- collective predictive coding ＝ symbolic emergence (springing up from a seedbed, esp. as a result of an explosion of symbols)
- Collective / prediction / encoding
- Tadahiro Taniguchi
tanichu NEW Single Author, Japanese, Outlook Paper “Cognitive Science,” “Dynamics of Language and Cognition Based on Collective Predictive Encoding: Toward New Developments in Symbolic Emergence Robotics,” https://doi.org/10.11225/cs.2023.064…
It is like Taniguchi’s current understanding of the world and his vision of it, written to the best of his ability alone.
① Isn’t language emergence a collective predictive encoding? ② If “Language” is taken as the subject, humans come to a subordinate position (language-subordinated worldview). If we consider “language” as a subject, language emergence is representation learning of external representation. Therefore, language emergence can be expressed as a generative model. ⑤ Language Games is a form of decentralized Bayesian inference algorithm. ⑥ Therefore, it can be straightforwardly developed as AI and robotics research, including to the social level. (Socialization of the world model) ⑥ Linguistic emergence may therefore be seen as based on the free energy principle “in society”. ⑦ Large-scale language models internalize the structure of the results of the above-mentioned learning by group representation, so they internalize the structural information of the world. Therefore, it “knows the world” unexpectedly in a way that integrates the human cyclic world. I think that a (predictive) encoding such as next token prediction is sufficient for this learning.
I am writing this based on this kind of world perception, although there are some meanings. Please have a look.
Also, in English, it is preprint, but Taniguchi, T. (2023, August 15). Collective Predictive Coding Hypothesis: Symbol Emergence as Decentralized Bayesian Inference. https://doi.org/10.31234/osf.io/d2ty6… I have written about collective predictive coding in > (UNDER REVIEW).
By the way, Mr. Ueda of Tokyo University @CsWeddy Lewis’s Signaling Game as beta-VAE,” which was accepted by ICLR2024, is a grounded formulation of language (symbol) emergence as such expressive learning and a discussion of the implications it brings. https://x.com/CsWeddy/status/1762044721160155267?s=20…
We are also publishing a bird’s eye view survey of language emergence at the same time, thank you. https://x.com/tanichu/status/1768427404261982241?s=20…
tanichu It is still a “vision” and needs to be supported and applied by many studies, but since it is a vision paper, I would be happy to give inspiration to many people and expand and deepen the discussion further. I would be happy if I could give inspirations to many people to further expand and deepen the discussion. …
:
This page is auto-translated from /nishio/集合的予測符号化＝記号創発 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.