From: J. Maynard Gelinas <j.maynard.gelinas.nul> Date: Fri, 10 May 2013 12:34:58 +0800 Archived: Fri, 10 May 2013 08:27:34 -0400 Subject: Critique Of Dolan's Sourcing & Conclusions Richard Dolan's UFOs And The National Security State series is highly regarded among the UFO research community and deservedly so. It's probably the best compendium of the history of 'UFOlogy' in print today. His wide range and large numbers of open sources gives the work a depth that only - perhaps - Timothy Good matched twenty five years ago with "Above Top Secret." It's good work and I'll be among the first in line to buy a copy when Vol 3 is released. But that doesn't mean Dolan's approach is beyond criticism. In fact, given his status as a leader in the community, it behooves everyone to question aspects of his work as much as possible, if only to tease out those parts where poor sourcing leads to untenable conclusions. I'd like to offer one specific instance of an almost certainly incorrect conclusion brought about from single sourcing in a statement made during the April 30th night session at the Citizen's Hearing. See Approximately from 35 to 38 minutes into the lecture. Dolan: "We're a couple of years away from computers that will challenge and literally surpass human intellect in many ways. [next slide] About a decade ago I got really turned on to the future of Artificial Intelligence. I read some books by Ray Kurzweil, who I think is a real visionary. He was the one who really started me thinking along these lines. He wrote a book about fifteen years ago called, "The Age of Spiritual Machines," which really grabbed ahold of my imagination. But not just him; many other AI people - and they're fairly mainstream, they're not like us, they actually get funding to do the things they do. They all say the same thing - with variations - which is essentially that within about twenty years from now your computer will be telling you it's a conscious, sentient, being. And you will very likely believe those claims. Now, will be be conscious the way that you are - no, probably not, but will it matter? It will seem like it. It won't need to sleep; it won't need coffee in the morning; it will have a relative IQ of 500 and, hey, as Kurzweil put it, 'it will have a God-like level of intelligence.' And, in that situation, ask yourself this: If you have a question you'd like to ask of your super-intelligent computer, let's name it 'Marvin', let's just say. [Reference to Marvin the Paranoid Android, "Hitchhiker's Guide to the Galaxy," D. Adams, fiction] You could say, "Hey Marvin, I've been thinking about all of this UFO stuff. I wonder if you could just tell me what you know." And Marvin will say whatever Marvin comes up with, but Marvin will probably say that it's real and that there's something going on. And, what is that line, I think by Schopenhauer, three stages of accepting new truths: Stage one, they ridicule you; stage two, it's attached violently; stage three, they say, 'yeah, we knew it all along'." This is a problematic assertion, which I'd like to challenge from a number of perspectives. To begin is the statement that '...many other AI people ... They all say the same thing - with variations - which is essentially that within about twenty years from now your computer will be telling you it's a conscious, sentient, being.' Actually, no. Few academic AI researchers would be willing to make such a statement on the record. And many would challenge it, not from a time standpoint but from a feasibility standpoint altogether. A good lay article challenging Kurzweil's latest book, A New Theory Of Mind was published in The New Yorker last year. "Kurzweil is so confident in his theory that he insists it simply has to be correct. Early in the book, he claims that =93the model I have presented is the only possible model that satisfies all the constraints that the research and our thought experiments have established.=94 He later declares that =93there must be an essential mathematical equivalence to a high degree of precision between the actual biology and our attempt to emulate it; otherwise these [A.I.] systems would not work as well as they do.=94 "What Kurzweil doesn=92t seem to realize is that a whole slew of machines have been programmed to be hierarchical-pattern recognizers, and none of them works all that well, save for very narrow domains like postal computers that recognize digits in handwritten zip codes. This summer, Google built the largest pattern recognizer of them all, a system running on sixteen thousand processor cores that analyzed ten million YouTube videos and managed to learn, all by itself, to recognize cats and faces=97which initially sounds impressive, but only until you realize that in a larger sample (of twenty thousand categories), the system=92s overall score fell to a dismal 15.8 per cent. http://tinyurl.com/cderjox Here is Jaron Lanier, early pioneer of 'virtual reality' and well known computer scientist: "One question I have about Ray's exponential theory of history is whether he is stacking the deck by choosing points that fit the curves he wants to find. A technological pessimist could demonstrate a slow-down in space exploration, for instance, by starting with sputnik, and then proceeding to the Apollo and the space shuttle programs and then to the recent bad luck with Mars missions. Projecting this curve into the future could serve as a basis for arguing that space exploration will inexorably wind down. I've actually heard such reasoning put forward by antagonists of NASA's budget. I don't think it's a meaningful extrapolation, but it's essentially similar to Ray's arguments for technological hyper-optimism. It's also possible that evolutionary processes might display local exponential features at only some scales. Evolution might be a grand scale "configuration space search" that periodically exhibits exponential growth as it finds an insulated cul-de-sac of the space that can be quickly explored. These are regions of the configuration space where the vanguard of evolutionary mutation experimentation comes upon a limited theater within which it can play out exponential games like arms races and population explosions. I suspect you can always find exponential sub processes in the history of evolution, but they don't give form to the biggest picture." http://www.edge.org/discourse/jaron_answer.html To explain how Lanier's statement relates to Kurzweil's thesis, one should understand that a foundational argument of Kurzweil's is that technology improvement is occurring at exponential rates, and that this improvement is mitigated by an evolutionary path that leads to greater functional complexity. Lanier is challenging this based on historical events where exponentially increasing returns to technology ultimately petered out. Because _nothing_ increases at exponential rates forever. Not bacteria in a petri dish - the food runs out; and not technological improvements. In computing, physical constraints of energy consumption and systems shrinkage, as seen in the limits of Moore's Law as we hit lithographic printing sizes that border on 10nm trace widths - two orders of magnitude away from the average size of an atom - show that exponentially increasing returns from computing performance will not continue without a radical shift in computing architecture. In the academic literature, here is another critique of Kurzweil's approach: "Goertzel, B. (2007). Human-level artificial general intelligence and the possibility of a technological singularity. A reaction to Ray Kurzweil's The Singularity Is Near, and McDermott's critique of Kurzweil. Artificial intelligence, 171, 1161-1173. Critique of Kurzweil by McDermott and replies by Goertzel: 1) Kurzweil does not give any proof that an AI Singularity is upon us. - Kurzweil does not claim to do so. 2) Even if we succeed in scanning the brain into a computer, we still won't understand human intelligence. [p1167] - Kurzweil's predictions are far in the future and take into account this learning process. - An uploaded, digitized human brain can be more easily manipulated and studied. [p1168] 3) Kurzweil says that machines will augment their capacities without limit, but this is unrealistic. - That is true, the laws of physics might impose their own limitations but progressive self-improvement might not get boxed in by universal laws. - An analogy for this augmentation can be the scientific community, a self-improving collective intelligence composed internally of human-level intelligences. [p1169] Kurzweil's route toward Singularity-enabling AGI can be summed up as scanning human brains, creating brain emulations, studying these emulations and creating AGI systems capable of self- improvement." The actual paper is behind a paywall, so here is a summary of the critique: http://www.jimdavies.org/summaries/Goertzel2007.html To understand this argument, one must recognize what Kurzweil proposes. He argues that by creating scans of the brain it should be possible to simulate the structure of those scans in a computer and replicate general intelligence in this way. Once the system reaches a threshold of intelligence, it should then be possible to assign the system the task of increasing its own intellectual capacity by studying itself. In time, via evolutionary methods, it would increase in intelligence to levels not recognized by humans who had originally developed the system. Stepping up from the individual neuron, he argues that collections of neurons work as "pattern recognizers" (see New Yorker critique), which when hierarchically connected in complex patterns form what we perceive as 'self directed conscious awareness'. As noted in prior critiques, this idea is not new. The founder of the MIT AI lab Marvin Minsky promoted a similar idea in "Society of Mind", way back in the early 1980s. So too did Tufts Philosopher, Daniel Dennett in "Consciousness Explained", published in the late 1980s. Ray Kurzweil used a similar method in developing his commercial Optical Character Recognition system back in the late 1970s (a tool for converting text to speech for the blind). This approach is how speech recognition engines have worked since the early 1990s. It uses stochastic statistical analysis for recognizing a pattern to then select best fit solutions for a desired result. It works, but as anyone who has used speech recognition systems knows, with a high error rate in comparison to actual brains and real ears. Improvement in the technique has not come from better algorithms but faster computers and better microphones. Computers are reaching their limits of raw speed improvement. Speech recognition is but one small task a general intelligent agent must perform. Each hierarchy of problem solving agent must interact with a tree of agents underneath, with errors in calling up the correct pattern of agents for a particular problem compounding with systems complexity. The hope is that massive parallelization in compute engines will solve the 'physical limits of computing' problem. And that brain scans at the neuron level will solve the problem of patterning these agents in a way that replicates human cognition. But that's a hope. And it's one that most certainly isn't a sure bet. But there are ongoing attempts at this. For example, the Blue Brain project is attempting to scan an entire mouse brain and model it using an IBM BlueGene computer. They've succeed at scanning a neocortical column of a mouse brain (about 1mm cubed) and have published results that show similarities in input and output between their simulation and a live mouse brain. But it takes an entire room full of computers to do so, and the simulation doesn't run real-time. http://bluebrain.epfl.ch/ The Obama Administration has announced a funding plan to attempt to map the entire human brain. The project is considered on the par of the Human Genome Project in funding and interdisciplinary complexity. But even if successful, that doesn't mean it will generate useful results. This is pure research, not applied. http://tinyurl.com/bvqh9s5 And then there's a problem with the central thesis that computers using the methods of 'self-insight' might somehow improve themselves intellectually. From a historical perspective, philosophers going back to Aristotle, Plato, and up through Hume and Freud have proposed self-insight as a mechanism for improvement. But that hasn't worked too well. At least not in terms of finding correct results or orders of magnitude improvement. What's the upshot of this as far as Richard Dolan's claims go? 1) Assuming Kurzweil is right, the timeframe Mr Dolan proposed is an incredibly risky claim. Most professional AI researchers would NOT be willing to predict on the record anything like, 'in 20 years and AI will surpass human intelligence'. 2) It's a bad idea to assume Kurzweil is right, because there is too much we don't know about how brains - or even general intelligence - actually works. For the last fifty years AI researchers - very much like fusion researchers - have been claiming that with incremental advances we'd achieve the AI Holy Grain in short order. And they've been wrong. In fact, fusion actually has a path to success with ITER (International Thermonuclear Experimental Reactor) that actually has a theoretical foundation for net positive energy production (more energy out than put in to start the reaction). AI has nothing of the sort but hopes and dreams. There is no hard theoretical foundation to back up Kurzweil's claims. Conclusion So I think Dolan is wrong in this prediction. It's OK to be wrong, and I write this note with all due respect to the gentleman. But the reason why I point to it is because it suggests a level of cherry-picking in his sources that leads to excessively hopeful and bombastic claims. Individually, this response is a minor criticism. If Dolan were to remove this anecdote from his logic chain, it wouldn't destroy his 'Breakaway Civilization' thesis. But it does - in a small way - chip away at it. And I think UFO researchers should critique the foundations of his claims in order to determine the likelihood of it being a rational conclusion. Not because Mr Dolan's work is poorly done or because he's a bad guy, but because it's a big conclusion that rests on a large foundation of claims across many disparate sources that individually might not be tenable as well. It's appropriate to challenge these foundational claims in order to determine if the logic chain holds overall. --M Listen to 'Strange Days... Indeed' - The PodCast At: http://www.virtuallystrange.net/ufo/sdi/program/ These contents above are copyright of the author and UFO UpDates - Toronto. They may not be reproduced without the express permission of both parties and are intended for educational use only.
[ Next Message | Previous Message | This Day's Messages ]
This Month's Index |
UFO UpDates - Toronto - Operated by Errol Bruce-Knapp