CG October 2023

cg@lists.iccs-conference.org

3 participants
15 discussions

Re: [ontolog-forum] Addendum to (Generative AI is at the top of the Hype Cycle. Is it about to crash?

by John F Sowa

Alex, Thanks for the list of applications of LANGUAGE-based LLMs. It is indeed impressive. We all agree on that. But mathematics, physics, computer science, neuroscience, and all the branches of cognitive science have shown that natural languages are just one of an open-ended variety of left-brain ways of thinking. LLMs haven't scratched the surface of the methods of thinking by the right brain and the cerebellum. The left hemisphere of the cerebral cortex has about 8 billion neurons. The right hemisphere has another 8 billion neurons that are NOT dedicated to language. And the cerebellum has about 69 billion neurons that are organized in patterns that are totally different from the cerebrum. That implies that LLMs are only addressing 10% of what is going on in the human brain. There is a lot going on in that other 90%. What kinds of processes are happening in those regions? Science makes progress by asking QUESTIONS. The biggest question is how can you handle the open-ended range of thinking that is not based on natural languages. Ignoring that question is NOT scientific. As the saying goes, when the only tool you have is a hammer, all the world is a nail. We need more tools to handle the other 90% of the brain -- or perhaps updated and extended variations of tools that have been developed in the past 60+ years of AI and computer science. I'll say more about these issues with more excerpts from the article I'm writing. But I appreciate your work in showing the limitations of the current LLMs. John ---------------------------------------- From: "Alex Shkotin" <alex.shkotin(a)gmail.com> John, English LLM is the flower on the tip of the iceberg. Multilingual LLMs are also being created. The Chinese certainly train more than just English-speaking LLMs. You can see the underwater structure of the iceberg, for example, here https://huggingface.co/datasets (1). Academic claims against inventors are possible. But you know the inventors: it works! It's funny that before that hype LLM meant Master of Laws:-) Alex (1)

1 year, 5 months

Re: [ontolog-forum] Addendum to (Generative AI is at the top of the Hype Cycle. Is it about to crash?

by John F Sowa

Alex, I'm glad that we finally agree. The main problem with the LLM gang is that they don't ask the fundamental questions: How is this new tool related to the 60+ years of R & D in AI, computer science, and the immense area of the multiple cognitive sciences? For example, Stanislas Dehaene and his students and colleagues have shown that there are multiple languages of thought, not just one. And every method of thinking has a different view of the world, of life, and of the fundamental methods of thought. For example, thinking and working with and about mathematics, visual structures, music, games, gymnastics, flying an airplane, building a bridge, plowing a field, etc., etc., etc. activate totally different areas of the brain than speaking and writing English. A brain lesion that knocks out one region may leave other regions unscathed, and it may even enhance performance in those other regions. The LLM gang knows nothing about these issues. They don't ask the right questions. In fact, they're so one-sided that they don't even know what questions they should be asking. Somebody has to educate them. The best way to start is for us to ask the embarrassing questions. Just before I read your note, I came across another article by the Dehaene gang: https://www.science.org/doi/pdf/10.1126/sciadv.adf6140 Does the visual word form area split in bilingual readers? Minye Zhan, Christophe Pallier, Aakash Agrawal, Stanislas Dehaene, Laurent Cohen In expert readers, a brain region known as the visual word form area (VWFA) is highly sensitive to written words, exhibiting a posterior-to-anterior gradient of increasing sensitivity to orthographic stimuli whose statistics match those of real words. Using high-resolution 7-tesla fMRI, we ask whether, in bilingual readers, distinct cortical patches specialize for different languages. In 21 EnglishFrench bilinguals, unsmoothed 1.2-millimeters fMRI revealed that the VWFA is actually composed of several small cortical patches highly selective for reading, with a posterior-to-anterior word-similarity gradient, but with near-complete overlap between the two languages. In 10 English-Chinese bilinguals, however, while most word-specific patches exhibited similar reading specificity and word-similarity gradients for reading in Chinese and English, additional patches responded specifically to Chinese writing and, unexpectedly, to faces. Our results show that the acquisition of multiple writing systems can indeed tune the visual cortex differently in bilinguals, sometimes leading to the emergence of cortical patches specialized for a single language. This is just one of many studies that show why LLMs based on English may be inadequate for ways of thinking in other languages or in non-linguistic or pre-linguistic ways of thinking, working, living, etc. Furthermore, language is a left-brain activity, and most of our actions and ways of behaving and working are right-brain activities. The current LLMs are based on ways of thinking by an English speaker whose right brain was destroyed by a stroke. None of the writings about LLMs ask or even mention these issues. In this mini-series on generative AI, we have to ask the embarrassing questions. Any science that avoids such questions is brain dead. John ---------------------------------------- From: "Alex Shkotin" <alex.shkotin(a)gmail.com> JFS: "Now is the time to ask deeper questions." Exactly, and these questions should be scientific :-) And we have a scientific phase with these creatures, GenAI in general and LLM in particular: experiments ;-) Alex

1 year, 5 months

Generative AI is at the top of the Hype Cycle. Is it about to crash?

by John F Sowa

Anatoly, Stephen, Dan, Alex, and every subscriber to these lists, I want to emphasize two points: (1) I am extremely enthusiastic about LLMs and what they can and cannot do. (2) I am also extremely enthusiastic about the 60+ years of R & D in AI technologies and what they have and have not done. Many of the most successful AI developments are no longer called AI because they have become integral components of computer science. Examples: compilers, databases, computer graphics, and the interfaces of nearly every appliance we use today: cars, trucks, airplanes, rockets, telephones, farm equipment, construction equipment, washing machines, etc. For those things, the AI technology of the 20th century is performing mission-critical operations with a level of precision and dependability that unaided humans cannot achieve without their help. Fundamental principle: For any tool of any kind -- hardware or software -- it's impossible to understand exactly what it can do until the tool is pushed to the limits where it breaks. At that point, an examination of the pieces shows where its strengths and weaknesses lie. For LLMs, some of the breaking points have been published as hallucinations and humorous nonsense. But more R & D is necessary to determine where the boundaries are, how to overcome them, work around them, and supplement them with the 60+ years of other AI tools. Anatoly> When you target LLM and ANN as its engine, you should consider that this is very fast moving target. E.g. consider recent work (and imagine what can be done there in a year or two in graph-of-thoughts architectures) . . . Yes, that's obvious. The article you cited looks interesting, and there are many others. They are certainly worth exploring. But I emphasize the question I asked: Google and OpenAI have been exploring this technology for quite a few years. What mission-critical applications have they or anybody else discovered and implemented? So far the only truly successful applications are in MT -- machine translation of languages, natural and artificial. Can anybody point to any other applications that are mission critical for any business or government organization anywhere? Stephen Young> Yup. My 17yo only managed 94% in his Math exam. He got 6% wrong. Hopeless - he'll never amount to anything. The LLMs have been successful in passing various tests at levels that match or surpass the best humans. But that's because they cheat. They have access to a huge amount of information on the WWW about a huge range of tests. Bur when they are asked routine questions for which the answers or the methods for generating answers cannot be found, they make truly stupid mistakes. No mission-critical system that guides a car, an airplane, a rocket, or a farmer's plow can depend on such tools. Dan Brickley> Encouraging members of this forum to delay putting time into learning how to use LLMs is doing them no favours. All of us love to feel we can see through hype, but it’s also a brainworm that means we’ll occasionally miss out on things whose hype is grounded in substance. Yes, I enthusiastically agree. We must always ask questions. We must study how LLMs work, what they do, and what their limitations are. If they cannot solve some puzzle, it's essential to find out why. Noticing a failure on one problem is not an excuse for giving up. It's a clue for guiding the search. Alex> I'm researching how LLMs work. And we will really find out where they will be used after the hype in 3-5 years. Yes. But that is when everybody else will have won the big contracts to develop the mission-critical applications. Now is the time to do the critical research on where the strengths and limitations are. Right now, the crowd is having fun building toys that exploit the obvious strengths. The people who are doing the truly fundamental research are exploring the limitations and how to get around them. John

1 year, 5 months

Generative AI is at the top of the Hype Cycle. Is it about to crash? (was Consciousness

by John F Sowa

Ricardo, Alex, Anatoly, and anybody who is working with or speculating about LLMs for generative AI, LLMs have proved to be valuable for machine translation of languages. They have also been used to implement many kinds of toys that appear to be impressive. But nobody has shown that LLM technology can be used for any mission critical applications of any kind -- i.e. any applications for which a failure would cause a disaster (financial or human or both). Question: Companies that are working on generative AI are *taking* a huge amount of money from investors. Have any of them produced any practical applications that are actually *making* money? Generative AI is now at the top of the hype cycle. That implies an impending crash into the trough of disillusionment. When will that crash occur? Unless anybody can demonstrate applications that make money, the investors are going to be disillusioned. To Ricardo> Those are interesting hypotheses about consciousness in your note below. But none of them have any significant implications for AI, ontology, or the possibility of money-making applications of LLMs. One important point: Nobody suggests that anything in the cerebellum is conscious. The results from the cerebellum that are reported to the cortex are critical, especially since the cerebellum has more than four times as many neurons as the cerebral cortex. There is also strong evidence that the cerebellum is essential for complex mathematics. (See Section 6.pdf) Implication: AI methods that simulate processes in the cerebral cortex (such as natural language processing by LLMs) cannot do the heavy duty computation that is done by neurons in the cerebellum -- and that includes the most complex logic and mathematics. See the summary in Section6.pdf and my other references below. John ---------------------------------------- From: "Ricardo Sanz" <ricardo.sanz.bravo(a)gmail.com> Hi, JFS>> What parts of the brain are relevant for any sensation of consciousness? So far, the question of neural correlates of consciousness (NCC) is still unresolved. This was the theme of the Chalmers-Koch wager. There are too many theories and no relevant enough experimental data to decide. The most repeated theory is that consciousness is hosted in thalamo-cortical reentrant loops. The cortex (sensorimotor data processor) and the thalamus (the main relay station of the brain). This is yet to be demonstrated. Another widely repeated theory was that the NCC was a train of 40hz signal waves across the whole brain. The boldest to me, however, is the quantum macroscopic coherence in the axon microtubules. This is called the Orchestrated Objective Reduction theory (Orch-OR). Best, Ricardo On Mon, Oct 2, 2023 at 5:40 AM John F Sowa <sowa(a)bestweb.net> wrote: That article shows several points: (1) The experts on the subject don't agree on basic issues. (2) They are afraid that too much criticism of one theory will cause neuroscientists to consider all theories dubious. (3) They don't have \clear criteria for what kinds of observations would or would not be considered relevant to the issues. But I want to mention some questions I have: What parts of the brain are relevant for any sensation of consciousness? All parts? Some parts? Some parts more than others? Which ones? From common experience, we know that complex activities require a great deal of conscious attention when we're first learning them. But after we learn them, they become almost automatic, and we can perform them without thinking about them. Examples: Learning to ski vs. skiing smoothly on moderate hills vs skiing on very steep or complex surfaces. The same issues apply to any kind of skill: driving a car, driving a truck, flying a plane, swimming, dancing, skating, mountain climbing, working in any profession of any kind -- indoors, outdoors, on a computer, with any kinds of tools, instruments, conditions, etc. In every kind of skill, the basic techniques become automatic and can be performed with a minimum of conscious attention. There is strong evidence that the effort in the cerebrum (/AKA cerebral cortex) is conscious, but expert skills are controlled by the cerebellum, which is not conscious. There is brief discussion of the cerebellum in Section6.pdf (see the latest excerpt I sent, which is dated 28 Sept 2023). For more about the role of the cerebellum, see the article and video of a man who was born without a cerebellum and survived: A Man's Incomplete Brain Reveals Cerebellum's Role In Thought And Emotion. https://www.npr.org/sections/health-shots/2015/03/16/392789753/a-man-s-inco…; John

1 year, 5 months

Consciousness

by John F Sowa

That article shows several points: (1) The experts on the subject don't agree on basic issues. (2) They are afraid that too much criticism of one theory will cause neuroscientists to consider all theories dubious. (3) They don't have \clear criteria for what kinds of observations would or would not be considered relevant to the issues. But I want to mention some questions I have: What parts of the brain are relevant for any sensation of consciousness? All parts? Some parts? Some parts more than others? Which ones? From common experience, we know that complex activities require a great deal of conscious attention when we're first learning them. But after we learn them, they become almost automatic, and we can perform them without thinking about them. Examples: Learning to ski vs. skiing smoothly on moderate hills vs skiing on very steep or complex surfaces. The same issues apply to any kind of skill: driving a car, driving a truck, flying a plane, swimming, dancing, skating, mountain climbing, working in any profession of any kind -- indoors, outdoors, on a computer, with any kinds of tools, instruments, conditions, etc. In every kind of skill, the basic techniques become automatic and can be performed with a minimum of conscious attention. There is strong evidence that the effort in the cerebrum (/AKA cerebral cortex) is conscious, but expert skills are controlled by the cerebellum, which is not conscious. There is brief discussion of the cerebellum in Section6.pdf (see the latest excerpt I sent, which is dated 28 Sept 2023). For more about the role of the cerebellum, see the article and video of a man who was born without a cerebellum and survived: A Man's Incomplete Brain Reveals Cerebellum's Role In Thought And Emotion. https://www.npr.org/sections/health-shots/2015/03/16/392789753/a-man-s-inco… John ---------------------------------------- From: "Nadin, Mihai" <nadin(a)utdallas.edu> Dear and respected colleagues: The issue does not go away: https://theconversation.com/consciousness-why-a-leading-theory-has-been-bra… I have no dog in this race! Mihai Nadin https://www.nadin.ws https://www.anteinstitute.org Google Scholar

1 year, 5 months

2025

2024

2023

2022

2021

CG October 2023