| Project/Area Number |
22K00015
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 01010:Philosophy and ethics-related
|
|---|
| Research Institution | Akita International University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2022-04-01 – 2025-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥3,250,000 (Direct Cost: ¥2,500,000、Indirect Cost: ¥750,000)
Fiscal Year 2024: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2023: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2022: ¥1,950,000 (Direct Cost: ¥1,500,000、Indirect Cost: ¥450,000)
|
|---|
| Keywords | artificial intelligence / scientific explanation / machine learning / philosophy of science / game of Go / data mining / scientific method / deep learning / epistemology |
|---|
| Outline of Research at the Start |
Due to advances in artificial intelligence, especially in deep learning, we now have access to automatically generated high-quality statistical knowledge beyond human expert intuition in many fields. However, the representation is not human-friendly: an opaque mass of pure associations instead of narrative, causal explanations. It gives only answers, but no explanations. We investigate how we can use these AI systems for improving human understanding. How can we learn from AIs? We combine empirical, computational, and theoretical research and focus on the game of Go as our testbed application.
|
| Outline of Annual Research Achievements |
We continued to work on the analysis of cost of passing measure, and published an extended version of the paper. It includes the discussion of the measure of efficiency, fingerprinting of a game record database, and a more precise numerical characterization of game stages.
We did engineering work on the analysis software: changing the visualization library and separating the analysis tools from the Go engine and game management modules.
Further (originally unplanned) research was done on the algebraic automata theory analysis of games. This makes the definition of ground truth in game worlds precise and thus determines the available room for knowledge growth.
Started work on a new theory of explanation based on the idea of compatible operations (algebraic homomorphisms).
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
Due to the unforeseen problems with the game record database (the rule sets and komi settings are not consistent) the historical analysis is slower than expected.
The best practices survey has been rescheduled due to the extra work on the algebraic game theory.
Regarding the spending, I have received generous funding from the University of Waterloo, thus there was no need for using the budget for my travel.
I had an accepted talk scheduled for the 1st International Go Studies Conference, but my presentation was cancelled last minute; the real reasons never disclosed. This did not affect the progress of the project directly, but psychologically it was damaging.
However, despite the minor setbacks and reorganizations, there is no reason to think that the project will not finish on time.
|
| Strategy for Future Research Activity |
Applying the theory of explanation based on morphic relations (as in algebra) to the game of Go. This will address the question, what is a good explanation and how can we create them from the non-explanatory but high-precision AI output. This is the focus of this project: verbalizing AI knowledge for human understanding. We will develop the algebraic/category theoretical ideas in concert with a scholarly study of the most recent literature on scientific explanation.
Finishing the historical game analysis with semi-automated (partially manual) detection and correction of rule sets for game records. We plan to create summary visualizations for a large number of games.
Finishing the best practices survey and detecting any deviations between current practice and our scientific recommendation.
|
