This is a link to an essay by Stevan Harnad titled "Post Gutenberg Galaxy: The fourth revolution in the means of production of knowledge".

Who is Stevan Harnad, and what kind of psychedelic essay is this, you may ask?

Harnad is a cognitive scientist and can be considered the father of the Open-Access Movement, a movement that evangelizes the free publication and free distribution of academic writings; something akin to the Free Software movement, but focusing on academic output instead of computer software. As a matter of fact, Harnad is sometimes likened to Richard Stallman. In the essay linked above he presents his vision that the free flow of academic knowledge, empowered and accelerated by the modern electronic communication technology, and the global network, will result to a revolution in the way knowledge is produced. This is what he calls the "fourth revolution" (the previous three being: speech, writing, and typography).

In 1994, Harnad published a subversive proposal (Scholarly Journals at the Crossroads: A Subversive Proposal for Electronic Publishing), presenting methods that can be used to hasten the arrival of the day when esoteric, peer-reviewed, electronic publishing becomes ubiquitous. From this proposal is the quote that follows:

We have heard many sanguine predictions about the demise of paper publishing, but life is short and the inevitable day still seems a long way off. This is a subversive proposal that could radically hasten that day. It is applicable only to ESOTERIC (non-trade, no-market) scientific and scholarly publication (but that is the lion's share of the academic corpus anyway), namely, that body of work for which the author does not and never has expected to SELL the words. The scholarly author wants only to PUBLISH them, that is, to reach the eyes and minds of peers, fellow esoteric scientists and scholars the world over, so that they can build on one another's contributions in that cumulative, collaborative enterprise called learned inquiry. For centuries, it was only out of reluctant necessity that authors of esoteric publications entered into the Faustian bargain of allowing a price-tag to be erected as a barrier between their work and its (tiny) intended readership, for that was the only way they could make their work public at all during the age when paper publication (and its substantial real expenses) was their only option.

Source: lwn

In Basel, Switzerland, between the 13th and 15th of January, 2006, an international symposium was held on the occasion of the hundredth birthday of Albert Hofmann.

Hoffman worked at the laboratories of Sandoz in Basel, investigating the properties of the plants Scilla and Ergot, as part of a plan to isolate and synthesize their active ingredients that could be used as pharmaceuticals. His study of the shared component of the Ergot alkaloids (Lysergic Acid), lead to the synthesis of the substance known as LSD-25, in 1938.

Though bent by age, Hoffman did participate in the conference. Apart from the speeches there were presentations of electronic music and psychedelic art by painter Alex Grey. Participants, were asked to contribute their experiences with psychoactives in the library of Erowid; the website of the member-supported society that collects information about various psychoactives, and who state their collective vision as:

A world where people treat psychoactives with respect and awareness; where people work together to collect and share knowledge in ways that strengthen their understanding of themselves and provide insight into the complex choices faced by individuals and societies alike.

Hofmannn said millions of people have taken LSD, but some had bad reactions when they took counterfeit drugs. He would like to see a modern Eleusis, the ancient Greek site that held the rituals of Eleusinian Mysteries which took place for two millennia beginning in 1500 BC. During the LSD symposium, mythologist Carl P. Ruck and chemist Peter Webster presented their research suggesting that an ergot preparation was the active ingredient for the Kykeon beverage used during the ritual.

justbeast

You clearly seem to believe that intelligence is a binary, on-or-off kind of trait, and that only humans possess it. This is where our terminology differs, fundamentally. I belive that intelligence is a smooth continuum, and that other systems (monkeys, dogs, dolphins, ant colonies, computer AI opponents in video games, etc, etc) possess some forms of intelligence -- more limited than ours, but intelligence nevertheless.

We seem to be going around in circles. I reiterate my arguments, obviously unable to clearly express what I mean, and as a result you keep reiterating the same misconceptions regarding what I said. Let's try to focus, then, on some fundamentals, and see if this gets us somewhere.

I said that if a machine acts indistinguishably from a human, then I accept that this machine is intelligent. The point of this statement, is not that I have a "binary" notion of intelligence. The point is simply to dissuade any mystical or sentimental precepts from my rejection of behaviorism. A machine does not have to be indistinguishable from a human in order for me to accept it as intelligent. I never said "and only if". I do agree with you that intelligence is a whole spectrum, and that there can be more intelligent and less intelligent entities, simpler and more complex ones, that differ from each other to great extends. I also believe that by increasing the complexity of its structure, a machine can move forward in this continuum of intelligence, and eventually reach to a point very close to where we humans stand, or even beyond. My point is that I will not accept the external similarity between human actions and machine-actions as an indication of the machine's intelligence! Not by itself!

- This machine can reach to conclusions given a set of statements, so it must be more intelligent than one that doesn't.

- No it does not! This fact alone is not enough.

- If we give it more statements of fact, then it will behave more like a human.

- Probably

- So it will become more intelligent.

- Not necessarily!

- Aha! you must be a mystic! You must believe in the existence of a soul, or in the innate superiority of the human being, and other such crap.

- No I don't. Because if you give me a system that behaves exactly as a human I will accept it as intelligent.

Our basic difference at this point is I believe this: For me intelligent behavior can be mimicked. For you intelligent behavior necessarily indicates the presence of some form of intelligence. From the following computer interaction:

   Hello how are you.
   Would you like me to add two numbers? Yes.
   Which is the first number? 12
   Which is the second? 10
   After consideration I reckon that the result is 22.

From this alone I cannot not assume a more (not one single iota more) intelligent system, than from the one below:

   expr 12 + 10
   22

Regardless of the fact that the first interaction is obviously more close to what would be considered the behavior of a human. I have to ask: What goes-on inside? And does this resemble the way humans think?

Furthermore, I believe that the problem of defining and modeling intelligence, is very very complex. Immensely complex. Any present system that exhibits behavior resembling the higher-order functions of the human intellect makes me worry that there is some sort of mimicking going-on; that it is in-fact nothing more than a glorified version of the adder shown above. On the other hand, a system that acts within its own domain, however limited, with a fluidity and adaptability resembling that of human thought-structure, such a system I would more easily call intelligent. A clever adaptive routing algorithm, yes; an expert system reasoning about philosophy, no. An operating system within its micro-domain of processes, tasks, queues, etc, can be intelligent if it can handle this ontology with a fluidity similar to that of a human. But you see, with such systems functioning in their micro-domains it becomes evident how far behind human intelligence our machines are; even in such limited universes, our machines are not yet capable of presenting an intelligent behavior. If the domains become more complex, then they are hopelessly lost.

And this brings us to what, I believe, is the core of our disagreement: Do you believe that intelligence exists? Put so blatantly the question sounds peculiar, so let me elaborate. There obviously exist intelligent entities (like you and me). These entities act in a multitude of ways, some of which we consider intelligent. The fact that we cannot exactly define which specific actions are intelligent, and which are not, is of little importance; we all agree that some of their actions are. The question is: Are these actions unrelated to each other, or are they the result of a certain structure (a certain characteristic) which we call "intelligence"? Is the term an arbitrary grouping of a set of otherwise unrelated behaviors, or does it correspond to a certain structure responsible for all these phenomena? Is it possible for an entity to exhibit one class of behaviors characterized as intelligent, without exhibiting any of the others, or are they all (or most of them) emergent properties of a certain quality (so far unknown to us) which the term refers to? Does intelligence exist, or is it simply a convenient word without much meaning? If (unlike myself) you believe that intelligence does not really exist, then I can see how your vision of sentient machines makes sense: "We have a machine that can walk in a room. Check. One type of behavior coded, a few more million to code." On the other hand, I believe that there really exists such an "intelligence quality". This quality can manifest in many levels of complexity, and possibly in several variations (thus creating a multidimensional continuum); but once it does, human-like behaviors start appearing as a result of its presence. You don't have to provide for each and every type of behavior explicitly as they are all metaphenomena of the presence of this structure. If you want do judge if a machine is intelligent, you have to ask if it possesses this quality. You have to ask if the machine's structure persuades us that it is, in fact, a manifestation of this quality. The mere actions of a machine can be misguiding.

This is a reply to justbeast's comment on my previous post. Posted as a separate entry mostly due to the apparent size limit on the replies...

some useful piece of technology comes out of the field of artificial intelligence research, gets quickly adapted and assimilated into computer science in general, and then people say "oh, /that/ isn't artificial intelligence, it's merely the so-and-so algorithm, or merely an expert system, or a neural net..."!

I never intended to imply that nothing useful has come out of the field of AI. I agree with you that several useful technologies have resulted from AI research. My only point was that we are nowhere near understanding, or modeling, human-like intelligence.

And of course the question comes to what one means by "human intelligence". You are asking if I consider "planning", "problem solving", or "reacting to the environment" real intelligence. Well, remove the quotes, and the ability of a machine to perform any of the above functions would be more than enough to make me consider it "intelligent". But by removing the quotes you have to also remove all behavioristic barriers. That is, it is not enough for a machine to simply solve a given problem, or even a given set of problems. In order to consider it humanly intelligent it has to persuade me that it maintains a mental model of similar nature and analogous complexity to the mental model maintained by a human.

Giving to a machine the predicates "All humans eventually die", "Socrates is human", and then have it reason that "Socrates eventually dies" is not a demonstration of intelligence, no matter how complicated the logical steps required to reach from the assumptions to the conclusion are. The human mind is capable of performing this form of reasoning, but this is an insignificant part of what we normally evoke when we use the term "intelligence". The "reasoning" step in this case is unbelievably minuscule compared to magnitude of the intelligence required to "impregnate" the words with meaning. In this example the machine would have to persuade me that it understands what "die" means, what "Socrates" is, what "eventually" means. That it understands that "eventually" has something to do with events and that "dies" is an event, therefore "eventually dies" is not unreasonable. For this of-course, it would have to understand what an "event" is, and be able to associate the word "event" with stimuli from its sensorial system. The same goes for "death". Then it would have to persuade me that it understands the tenses in the sentences, and that this understanding maps to a model of "time" as conceived by humans and reflected in the linguistic structure... and so on. As a matter of fact, though I lack the terminology required to argue about it, I do not believe that any intelligent process that has to do with the understanding of natural language is possible without a rather complete modeling of the human nature itself. This modeling is in turn impossible without also considering the human sensorial and perceptual system, as both are fundamental to the formation of language. It is easy, of course, to create behavioristic facades where a machine behaves as if it possesses intelligence (or understanding) while the only intelligence is that of the human interpreting the machine's output.

So, calling "intelligent" any present---or even near-future---machine that operates within a human domain (real-word domain) is, I believe, an exaggeration. There can, of course, be machines that arrange, rearrange, organize, and link-together meaningless objects in such ways as to help a human intelligence extract meaning from them; this is very useful, but it is still the human that is intelligent, not the machines. So, half-joking, half being serious, I find the maxim "If it works, it ain't AI!", quite accurate.

Do I believe that there is no "hope" for AI? There is, as long we understand how pre-embryonic our state of understanding is. A couple of years ago I was truly fascinated by a book by Douglas Hofstadter, where it presented some of the research results of his "Center for Research on Concepts and Cognition". Instead of grandiose aims for building "intelligent" computers that solve strategic conflicts, or predict the future of global-economies, he focused instead on the attempt to understand the fundamental processes behind human intelligence. To do this he concentrated on what he calls "micro-domains"; pseudo-universes with ontologies so simple that it is possible---at present---to construct programs that truly live within them, and exhibit behaviors one could characterize as genuinely intelligent. One of his most elaborate programs, for example, does not prove scientific theorems, neither it argues about Nitchean values: Given "abc --> def" it can argue that "kml --> nop" , and that "xyz --> abc", but also that sometimes "xyz --> uvw" (and it can also say that it doesn't feel very comfortable about the last one). Without claiming to be an expert, for some reason I do believe that this kind of research corresponds much better to the true state of affairs regarding our understanding of human intelligence.

In a previous post I quoted a part of Vernor Vinge's essay, where he speculates that humanity might be approaching some monumental turn-point, that could very well mean the end of history as we know it. Recently I had a discussion about this with a friend of mine, and today I decided to write down some of my thoughts on the subject; What follows is mostly a stream of consciousness, of the kind one produces when he has a drink, or a smoke, too many. Don't expect everything to be totally logical, fully supported, or even consequential...

So, do I really believe that we are---as a result of the pace of technological evolution imposed by Moore's law---really heading towards a singularity horizon that will be intercepted within our lifetimes? In one word, I believe the answer is no, we 're not even close to this danger, at least in the sense outlined by Vernor Vinge. That is, in the sense of creating an artificial intelligence with capabilities in the same "order of magnitude" as those of the human intellect. The current state of AI research is Bextremely pre-embryonic, for this fear to materialize. First there are several "macroscopic" reasoning engines (sometimes described as "expert systems") that from a distance, seem to exhibit a rather "intelligent" behavior; in reality though, they have almost nothing to do with intelligence, and attributing to them the property of "AI" is simply a historical accident; they are nothing more that glorified "predicate calculus" engines, not really more intelligent than a refrigerator, or a library cataloging system. They may handle facts in a seemingly "intelligent" manner, but in effect they merely juggle tokens, empty of meaning, in ways totally predefined by their hardcoded "reasoning systems"; the only intelligent thing in the behavior of an expert system is the way in which the human operators interprets the results and attribute meaning to them. So we may have computers that are close to winning chess grandmasters, but in fact these computers do not have clue about the game itself. In the other end of the spectrum we have neural-net research which, at the current state, can produce little more than glorified signal-processing nodes, or elementary automata, that could---after much evolution---perform functions similar to the human parasympathetic nervous system, but higher intelligence... please! In the middle ground we have research like this of Douglas Hofstadter and his Fluid Analogies Research Group, who seems to really understand the problems of emulating intelligence, and what becomes obvious if you follow their line of thought is that, presently, we don't fucking understand what intelligence is. We don't have a clue of how the phenomenon "works", on what elementary laws it is based on, and what it really means for something to be intelligent. All we can do is create virtual microcosms, and processes within them that exhibit a behavior characterized by a fluidity normally attributed to intelligence. By studying these micro-domains we may some day come to understand what intelligence is, but this will take much work, and Moore's law is irrelevant to this kind of evolution.

So it seems to me, that there is no chance in hell that we can develop a realistic theoretical model of higher intelligence within our lifetimes, regardless of the amount of processing-power available to us; processing power is irrelevant, as we still don't know what questions to ask, and the application of processing power can only give answers if you know the questions. But what if there is a shortcut? What if we can simulate intelligence without really understanding it? Given a massive amount of processing power we could simulate the operation of, say, the human brain, before we come to understand how higher-intelligence works. This will require a lot of computing power, but Moore's law is on our side this time. Could this be the... express to the Singularity? I still don't think so. You see the simple word "simulation" hides a lot of intricate details, and makes things sound so much simpler than they really are. What it hides is that there are several abstraction levels at which one can simulate a complex system, and there is a vicious trade-off between the depth of the simulation level, the knowledge of the system needed to achieve simulation at this level, and the computing power required. The deeper you go in the stack of abstraction levels, the less you need to know about the simulated system, but the more computing power you require in order to achieve the simulation. It may very well be that the computing-power consumption increases exponentially as you go deeper in the stack of abstraction levels (as a result of the existence of non-polynomial-complexity problems, which cannot be bypassed), thus canceling the effects of Moore's law.

Let's think of an example, where we are called to simulate something much simpler than the human brain (or the human intellect): A digital computer. It is a fact that a modern workstation can easily simulate a computer based on an 8-bit microprocessor, like the Zilog Z80, operating at 4.77MHz, in real time. Or is it? It all depends on the level of abstraction upon which the simulation will be based: If one had to simulate the behavior of BASIC programs running on a Sinclair ZX81, then a modern workstation could do it, several orders of magnitude faster than the real thing. To construct such a simulator one would have to simulate, in detail, the specific semantics of the BASIC language found in the target machine, and nothing more. But for these semantics to be extracted one would have to understand what parts of the simulated machine are irrelevant to this specific level of abstraction, and how they nevertheless contribute to it: The power consumption of the machine's components is irrelevant, the lithographic process used to etch the microprocessor's silicon is irrelevant, the width of the coper traces on the PCB are irrelevant, the way the NAND gates are constructed from transistors inside the processor are irrelevant, architecture of the processor (e.g. number of registers, width of registers, arithmetic model) may or may not be relevant depending on whether the BASIC interpreter exposes some of these properties, or not. Also irrelevant is the parser technology used in the BASIC interpreter, the exact implementation of the interpreter, the memory management model of the runtime environment, the interrupt handling policy, and so on. Now imagine that the computer was delivered to us as an "alien artifact", without user, service, or engineering documentation (as is the case for every natural object). All this "irrelevant" information would have to be exposed and understood before it would be possible to simulate the machine at the level suggested above. Alternatively one could decide to simulate the machine at a lower level, say for example at what is called the "architectural level". For this there is no need to understand all the details about the BASIC language, about the way the interpreter is constructed, about the interrupt handling policies, about the operating system of the machine, or the memory management model, and so on. In this case all these could be simulated. But all these extra simulated stuff induce a performance costs to the simulator. It would no longer be possible to run the simulator several orders of magnitude faster than real-time on a modern workstation. One could of course go even lower in the stack of abstraction-levels and simulate the machine at what is known as the "RT" (register transfer) level; this will do away with having to understand most parts of the architecture of the processor (e.g. the semantics of the machine language). At this level the machine could marginally be simulated by a modern workstation in real-time. And we can go even lower than this: Simulate the processor at "gate-level". Now you don't have to understand the processor architecture at all; all you need is to extract a logic-diagram of the processor's circuit (without understanding it). But now your simulator runs slower than realtime, probably many times slower, and this is the lowest level one would practically attempt to simulate a microprocessor today. Theoretically, one could attempt a simulation at the electrical level (simulate every transistor), or even at the quantum-mechanical level (simulate quantum effects in the body of the semiconductor), but any of these levels would be exponentially more costly than the previous one. So the answer to the question of whether it is possible to simulate an 8-bit processor on a modern workstation is: "it depends on the abstraction-level on which the simulation will be based". The deeper the abstraction level is, the less are the things you have to understand about the simulated object, but the more are the things you have to simulate, and thus the more costly the operation becomes.

This reasoning can be extended to the simulation of the human brain or of the human intelligence. In order to simulate the human intelligence (which I believe could be computationally feasible using a near-future computer), one would require a concrete model of it, and I wouldn't expect this to be developed within our lifetimes. In order to simulate the human brain, at, say, the chemical level, or even at some abstract neuro-physiological level, the processing power required would be immense (enough to put it beyond our lifetimes). Furthermore there is the problem of "interfacing", which becomes more and more complex as you descent in the stack of simulation levels: The lower the simulation level lies in the "stack", the harder it is to interface the simulator with external entities.

So, fascinating as I find Vernor Vinge's argument about humanity crossing a technological singularity, I don't believe you should rush to liquidate your assets, burn your computer (as a symbolic gesture), and go live in a remote retreat, away from digital technology, or even away from electricity... About your great-grandchildren, I don't know.

It is amazing how easily people get carried-away by scientific-sounding theories that promise hidden knowledge---usually of incredible value---and throw-in a good deal of conspiracy paranoia in order to justify why this knowledge, or any of its fruits, are not widely available.

One of the most prominent of these pseudo-scientific cults are the followers of Nikola Tesla (there are many around). Being confronted by yet-another Tesla believer, and trying to find a concrete scientific refutation of Tesla's wireless power-transmission theories, I came upon this interesting article, written by Paul Nicholson and titled The real science of non-Hertzian waves.

I found the article hosted in a site called Teslamania, which features several "Teslian" inventions, and documents several experiments that would have made the Serbian inventor proud (mind you: this is not a nutcase site; the machines presented, and the experiments described do work!). Among other things you will be able to find:

• Evidences of the operation of powerful Coin Shrinking machines
• Captured Lightning displays
• Tesla coils, generating impressive sparks

I've been reading Charles Stross' books the last few days (namely "Atrocity Archives", and "Toast", and I'm just now starting "Singularity Sky"). I greatly enjoyed them, in part because they are in fact very good, and in part because it has been some time since I've last had the pleasure to indulge myself in a treat of hard SF...

Instead of writing something about the books (you can find several reviews on the net), I will simply post a quote from Vernor Vinge's extremely relevant paper, about mankind's imminent collision with the Singularity:

[ ... ] The acceleration of technological progress has been the central feature of this century. I argue in this paper that we are on the edge of change comparable to the rise of human life on Earth. The precise cause of this change is the imminent creation by technology of entities with greater than human intelligence. [ ... ] Progress in computer hardware has followed an amazingly steady curve in the last few decades. Based largely on this trend, I believe that the creation of greater than human intelligence will occur during the next thirty years [...]

What are the consequences of this event? When greater-than-human intelligence drives progress, that progress will be much more rapid. In fact, there seems no reason why progress itself would not involve the creation of still more intelligent entities---on a still-shorter time scale. The best analogy that I see is with the evolutionary past: Animals can adapt to problems and make inventions, but often no faster than natural selection can do its work---the world acts as its own simulator in the case of natural selection. We humans have the ability to internalize the world and conduct "what if's" in our heads; we can solve many problems thousands of times faster than natural selection. Now, by creating the means to execute those simulations at much higher speeds, we are entering a regime as radically different from our human past as we humans are from the lower animals.

From the human point of view this change will be a throwing away of all the previous rules, perhaps in the blink of an eye, an exponential runaway beyond any hope of control. Developments that before were thought might only happen in "a million years" (if ever) will likely happen in the next century. [ ... ]

I think it's fair to call this event a singularity [ ... ]