https://arxiv.org/abs/2409.00102
This paper proposes a new conceptual framework, CPC as Model of Science (CPC-MS), which models scientific activities in terms of Collective Predictive Coding (CPC). The main points are as follows:.
- The CPC concept is applied to scientific activities, modeling science as a distributed Bayesian inference process with multiple researchers.
- Describes the process by which the partial observations and internal representations of individual scientists are integrated into external scientific knowledge that is shared through communication and peer review.
- Mapping aspects of scientific activities such as experimentation, hypothesis formation, theory building, and paradigm shifting to the elements of a probabilistic graphical model.
- Provides new insights into issues such as social objectivity in science, scientific progress, and the impact of AI on science.
- He presents a perspective that rethinks science as a generative process rather than a confirmatory process.
- Provides a theoretical foundation for attempts to automate the entire scientific activity.
- Discussion of possible extensions to the model, including consideration of network structure and introduction of non-stationarity.
The framework aims to understand the creation and advancement of scientific knowledge as a collective cognitive process and to provide a new perspective on the integration of AI into science and the automation of scientific activities.
HiroTHamadaJP Collective Predictive Coding as a Model of Science, which I co-authored, has been published on arxiv! The co-authored paper is now available on arxiv! Taniguchi-san @tanichu Lead paper on applying collective predictive coding to scientific actions. - Tadahiro Taniguchi We are proposing to consider collective predictive coding as a theoretical framework for the state of science for a wide range of range from philosophy of science to science of science and DeSci! The article is available here: https://arxiv.org/abs/2409.00102 Please pick up a copy of Mr. Taniguchi’s recent book “Symbolic Emergent Systems” as well! Hamada has been involved in the collective search section and theoretical projection! Thanks to the following for leading further new papers…
Singular CPC (Singular CPC) is an extension of the regular CPC model, which deals with the case where a singularity exists in the parameter space of the model. This concept is introduced to better explain scientific progress and paradigm shifts (paradigm shift). The main features are as follows:
- Presence of singularity:
- In Singular CPC models, singularities exist in the parameter space. These singularities have important effects on the structure and behavior of the model.
- Discontinuous updates:
- In the regular CPC model, the update of the posterior distribution is gradual and continuous, whereas in Singular CPC, there can be discontinuous jumps between singular points.
- Similarity to phase transitions:
- These discontinuous jumps are analogous to phase transitions in physics and are suitable for modeling paradigm shifts in science.
- A new interpretation of scientific progress:
- Singular CPC allows us to view scientific progress not as a mere accumulation of knowledge, but as a fundamental change in model structure.
- Relation to Kuhn’s paradigm shift theory: - The concepts of “normal science” and “scientific revolution” proposed by Thomas Kuhn. can be expressed mathematically.
- Similarities to Deep Learning:
- This concept was introduced because singularities also exist in the parameter space of deep learning models and similar phenomena are observed during the learning process.
Singular CPC aims to better model the rapid theoretical changes and emergence of new paradigms seen in the history of science. This allows us to understand scientific progress not only as a continuous process, but also as a process that is sometimes accompanied by dramatic changes.
Deep learning models are suddenly smarter because they’ve created a paradigm shift and are more geocentric than celestial.世界の記述コストが安いことに気づくからなわけか
personal https://claude.ai/chat/a63f2f50-25f4-4db4-963e-01e4c48e52ab
This page is auto-translated from [/nishio/Collective Predictive Coding as Model of Science: Formalizing Scientific Activities Towards Generative Science](https://scrapbox.io/nishio/Collective Predictive Coding as Model of Science: Formalizing Scientific Activities Towards Generative Science) 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.