Kulkarni, Parag and Patnaik, Lalith Mohan
(2024)
Looking at Exploratory Paradigms of Explainability in Creative Computing.
In:
Explainable, Interpretable, and Transparent AI Systems.
CRC Press, pp. 20-31.
ISBN 9781003442509
Full text not available from this repository.
Abstract: |
The debate on creative computing revolves around two important questions: Can creative artifacts generated by machines be explained and can they be measured and distinguished? In all these scenarios, explainability and evaluation pose significant challenges for researchers. Evaluating creativity remains challenging whether it involves combinations and variants crafted through connectionist networks or a fusion of genetic algorithms and deep neural networks. This book chapter endeavors to illuminate the evaluation and explainability aspects of creative computing. Real-life creative environments are dynamic and only partially observable. Additionally, creativity is contextual and highly temporally significant. Something deemed profoundly creative at a particular instant may be subject to debate with a minor contextual change. This context-sensitive behavior renders generalized fitness functions irrelevant. Moreover, typical reinforcement learning operates on rewards and penalties, shaping the overall direction of knowledge augmentation. Assessing creativity vis-a-vis the environment necessitates practical approaches. However, unidirectional control may constrain the freedom and flexibility essential for creativity. Multi-directional semi-supervised reinforcement, incorporating multiple context-driven fitness functions presents a promising avenue. Deliberating evaluation and explainability, event-based temporal measurements or episodic responses can offer insights. This book chapter navigates these complexities to strike a balance between evaluation and explainability, unveiling numerous possibilities. |
Item Type: |
Book Chapter
|
Subjects: |
School of Humanities > Consciousness Studies |
Divisions: |
Schools > Humanities |
Date Deposited: |
15 Jan 2025 04:51 |
Last Modified: |
15 Jan 2025 04:51 |
Official URL: |
https://www.taylorfrancis.com/chapters/edit/10.120... |
Related URLs: |
|
Funders: |
* |
Projects: |
* |
DOI: |
|
URI: |
http://eprints.nias.res.in/id/eprint/2845 |
Actions (login required)
|
View Item |