2023 Fiscal Year Annual Research Report
Human-Centric Autonomy from Agentic Dynamics
| Project/Area Number |
23H03455
|
|---|
| Allocation Type | Single-year Grants |
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
Rafik Hadfi 京都大学, 情報学研究科, 特定准教授 (30867495)
|
|---|
| Project Period (FY) |
2023-04-01 – 2026-03-31
|
| Keywords | Autonomy / Agency / Artificial Intelligence / Trustworthy AI / Game Theory |
|---|
| Outline of Annual Research Achievements |
The current achievement involves surveying the two main concepts of the project (agency and autonomy) and developing algorithms that use information-theoretic metrics, precisely mutual information, to define autonomy and agency. The resulting algorithms were tested on automata that simulate systems of interacting players (in the game theoretic sense). The findings from these tests are now being prepared for submission to a journal. The second achievement is the presentation of the theoretical model of the research and its implications for the study of agency and autonomy from an enactive perspective. This theoretical model results from the surveys of the concepts in cognitive sciences, particularly in 4E cognition. The last finding was presented at the CHAIN Philosophy Workshop "Frontiers of Enactivism in East Asia" at Hokkaido University.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The initial phase of the project is primarily theoretical. It involves surveying various accounts of agency and autonomy and developing mathematical and algorithmic formulations of these concepts.
|
| Strategy for Future Research Activity |
The plan consists in three main phases, starting with the theoretical characterization of autonomy. The first phase involves formulating autonomy using information-theoretic metrics, specifically conditional mutual information, and translating these metrics into algorithms capable of quantifying autonomy in systems composed of two agents. These algorithms will then be tested on automata that simulate real systems of interacting game-theoretic players. This initial implementation will be conducted with the assistance of a research assistant (RA) and primarily using a laptop. The ongoing second phase focuses on constructing a game-theoretic framework for autonomy that incorporates the previously developed metrics into various games, such as Iterated Prisoner's Dilemma (PD). This phase involves algorithmic implementations, which will be coded in Python with the support of an RA.
|
