Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data

understanding of the world is, again, based on the fact that we are embodied beings. Now, these are not newly edited books, and no, there is no real reason to think that because they are all +30 years old or even more, they are outdated. They are not. If you only look at the publication date, then Ramon y Cajal works would be totally crap (by the way, still one of the most cited works in neuroscience). It is from early 1900s. To complete the picture a bit, you could read: On the mode of existence of technical object, by Gilbert Simondon Technics and Time, by Bernard Stiegler Meditation on the technique, by Jose Ortega y Gasset The question concerning technology, by Martin Heidegger These works will give an understanding of how technique in general (technology in particular) is completely anthropomorphisized, which is what ultimately leads to the assumptions present in the AI research programme. Also, A history of philosophy, by Frederick Copleston. Although extensive, reading volume I (greek and roman philosophy) is essential. More citations (again, if you really measure the quality or relevance of a philosophical/scientific work by its publication date, you are missing the picture): Arbib, M. A. (2025).* Artificial intelligence meets brain theory (again). Biological Cybernetics, 119, 16. https://doi.org/10.1007/s00422-025-01013-5 Farkaš, I., Vavrečka, M., & Wermter, S. (2025).* Will multimodal large language models ever achieve deep understanding of the world? Frontiers in Systems Neuroscience, 19, 1683133. https://doi.org/10.3389/fnsys.2025.1683133 Lin, Z. (2025).* Six fallacies in substituting large language models for human participants. Advances in Methods and Practices in Psychological Science, 8(3), 25152459251357566. https://doi.org/10.1177/25152459251357566 Seth, A. K. (2025).* Conscious artificial intelligence and biological naturalism. Behavioral and Brain Sciences, 1–42. https://doi.org/10.1017/S0140525X25000032 Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2024).* Dissociating language and thought in large language models. Trends in Cognitive Sciences, 28(6), 517–540. https://doi.org/10.1016/j.tics.2024.01.011 Mitchell, M., & Krakauer, D. C. (2023).* The debate over understanding in AI's large language models. Proceedings of the National Academy of Sciences, 120(13), e2215907120. https://doi.org/10.1073/pnas.2215907120 Bowers, J. S. (2025).* The successes and failures of artificial neural networks (ANNs) highlight the importance of innate linguistic priors for human language acquisition. Psychological Review. Advance online publication. https://doi.org/10.1037/rev0000595 Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2024).* Dissociating language and thought in large language models. Trends in Cognitive Sciences, 28(6), 517–540. https://doi.org/10.1016/j.tics.2024.01.011 Bolhuis, J. J., Crain, S., Fong, S., & Moro, A. (2024).* Three reasons why AI doesn't model human language. Nature, 627(8004), 489. https://doi.org/10.1038/d41586-024-00824-z Bender, E. M., & Koller, A. (2020).* Climbing towards NLU: On meaning, form, and understanding in the age of data. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 5185–5198. https://doi.org/10.18653/v1/2020.acl-main.463 Everaert, M. B. H., Huybregts, M. A. C., Chomsky, N., Berwick, R. C., & Bolhuis, J. J. (2015).* Structures, not strings: Linguistics as part of the cognitive sciences. Trends in Cognitive Sciences, 19(12), 729–743. https://doi.org/10.1016/j.tics.2015.09.008 Hauser, M. D., Chomsky, N., & Fitch, W. T. (2002).* The faculty of language: What is it, who has it, and how did it evolve? Science, 298(5598), 1569–1579. https://doi.org/10.1126/science.298.5598.1569 Johnson, M., & Lakoff, G. (2002). Why cognitive linguistics requires embodied realism. Cognitive Linguistics, 13(3), 245–263. https://doi.org/10.1515/cogl.2002.016 Lakoff, G. (2012). Explaining embodied cognition results. Topics in Cognitive Science, 4(4), 773–785. https://doi.org/10.1111/j.1756-8765.2012.01222.x Harnad, S. (1990). The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3), 335–346. https://doi.org/10.1016/0167-2789(90)90087-6 Placani, A. (2024). Anthropomorphism in AI: Hype and fallacy. AI and Ethics, 4, 691–698. https://doi.org/10.1007/s43681-024-00419-4 Salles, A., Evers, K., & Farisco, M. (2020). Anthropomorphism in AI. AJOB Neuroscience, 11(2), 88–95. https://doi.org/10.1080/21507740.2020.1740350 Floridi, L. (2025). AI as agency without intelligence: On artificial intelligence as a new form of artificial agency and the multiple realisability of agency thesis. Philosophy & Technology, 38(1), 1–27. https://doi.org/10.1007/s13347-025-00858-9 <<< I am not convinced by his position, but it is nonetheless relevant since it splits the debate in two: intention and consciousness Dreyfus, H. L. (2007). Why Heideggerian AI failed and how fixing it would require making it more Heideggerian. Philosophical Psychology, 20(2), 247–268. https://doi.org/10.1080/09515080701239510 Bengio, Y., & Elmoznino, E. (2025). Illusions of AI consciousness. Science, 389(6765), 1090–1091. https://doi.org/10.1126/science.adn4935 Dotov, D. G., Nie, L., & Chemero, A. (2010). A demonstration of the transition from ready-to-hand to unready-to-hand. PLOS ONE, 5(3), e9433. https://doi.org/10.1371/journal.pone.0009433 Mitchell, M. (2021). Why AI is harder than we think. Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2021). https://doi.org/10.1145/3449639.3465421 I haven't read all of them yet. Feel free to discuss. Now, there are two problems I see in the community regarding the critique of AI. One is the problem of the increasing capability of models. The other is the idea that GOFAI and ANN-based systems (like LLMs) are fundamentally different. Let me explain. 1) The increasing capability of models: it is difficult to engage in any meaningful discussion if the metrics are the capability of models. One should look at how models structurally encode information and what the learning process looks like from an epistemic point of view. As far as I know, and correct me if I'm wrong, these two issues have not changed and are likely not going to change. 2) The idea that GOFAI and ANN-based systems are fundamentally different: this, I recognize, is a controversial claim. But one should not look at how GOFAI and ANN-based systems encode knowledge (explicitly curated and written rules vs statistical learning), but at how the learning material is selected, curated and presented to the system, and the problem of 'closure' and self-reference in datasets. In this regard (which we could call epistemic) there should be no difference between these two technologies. Again, we should not look at how they are implemented, but at how we relate to them from an epistemic point of view. But going back to my initial comment, this whole thread feels like proving my point. For those not wanting to get involved in philosophy despite willing to engage in AI research discussions, keep in mind that philosophy has always been a guiding light for science. As a final note: the whole discussion about AI is on whether computational theories of mind are actually solid or not. But it is really difficult to engage in this conversation without at least some background in philosophy, but preferably a strong background in it. I'm getting a bit tired of coming back to this thread. Reach me out if you want to discuss more. Glad to help and glad to learn. @federico_ricca https://www.linkedin.com/in/federico-ricca

core ideas which are the justification for the 2021 paper that actually introduced the rhetoric of 'stochastic parrot') made many clear, falsifiable statements about what LLMs would never be able to do; here's one of them: https://aclanthology.org/2020.acl-main.463.pdf#page=14 > To get a sense of how existing LMs might do at such a task, we let GPT-2 complete the simple arithmetic problem 'Three plus five equals'. The five responses below, created in the same way as above, show that this problem is beyond the current capability of GPT-2, and, we would argue, any pure LM. One could say many things about this claim, but not that it is a 'strictly philosophical one'.

eventual full-scale solution to human-analogous NLU, they are not nearly-there solutions to this grand challenge. We argue in this paper that genuine progress in our field — climbing the right hill, not just the hill on whose slope we currently sit —depends on maintaining clarity around big picture notions such as meaning and understanding in task design and reporting of experimental results.

that have never existed before. Meanwhile, it's failing at basic tasks: you can find plenty of examples of it failing basic logic, anything with math or arithmetic, a lot of ethics/bias concerns stemming from the training data, etc. I think when we look forward to an AI that "reasons" this is not what anybody would mean. Current AIs are bullshit engines. They are very impressive, and probably even useful. They are a milestone. But they are not reasoning in any meaningful way. And if you look at the math behind them there's really no reason to think they would. So I guess given a methodology that seemingly shouldn't produce reasoning ability, and no evidence that it has so far, sure, maybe scale will magically unlock it. There's always a chance I guess. But it doesn't really seem too sensible. See Sam Altman's take as well on Twitter here [3] [1] https://en.wikipedia.org/wiki/Chinese_room [2] https://aclanthology.org/2020.acl-main.463.pdf [3] https://twitter.com/sama/status/1601731295792414720