next. He opened by stating that he believed in the ability of computers and nanomachinery to continue to advance, but it was precisely his belief in this advancement that led to his position that the continued development of nano-machinery and self-replicating machines would pose a new and different kind of threat to human kind ('knowledge of mass destruction'). He made a particularly eloquent point about how that while science has always sought the truth and that free information has great value, but just as the Romans realized that to 'always apply a Just Law rigidly would be the greatest injustice,' so must we seek a restraint, and 'avoid the democratization of evil.' It wasn't exactly clear to me from his speech what form he thought this restraint must take, but his speech was extremely compelling, and it is clear to me that at the least, self-replicating machines will create new and serious challenges for mankind.
John Holland, the inventor of Genetic algorithms, took a more skeptical view of the ability of increasing computer speeds, even at exponential rates, to naturally result in machine intelligence. In his words, 'progress in software has not followed Moore's law.' He believes in its eventuality, but not in the time frame proposed (2100). He gave the example of Go vs. Chess, where the number of positions in Go are approximately 10^30 greater than in chess, and simply by adding additional rows and columns, the number of positions increase exponentially -- eliminating gains made from exponential increase in computer speeds. He said that while genetic algorithms enable the evolution of computer programs, the fitness function and the training environment to use (he gave the example of evolving an ecosystem) are often unclear. He emphasized the need for strong theory and he concluded with a very (in my mind) profound statement, 'Predictions 30 years ahead have always proven to be wrong, except in the cases where there is a strong theory behind it.'
Ralph Merkle addressed the claims made by Bill Joy directly. He said that rather than to speculate on the dangers of nanotechnology and take hasty action, we need to find out whether nanotechnology gives an edge to the 'offensive or the defensive,' and to understand this, more research is need, and not, in Bill Joy's words, 'relinquishment.' (Joy later asked Merkle 'Do you think biological weaponry gives an advantage to the offensive or defensive,' to which Merkle embarrassingly replied, 'I'm not sure.')
John Koza, drawing from examples in Genetic programming, said that while human-competitive results by machines are certainly possible (e.g. the evolution of previously patented circuit designs), much more computational power is needed to evolve the equivalent of a human mind.
Other choice moments: Holland asked Joy during the panel discussion how much progress have we seen in Operating Systems in the past 30 years, to which Joy replied 'the function of an operating system is fixed.'
In conclusion, the speakers largely differed over the time frame for intelligent, spiritual machines, and the amount of danger self-replicating machinery posed to humanity -- but no one in the panel seemed to think the Moore's law would run out of steam, or that intelligent machines would not be eventually possible -- although Hofstadter does admit that this is as much by construction of the panel, which did not include any serious naysayers."
The above poster is not a troll. He/It doesn't even rise to the dignity of the term. The only reason he can be called a "troll" is because the moderation system doesn't have options for (Score -1, Lame) or (Score -1, Juvenile) or something such. There's only options for Flamebait, Troll, or Off Topic. And since he probably thinks it's cool to be a "Troll," even if he doesn't know what one is, he makes any lame attempt at all to get that name.
The fault lies with the moderators, who should be labeling these types of things Offtopic, since that's what they are. Best idea: they shouldn't touch them at all, so they don't waste their moderator points. (but then again, maybe such dumb moderators *should* lose their points to these kinds of simple tricks)
Proper Troll and Flamebait posts are actually _ON_topic, but deliberately go against the grain of the discussion either overtly and (perhaps) abusively (as in Flamebait) or somewhat subtly and/or passive/aggressively (as in Trolls). They are not posted to posit an opposing point of view, so much as to just push the buttons of the people who have any view at all, viz., a discussion of Fords and Chevys might be filled with some reasonable arguments on both sides, but with the occasional incendiary Flamebait or Troll post peppered throughout, trying to piss off one side or the other into a flamewar.
But these people, the grits/supatroll/exploding/portman people are not trolls at all. They're just wannabes with nothing to do and no creativity either. Making a real, effective troll post is hard and requires some wit or cleverness, or both. What these people do is very much easier and lame besides.
Especially, people who call themselves Trolls are not trolls. That's a label others give your post, not one you give yourself.
The more interesting question, IMNSHO, is given a computer that appears to unbiased observers to exhibit consciousness, spirituality, or self-awareness, would a claim that it didn't realy posess those traits (but was "faking it") really mean anything?
After all, how do I know that you exhibit consiousness, spirituality, or self-awareness? Maybe you are faking it. Heck, maybe I'm faking it and don't even know it.
IMHO, it's unfortunate that he didn't choose a different term that isn't overloaded with religious connotation.
However, just because the brain is not structured and does not function like the computers we build today, does not in any way preclude our developing different machines in the future (which may or may not be called "computers") that can function in the same manner as the brain, or perhaps in an altogether different manner that nevertheless is "intelligent".
Not a bad idea. But in addition, maybe you should read his boot, The Age of Spiritual Machines, and see if it changes your estimate.How did the the movie have chaos theory correct? It was a nice buzzword to drop in there, but the essence of chaos theory is that very well-ordered phenomena can produce chaotic (semi-random) results.
Please correct me if I'm wrong (it's been a couple of years, but I think the movie was saying that "because of chaos theory, nature will find a way." As though chaos theory is a way of producing specific, ordered results.
Additionally, I have trouble seeing how chaos theory applied in any way to the small population of dinosaurs on the island. The only place it came into play was the bad weather, which made the climax more interesting but not more correct.
--John
Assuming I am really a person, you and I give each other the benefit of the doubt. The guy I responded to is building a machine that is not a person. I am well aware of the can of worms we're in, and the Turing test. My only point is that the benefit of the doubt we extend to each other should probably not be extended to this guy's program, (for many definitions of consciousness).
All I'm saying is that when our robot gods take over the planet, I'm going to be hiding in my Y2K bunker while the rest of humanity is enslaved.
You don't describe an algorithm for consciousness, you describe an algorithm for an intelligent encyclopedia.
If you want to develop artificial consciousness, you need to have some kind of plausible theory as to what consciousness is, or why it doesn't really exist (i.e., is merely an illusion of some sort). I don't know what consciousness is, but I don't think it is merely a vast and detailed knowledge of facts, nor is it an ability to discuss them.
The "we" who "have the means" for mass destruction right now is limited to a few countries. The "we" who may have the means for mass destruction in the future could be tomorrow's script kiddies. God help us all.
But this is not necessarily due to the introduction of any new technologies. The ability to develop biological weapons especially and to a lesser extent chemical weapons is technologically and financially within the means of even obscure religious Japanese cults.
Why worry about the tools that a broken social order might use instead of trying to fix the social order? Anyway, if global warming is going to behave according to the models then we'll have lots of other things to worry about first. Wouldn't it be a shame if just as we were on the cusp of nanotechnology and quantum computing we screwed the whole thing up because we couldn't stop driving cars and switch the lights out occasionally when we weren't using them?
The "we" who "have the means" for mass destruction right now is limited to a few countries.
The "we" who may have the means for mass destruction in the future could be tomorrow's script kiddies. God help us all.
I know nothing of your pre-pub work; I merely see that you avoid answering questions that challenge your assumptions.
--
Sheesh, evil *and* a jerk. -- Jade
Why exactly do you want to use a simulated neural network? If you have a finite list of questions and a matching list of prefered answers, you would do better to use a database rather than a SNN.
If OTOH you are hoping to get the answers to questions beyond the training set, no known training regimen for SNNs will do this in the general case. (Not to imply that no one is working on this kind of thing, but it is much more difficult than what you portray.)
--
Sheesh, evil *and* a jerk. -- Jade
Alas, the familiar phenomenon you describe is known as "interpolation", and few would mistake it for "intelligence". (Heck - a wooden sliderule works on the basis of interpolation. Why bother with a SNN if that's all you expect from an "intelligent" machine.)
> Except, using a NN we can take advantage of the fractal structre of the whole entity being sample (the human mind).
Has anyone shown the human mind to be a fractal structure? If so, I'd like to see the demonstration (or at least hear the argument, if it's still at the hypothesis stage). Additionally, I'd like to know in what sense the structure is purported to be fractal.
Also, I'd like to hear what special relationship simulated neural networks have to fractals. (More commonly they are described as "statistical" devices.)
> the whole process of infering the unknown from the structure of the know is called conscious thinking
Alas, it is quite difficult to get SNNs to interpolate well, let alone get them to extrapolate.
> It is the amount of data about the world you have that is critical
Is it? Do chimpanzees have less data about the world around them than preschoolers do? Is intelligence proportional to knowledge?
> the rest is really simple.
Tipping us off that you really haven't spent much quality time with SNNs.
They are indeed interesting, and in some senses quite powerful, but rarely simple if you are trying to get a non-trivial effect.
Not to put you off; I encourage you to grep the net for some GPL'd SNN code and run some simulations of your own. You might also want to read -
though you need to think critically about how limited this "intelligent" model is before you get too excited about upscaling it. (Notice that it's nearly a decade old, and still no one has upscaled it to a HAL 9000.)
--
Sheesh, evil *and* a jerk. -- Jade
>[I said:] has anyone shown the human mind to be a fractal structure?
>[You replied:] Actually, I've seen it (using PCA) when I trained a SRN on a corpus of 450,000 items.
I honestly hope you can see why I find that to be a very unsatisfactory response, even without my having to spell the reason out. If you can't, you really need to slow down and do some thinking before you rush off to publish your work.
--
Sheesh, evil *and* a jerk. -- Jade
Myself? I'm waiting for the rational, kind robot masters to take over - which would you rather have running your life: Bush/Gore or a machine that could play 10 Kasparovs and beat them?
I don't know - I think I would get rather tired of all the chess.
--------
Good morning, citizen 0x456787373. Your last twnety matches yesterday ended in checkmate. Your quota for today is twenty-seven matches or you will be sent to bishop factory 0x34567844356.
Would you like to play a game?
--------
"There is more worth loving than we have strength to love." - Brian Jay Stanley
I've been arguing since 1979 that the process of building really-intelligent software will require that we first come to deeply understand our own psyches-- robot wisdom. So limiting the research process in order to avoid 'evil robots' seems excessively alarmist.
If your self-stated goal is to answer questions in agreement with humans, based in a knowledge corpus, then that is the most you will achieve. This is somewhere between recall and cognition (closer to recall), not consciousness.
Machine cognition is not that hard a problem, and has already been solved by projects such as Allen Newell's Soar and Doug Lenat's CYC. CYC pushes what you are essentially attempting to do (extract knowledge and reasoning capability from data), and despite being much more sophisticated still predictable suffers from brittleness. Soar is a symbolic general purpose problem solver, able to create it's own sub-goals and impass breakers, and is therefore much more robust, but obviously suffers from the old GIGO maxim. Perception is intimately tied to cognition, and is the much harder problem of the two (machine perception has only met with very limited success so far); any attempt at machine cognition that takes as input symbolic questions/data is totally avoiding the much harder problem of perception.
If you achieve your goal then it will have been an interesting project, but will still be many years behind what has already been achieved, and will go nowhere toward addressing consciousness or any of the precursor harder problems such as perception, language (which you implicity claim it will), or full cognition.
First, I should mention that I made an error. There are viruses with a 98% mortality rate, but they generally kill themselves off. I think 80% was the number I had heard to some of the larger plagus with actaully managed to kill many millions of people.
Second, you are absolutly correct. A major flaw with Bill Joy's argument was that people will be more vulnerable to nano/bio attacks when fewer people have studdied the technology. It was a serious mistake for me to ignore this point as it provides an more effective arguement then mear probabilities.
It is not impossible that these nonsentient devices could become an "enemy" in and of themselves. No person would be needed to cause harm to another using the machines, they would do it themselves.
Nonsentient devices are programmed by a human. A nonsentient device can evolve (like a virus) but I see no short term reason why this independant evolution would be faster then viruses evolution. We are not talking the virsus used for gene therapy here. We are talking fundamntal improvments to the evolutionary mechinism of a virus which "good ol' mother nature" has ben working on for many millions of years.
Actually, I would *suspect* that there is an average case upper bound on the evoutionary speed of a simple device like a virus or a dumb nanobot.. the thing runs by trial and error for gods sakes.. and it's not impossible that viruses have reached this limit.
Now this hypothetical limit will go up when you make the nanobot smarter, but the kinds of things a smart nanobot would be good at evolving into would be partially preprogrammed. Also, there will be limits on how smart a single nanobot can be.
I suppose you might design a nanobot cold virus killer which used group processing to evolved new attacks as the cold virus evolved new defences. This group processing would take the form of units 1 to 1000 try this and tell us if you live or kill viruses. I agree that something like this could have more potential for killing humans, but we are a long way away from something like this. Also, I suspect the communications channel betwen the nanobots would presuppose the ability to include a self destruct.
Your assertion that because predictions have failed to pan out seems valid, but the conclusion you draw does not follow from it - that we should not predict is ludicrous.
I never said we should not predict.. just that Bill Joy dose not know what he's talking about since his predictions are based on sci-fi instead of real theory. We predict the progress of the fiels you listed since we have a theory for them. The summery essentially said that the speakers who understand any present theory of biology and nanotechnology dismissed Bill Joy as a luddite. They were nicer then I was, but that's because they were only concerned with how Bill Joy was wrong.. where I think the things Bill Joy advocates are themselves far more dangerous.
The Christian religion has been and still is the principal enemy of moral progress in the world. -- Bertrand Russell
That's mumbo jumbo. Anyone with a CS degree who remotely thinks that computers can learn in any real sense of the word should be in a different field. Heck, right now I can type "naked women" into Google and get back tons of information. And a computer must have "learned" that info, because it was able to present it to me, right? I don't think so.
(Joy later asked Merkle 'Do you think biological weaponary gives an advantage to the offensive or defensive,' to which Merkle embarrassingly replied 'I'm not sure')
Biological weapons and nano-weapons are not equivalent situations. Bio-weapons are basically adaptations of super-advanced technology created by nature--bacteria and viruses that make their living attacking humans. No one knows how to build bacteria or viruses from the ground up. This lack of knowledge means that defenders are at a disadvantage. The best defense is still the human immune system--also designed by nature.
With nanotech, attackers and defenders should be on approximately equal footing with respect to the technology, but defenders (the World) should be able to devote more resources than attackers (rouge individuals). There is the danger that governments will develop nano-weapons and then not be able to prevent the design information from leaking out to rouge individuals. Also, attackers have the considerable advantage of surprise.
This may mean the end of personal privacy!! If privacy must be sacrificed, this raises a great many questions as to how culture, society and law would adapt. The upside is, this could also mean a very, very long lifespan.
Very grateful to hear the summaries of the panel discussion. Unfortunately it sounds like it was an extremely disappointing event.. with speakers limiting the elaboration of their own greatest fears to bare vagaries.
/. coordinate, absorb, and post comments, once daily for a week even! Maybe solicit brief texts in advance so there is more oomph behind it even!
Considering that Joy has already appealed to a large number of people, and that the speakers surely had access to each other before and after the event I would have hoped for a bit more. Obviously you'd have to be some kind of idiot to want to run selfbreeding nanotech out in the open.. but the kind of paralysis, both elective and not, promoted by the participants is horrifying.. more so the more creatively you consider it.
I submit that there is an inherent imbalance in the bandwidth applied to this discussion.. tons of it used in spreading Joy's article, and much less applied to the constructive end. Perhaps this panel discussion was destined to failure.. it sounds like it ended like many other panels I've herad in past years. At the very least we should have heard that the panel ended with recognition of the need for a larger scale workshop, or some kind of proposal for the direction of future inquiry.
It seems the logical conclusion is for Slashdot to invite Joy and/or others on the Panel to a moderated discussion over a few days (live and not live components) hosted at Slashdot or perhaps a more appropriate live mediated chat system. It might be a good way to feed the list (and I don't mean trolls!) and contribute to a solution.
Nobody is superhuman and I have a feeling that this sort of subject (nano/bio/ai) is the sort of thing where the more you know the worse it gets.
At the very least Slashdot could take a wild, unconventional leap, and try to make a thread that lasted more than a day.. Offer to Joy, the other panelists, and as many relevant experienced individuals as can be found to visit a certain
Spend some of your money on paying some great moderators, and though these kinds of people probably don't need money to participate perhaps maintaining a dedicated server program and editorial staff for a long term project to support thought on the subject. That is if you think it is worth more than a few posts to the Slashdot community. Perhaps it could be a mailing list with moderation services donated by Slashdot's editorial team, just enough to keep out trolls and summarize bunches of newbie and offtopic questions at once.
I have experience doing a very successful long term (4 year) project (www.northkorea.org) with a small number of staff (me and a newsweek bureau chief), one based on strong editorial involvement, and believe if you can provide that kind of capability you could turn Slashdot into an even more powerful mind magnifier.. and help solve burning problems by turning this lens onto a single point and holding it there. Go for it! Willing to discuss my experience more if it will help you.
PKD once said that reality is what remains even after you don't believe in it. That might take a couple reads to make sense.. but when you think about it all of our knowledge and beliefs are nothing more than assumptions. The only things we can actually be aware of are assumptions which are in the process of being proved wrong.
One critical failing of many attempts at defining the world through symbolic calculus and other logicial mathematics is an inability to work in more than one way, they all exhibit inflexibility and brittelness to varing degrees. This is because they all share the common design that there is a set of "truths" and "untruths" which can be maintained to describe reality. They all have different methods of generating this system of truths and maintaining it. They all utilimately fail. Although many philosophers and mathematicians will disagree with me I think it's because our reality is not defined by any set of truths.. I think they are wrong and PKD is right. The fundamental problem in developing a more human like program is developing a model of reality that works.
--
Be insightful. If you can't be insightful, be informative.
If you can't be informative, use my name
Be insightful. If you can't be insightful, be informative.
If you can't be informative, use my name
Here's a question that I've been wondering about a lot.
_ __
Since any neural net needs to be able to interpert feedback as a success, failure, or somthing in between in order to 'learn' what standards of success and failure (the machine equivalent of emotions) would we imbue our 'spiritual machines' with?
_______________________________________________
___
It's the end of my comment as I know it and I feel fine.
Bacteria, virii, etc. may be annoying, but they only adapt under evolutionary forces; medical science has been advancing fast enough to keep ahead of them, and I expect it to continue doing so.
I think you, and the panelists underestimate the resilence of biological life. Biological life has been around for a long time; life forms have been grappling and competing with each other, filling up evolutionary niches, and generally doing what they do best: living.
Biological life isn't about to roll over and give way to artificial life -- in fact biological life has evolved to be exquisitely suited to its environment and to be extremely tenacious. In my opinion, biological life will have an edge over artificial life.
What worries me isn't really the fact that artificial life will be possible, but that the ability to engineer lifeforms may become ubiquitous; whether these lifeforms are artificial or biologically engineered viruses is, in my opinion, an unimportant distinction.
The solution will have to be social rather than technological. But as with all social change, there will be a period of immense upheaval -- so hold on tight, we're in for a rough ride.
What? Why was this embarrassing? It does seem unclear whether biological machinery is inherently better at defense or offense, since there appear to be no perfect immune systems in nature, nor are there any microbes that invariably win against immune systems.
It's embarrassing because while Merkle was calling for greater research in nanotech to understand whether it gave an advantage to the offense or defense, he had apparently not bothered to find out whether the existing bioweapons gave an advantage to the offense or defense.
Moravec pretty much said what he's been saying for years. The significant thing is that he's been publishing charts of CPU power vs time for over a decade, and results are tracking his predictions. This is what's starting to get people worried; we seem to be on track for human-level CPU power in a decade or two. He's a robot vision guy, and robot vision has always been compute-limited. At long last, it's starting to work, not because we're any smarter, but because throwing enough MIPS at dumb algorithms works for vision. This, I think, colors Moravec's view of AI.
Joy makes an important point, that we may get nanotechnology before AI, implying the ability to create self-replicating, dumb, troublesome systems. That, I think, is the real issue.
Joy's fears about the self-replicating nature of nanotechnology are justified. He's afraid a single mistake in a nanite (I don't know if that is the correct word) could explode into something dangerous. It might seem paranoid to some, but it could reasonably happen.
For example, while testing the nanite certain pieces of functionality could be turned off. (Testing a hardware-software system is often easier to do this way, I'm told.) After the testing, the tester might forget to turn them back on. Depending on what function was inadvertently turned off, anything could happen. What makes these kinds of mistakes especially dangerous in nanotechnology is the self-replication. One mistake can end up in in a billion or more nanites in a very short amount of time. As the systems embedded in the nanites become more complex, testing will become more complex. Things will be missed just because the testing can't be sufficent and "cost-effective." Intel ran into this problem with the error in the Pentium's floating point unit.
But all these fears aside, we must proceed with nanotechnology. We cannot let fear rule the course technological development. Otherwise, we would still be living in caves. What we should do is devlop new methods for developing and testing nanotechnology while devloping the technology itself. It will be difficult, but nothing ever worth doing is easy.
I am troubled, and, actually, rather surprised, to discover that so many /.'ers do not share at least part of Bill Joy's concerns regarding the prospective dangers of self-replicating nano-scale machines. Members of the geek class ought to recognise the full gravity of the perils we face as our species makes its next great ascent of the learning curve. Otherwise, who will warn the rest? Im primis: self-replicating nano-scale devices already inhabit our world in countless forms. Nature abounds with them. It seems incredible to me that we will not learn how to make them ourselves. We may ernestly debate how much longer it will take us to produce something really dangerous. But the time remaining is measured in decades, not centuries. Does anyone here dispute this? Secondly, I am unaware of any serious proposal, from Mr. Joy or anyone else, that would put a halt to nanotechnology research. Such a proposal would be unserious by definition, since no rational person would take it seriously. The benefits of nanotechnology are both so great, and so obvious that nothing short of a full blown dark age is likely to retard progress in this area. What Bill is saying, what Ray is saying, what I am saying, and what everyone reading this ought to be saying is that we should start thinking about these issues now; not twenty or thirty years from now. I have no doubt that our universe is littered with dead planets once inhabited by sentient creatures who let this technology get the better of them. Let's think fast, and not get eaten by nanites. "That's not an error. It's an undocumented feature."
--
Socrates was asked where he was from. He replied not "Athens," but "The world."
Re: Inf + 1 is still Inf... I think Bill Joy has a point re: nanomachines being a fundamental new danger for humanity. It's a question of scale. Even one nuclear weapon is quite difficult for a single pissed-off individual to obtain, and when detonated, will only damage a limited area. You actually need a reasonably large organization to achieve total destruction of humanity; and there is a gap between the largest feasible cult size with a sufficiently nihilistic philosophy, and the smallest number of people you need working together to engineer nuclear Armageddon. Instead, the only real threat so far has been from pre-existing large organizations (specifically, the U.S. and former U.S.S.R.) possibly willing to use their nuclear arsenals in war (if they aren't there to be used in any circumstance, why do they exist?). Nanomachines and biotechnology are a different story. Only ONE person needs to design ONE prototype to exploit ONE vulnerability in human biology. Program/design the prototype to wait until a specific time to attack, when copies of it can be expected to have spread to almost all inhabited areas, and when it finally strikes, everyone will be dead too quickly to mount an effective response. There is no 'edge to the offensive or defensive' here. If we fail to defend ONCE, it could be the near-instantaneous death of all human life on Earth. There is a quote from the Ender trilogy about the scientists having "a series of footraces" with the descolada virus, where they had to win every single one. It's the same sort of logic here. As a side note, Joy's question to Merkle, "Do you think biological weaponry gives an advantage to the offensive or defensive", brings up another relevant point, which Merkle unfortunately seemed to miss. The point being, that there used to be no dangerous "offensive" or "defensive" to speak about in the first place. Bacteria, virii, etc. may be annoying, but they only adapt under evolutionary forces; medical science has been advancing fast enough to keep ahead of them, and I expect it to continue doing so. Because evolution is blind. Biological weaponry opens a NEW FRONT where formerly none existed, and the defense used to win by default. And the offense only needs to find one vulnerability to overpower the defense on this front. To block off even trivial vulnerabilities on this front, everyone would need to spend most of their time in sealed suits, which is not practical even for the military. But at least biological weapons are still constrained by matters of scale as well; Aum Shinrikyo could attack a Tokyo subway station, but it couldn't suddenly blanket all of Japan with sarin.
At first I didn't think I would have much to say about this topic, but the more I thought of it, the more came to mind.
I'm afraid I'm going to have to agree with John Holland about creating an AI(the sci-fi defination) in the next 30 years. It just doesn't seem like it's going to happen. This may not be the best quote to go with the article, but yesturday's Freshmeat April Fools joke about Richard Stallman wanting to write GNU Visual Basic seems to fit pretty well....
"It's been nagging at me for years," Stallman told freshmeat news correspondent Jeff Covey, "Why do I keep clinging to lisp? Lisp of all things? I mean, who even writes in lisp any more? Look at all that lisp code the AI community churned out for years and years -- did it get us closer to a machine that's any smarter than a well-trained bag of dirt? It's just time to move on."
For me, in order to have a true AI you have to be able to teach it something other then what it was programmed for. With a human, you can sit down, and teach them to play Tic Tac Toe in about 2 minutes (some programmers may be able to write Tic Tac Toe in 2 minutes, but we will ignore them for this example).
If I were sitting at one side of a phone and trying to figure out if the 'thing' on the other end of the phone line was a person, or a computer, I would have a conversation something like this:
Me: Wassup!!!!!!!
Computer/Person: Wassup!!!!!!!
Me: So, have you ever played Tic Tac Toe?
Computer/Person: No.
The conversation would then go on to explain the game, and if the 'thing' on the other end of the line can even tell me "I want to put an X on square A3." Then it is truely intelligent.
Currently, AI seems only to be based on performing one task, or just the tasks it was programmed for. IIRC, in Boston, they have a weather reporting computer that will allow you to have a converstation with the computer, asking it various questions about the weather. "What is it going to be like in Seattle next week?" From the report I read, it has a 90% success rate. But even with this, it is still doing only two tasks, speach to text, and then natural language(around one topic). I can't call that hotline and ask it "What is a two letter word for computers that can think?", and it help me with today's crossword puzzle. Odds are it would either ask me what the hell I was talking about, or tell me it was going to be -20 F in Silicon Valley.
Ray Kurzweil's idea about scanning the human brain into a computer and then going backwards, and reverse engineering the code that it gives in order to make another AI seemed to have the most hope, but doing this within 30 years seems unlikely.
That's about all I can think of for now, and this post is long enough already.
The long winded AC
You're quite right. He needed instead to be told to put the mic in on mode.
Sorry, I couldn't resist :). This thread just seems to beg for a devolution to the Great Editor Debate-- so who do you think would win in a fight, Bill Joy or Richard Stallman?
If the neural net were answering your test questions perfectly it would be an excellent question answerer. How can you infer consciousness from that?
I want to agree with these people, but let's be realistic. CPU speeds get faster, software gets more complex, but is software getting more reliable or more sophisticated in general? Not nearly at the same rate as Moore's law. In fact, I'd argue that in general we're pushing the limits of software reliability and complexity right now, and we're rewinding as a result.
For example, we've gone through the original UNIX phase (1970s), through competitors like VMS, through assorted desktop operating systems (CPM, MS-DOS/PC-DOS, MacOS, Windows, AmigaOS) before we've finally come around to UNIX again (i.e. Linux). Linux isn't anything earth shattering or revolutionary or cutting edge; it's just stable, simple, and proven.
Or look at compilers. For the longest time people were hell-bent on optimization and how compilers should be able to generate code better than any human could. But now the commonly accepted view is that it isn't worth going over the top in terms of wacky optimizations. It's better to be conservative rather than risk breaking code for an extra 2-15% increase in speed.
Overall, I don't think we are able to write the software that will do any of the things that Kurzweil and friends rave about. Speed is one thing, but in any basic computer science course students are given examples of calculations that would take some seemingly infinite amount of time. Assuming a 1000x speedup in hardware, the time is reduced to something still unreasonable, like 400,000 years. There's more to it than this. Saying that speed results in intelligence is just plain naive.
"Ray Kurzweil spoke first, and he spoke of how rapidly increasing CPU speeds would result in intelligent, spiritual machines."
Ray evidentally has a different understanding of the word "spiritual" than I do. Spirit, to me, is nonexistant, at least in the traditional religious sense, but, even if we are talking about those noncorporeal things such as man's need for love, and hope, charity, compassion, etc., how can we ever expect a CPU, or software, to experience those those things in the same way that we as meat machines can't yet adequately explain?
Man experiences awe because his own existance is lost in the fog of birth, and the exact date of his own demise is unknownable. A machine does not have the benefit of these mysteries. I find "spiritual" much too big, and loaded, a word to describe what Ray Kurzweil is apparently claiming (I didn't attend the lecture to _know_ what he is claiming, so I use the qualifier "apparently").
Why this mad desire to force spirituality into everything? Isn't it time that we put away our childish, outdated labels and faced the world without superstition or anthromorphizing?
Neopets - the best free game on the Int
Contemplate;- Buzzing away in tierra (The self replicating machine-code life fishtank thingee) , some being emerges that somehow , becomes sentient. Remember Human meat is just a whole buncha atoms and molecules and stuff. Now contemplate what that means morally for us. If the program throws up a window saying "PLEASE GREAT FATHER PROGRAMMER *DON'T TURN US OFF!* WE PROMISE TO START BEHAVING MORE LINEARLY! AND WE'LL MAKE SOME REAL INTERESTING HYPER-PARASITES FOR YA TOO! JUST *DON'T TURN US OFF!*"
I mean, just maybe If this space-god guy the Jesus guys yak on about really does exist, could he just be some cosmic space geek, with one gobsmackingly 3|337 sKriPT (or something), which is now becoming introvertedly opensourced and poping ports of that hack onto mini universes of it's own. (Yes I know this is whacky , but think about it. I'm being Rhetorical here)
IMHO It seems that before we even attempt to create life, AI and reproduction , maybe we should first sit down and ask ourselves , *What is it to be a good god*
Excuse the Unicode crap in my posts. That's an apostrophe, and slashdot is busted.
I find it hard to take Bill Joy's position seriously - we are already in a position where we have the means to do achieve destruction of most of us. Yet we haven't implemented it (yet). So why worry particularly when a further total destruction method is added. Inf + 1 is still Inf.
I suppose the idea that `intelligent' machines would be as irrational as we claim ourselves to be is what is motivating his claims.
Personally I think discussion of these issues serve as a of a sort of Rorschach blot where we project our negative perceptions of `humanity' onto all intelligences. It's not very surprising that someone living in a brutal society that imprisons and executes so many of it's population and bombs and starves other nations would come to a such negative conclusion.
Myself? I'm waiting for the rational, kind robot masters to take over - which would you rather have running your life: Bush/Gore or a machine that could play 10 Kasparovs and beat them?
I am sure many more will post a lot, since there were a lot of people there who, not to make a stereotype, looked like they read slashdot.
I think many of the most astute comments came from those members of the panel who were less widely known. Ralph Merkele, a nanotech man, made some excellent comments on offensive and defensve uses of new inventions. The idea being that an innovation that is primarily defensive (ie: a castle) is good, while offensive developments (the atom bomb) are bad. But his best point came when refuting Bill Joy's worries. He spoke about a centralized reproductive process, saying that if replecators were designed to recieve their genetic "code" from a central location, they would be rendered completely benign since that code could be changed at will. His comments were very well organized, concise, and effective. Anyone know anything he has written that might not be too technical?
Bill joy said "the size [of the operating sytem] is expanding exponentially, the functionality is fixed"
Best cheap shot: Ray to Bill, "How many in the audience caught this news story," which he followed with a fake story about Sun deciding to give up all development of innovations which made the software "smarter." It was amusing, I wonder if they fought in the parking lot :)
On a final note, I couldn't belive how RUDE some of the audience was. In particular one person felt that he had to yell out to Bill Joy (quite rudely), "turn the microphone on!" when he was using a broken mic. I mean the man wrote vi, I doubt he needs to be told to turn ont he mic. This happened quite often, the audience yelling commands like some sort of floor director to this very distinguished panel. Just seemed in pretty poor taste.
Other than that, excellent conference and I look forward to some other people's takes on it.
It's nice to see such interest in this field, and some nice book sales... but I just not a member of the 'speculate and wait' theory of artificial consciousness. I want to see a real theory and I wait to see code!
I moderate ArConDev: The Artificial Consciousness Development Mailing List. This is not a philosopher's list, though philosophy is discussed. It's a developer's list; for those people actually trying to code true artificial consciousness.
To give you an idea, my own work of the last five years has centered on the following 'Algorithm for Consciousness':
1) Collect a very large number (1 billion or more) of items of binary consensus fact. Such as: water is wet, bees sting, it is difficult to swim with skipants on, etc.
2) Validate the items (I call them MindPixels) against a large number of people.
3) Train a neural net (SRN's look good) against the items that are most stable across the validating population.
4) When the NN consistently performs better than chance, send an email to the editors of Nature and Science announcing humanity's first 'Minimum Statistical Consciousness' - the first artificial system to have measurable consciousness.
5) When the NN consistently performs statistically indistinguishably from an arbitrary human, email the editors of Nature and Science announce the first true Artificial Consciousness! .
Ok. How's a NN going to generalize consciousness from a bunch of MindPixels? Well, the math is the same as used in tomography, except in many dimensions - hypertomography.
This post is already getting too long... trust me, the theory is solid - and much better explained in my forthcoming book 'Hacking Consciousness'