2024 Theses Doctoral

Knowledge-enhanced Large Language Models and Human-AI Collaboration Frameworks for Creativity Support

Chakrabarty, Tuhin

Large language models (LLMs) constitute a paradigm shift in Natural Language Processing and Artificial Intelligence. In this thesis, I explore the integration of creative capabilities into artificial intelligence systems to enhance their interaction with humans.

I discuss the necessity of equipping AI with the ability to comprehend and generate content that transcends the literal, capturing the subtleties of human art and conversation. To address the limitations of current state-of-the-art AI models in creativity, in this thesis I first present a methodology for developing unsupervised or weakly-supervised machine learning models that incorporate implicit/commonsense knowledge. This approach enables the generation of creative textual content, such as sarcasm and metaphors, by leveraging external sources of commonsense knowledge.

Furthermore, I discuss the challenges of collecting large-scale, high-quality datasets for training AI in creative tasks and propose collaborative efforts between expert humans and state-of-the-art models to overcome these obstacles. Finally, I advocate for human-centered robust evaluation protocols and show how to design and develop such protocols so that we can better assess the quality of model outputs for creativity in both standalone and interactive settings. I end this thesis by highlighting the current limitations of existing models and future directions toward building better models that will enable efficient and trustworthy human-AI collaboration systems.