* Dr. Arne Babenhauserheide <arne_...@web.de> [2025-03-24 02:00]: > Jean Louis <bugs@gnu.support> writes: > > It surprises me when someone as integral to programming seems lacking > > in basic social awareness or comprehension. > > > > If you have examples, show them. > > I told you about Alpha Go inventing new methods of playing Go. Methods > that human players took up and that changed how people play and learn go. > > If you can’t accept that as innovation, there is nothing you’ll accept > and this discussion is useless.
It is very sad to see a doctor with such dismissive tone answering to me, instead of providing some facts, references or Dr.-like reasoning. I did not expect that you relied solely on the media (of AlphaGo) to form your opinion, Dr. Arne. It is not about me not accepting, but me, I am searching for answers which I am not finding. The question was about examples, I have asked for examples of true innovation by LLM, and there were no examples shown. At that time point I was curious. Now I am curious about the future state, about new principles, rather than current, that are going to make the AI computers generate some innovations. And because nobody found examples, it was me who found references to works of other people teaching me more about the subject. I have gained little bit more of the fundamental knowledge related to current state of. I see you did not read references I have sent. As from your caliber I expect it. Let's go again. I understand that large language models (LLMs) operate using probabilistic methods; however, this is something I'm not questioning. - If users don’t understand any aspects of the probabilistic algorithms used in large language models (LLMs), they may find them amazing and refer to it as "intelligent" behavior. - When user finds something interesting, some new knowledge to the user on Wikipedia, or any other body of knowledge, can we consider that finding an innovation, just because one user didn't know about it, but it was available somewhere in the encyclopedia? I have already given references on that subject: Simple question that Microsoft Phi-4 Large Language Model (LLM) cannot answer: https://gnu.support/large-language-models-llm/Simple-question-that-Microsoft-Phi-4-Large-Language-Model-LLM-cannot-answer.html Where there are 3 good references on the end, and one is: https://www.researchgate.net/publication/370105098_Creative_through_AI_How_artificial_intelligence_can_support_the_development_of_new_ideas where it says: "Artificial intuition" When AlphaGo bet the world champion in the game of Go it played a move that no one would have expected. As AlphaGo developed its own strategy, that was clearly not learned from humans, it was discussed whether it showed intuition. Why? Because humans require intuition to play the game of Go, that means they not only consider what is logical but also what feels right. However, only because a human requires intuition to play Go, this does not mean that a machine requires it, even if it plays a move that seems intuitive to us. In fact, DeepMind revealed that AlphaGo played the surprising move because it exactly knew that the probability that a human would play this move was 1:10,000. “This explanation of an AI “deciding” to per- form a particular move, is rooted in logic rather than intuition”. So while AI might be able to mimic human intuition, it does not seem as if it would be able to show the same form of intuition as humans. Anyways, we do not even fully understand the human intuition process yet." Where it references: Johanssen, J.; Wang, X. (2021): Artificial Intuition in Tech Journalism on AI: Imagining the Human Subject. Human-Machine Communication, 2, 173-190. https://stars.library.ucf.edu/cgi/viewcontent.cgi?article=1023&context=hmc quote: "In fact, as Paolo Bory (2019) discusses, the move that led to the defeat of the world champion by the original AlphaGo (the previous version of AlphaGo Zero) was considered as creative, even beautiful by commentators. It was also seen as very un-humanlike." and "The other two quotations about AlphaGo Zero and DeepStack reproduced earlier similarly describe intuition as logical and math-based. Artificial intu- ition is seen as something that can surpass human intuition in advancing a form of intuition that is unlike those of humans. However close to intuition the moves by AlphaGo Zero or DeepStack may have been, the data that they trained themselves with were still accessible to them. Both systems consciously learned from data." and it is alright to counter those statements of those two people: - Jacob Johanssen from Institute of Business, Law and Society, St. Mary’s University, Twickenham, London, UK, and - Xin Wang from College of Arts and Media, Tongji University, Shanghai, People’s Republic of China by provding some facts. That was the initial question of me. Do you facts, not just adopted impressions from other people? As you see in the comment, indications are that observers got excited and that it created some public sensation. I don't think that average computer gamer would get excited because there is computer making some moves. It was programmed to work by certain principles, and among many others, even the company confirmed it was based on probabilistic computation. For probability to work, computer must have data. For data to exist it must be created by what is known in our human world. Probabilistic results can thus give responses only from the data that already exists. I personally do not believe this will be our stopping point in progress because precisely, it's the principle of generating probabilistic responses that impedes genuine innovation from Large Language Models. I really hope, as I am personally not capable of doing it in this moment, that developers are to find out new ways, something better, that is not based on the existing data only. Again I think that computer will need to have much more data, maybe even 100 times more than what is now, in order to use some different principles to truly create. -- Jean Louis --- via emacs-tangents mailing list (https://lists.gnu.org/mailman/listinfo/emacs-tangents)
