2013 Fiscal Year Annual Research Report
| Project/Area Number |
13J10032
|
|---|
| Research Institution | Nagoya University |
|---|
Principal Investigator |
アーノード リービフルヒャ 名古屋大学, 情報科学研究科, 特別研究員(DC2)
|
|---|
| Keywords | 心的表象 / 知能の進化 / 機械学習 / ニューラル・ネットワーク / 二次学習 / 進化と学習の相互作用 |
|---|
| Research Abstract |
Goal was to propose and refine a novel theory on the evolutionary origins of MR, and to attempt to verify this theory computationally, using techniques from Artificial Intelligence and Artificial Life to simulate evolution of cognition. The importance of the theory lies in the fact that if correct, it could resolve a long-standing issue in AI on the role of representation in cognition. Within cognitive science and the philosophy of mind, representation is generally seen as a central element of advanced cognition. However within AI, a substantial body of work has cast doubts on the assumption that representation as traditional ly envisioned is of much importance. This leads to somewhat of a paradox : if representation is unnecessary, then why did it evolve? Or is maybe our intuition that cognition is representational mistaken? The theory we propose aims to resolve this paradox by giving specific conditions for evolution of representational cognition. When these conditions obtain, evolut
… More
ion can be expected to favour representational solutions. When they don't, cognition remains non-representational. On this account our cognition is representational because in our evolutionary environment, the conditions were met. The conditions pertain to specific types of learning ability : learning with innate bias and second order learning. In environments that require these forms of learning, we can expect evolution of representational solution. This contributes to our understanding of the evolutionary origin of mind and intelligence, and furthermore can serve as a basis for the study of the evolution of mind (the problem that representational cognition failed to evolve in computational models has made many scholars doubt the applicability of evolutionary simulation work to the study of mind). I have worked towards dissemination and verification of this theory. After achieving promising results in computational work that applies the theory to a concrete case of spatial representation, I have this year mostly focused on its application to social cognition. Here I have shown that a model based on the theory successfully produced AI agents that can predict one another's behaviour after observation of past behaviour, in a way that requires representation, a finding that supports the theory and garnered interest at the conferences where I presented it. Less
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Progressing according to schedule. Expecting to complete the main research goals within schedule. Whether the optional additional goal of constructing a general model can be realized remains to be seen.
|
| Strategy for Future Research Activity |
1) Complete social representation work with full analysis of results.
2) Publish theoretical paper in philosophical journal (currently in review).
3) Further improve spatial representation model with insights gained from social representation work.
4) General model : If time allows, I will attempt construction of an evolvable neural network architecture that can be applied across many different representational task.
|
