Learning to Listen: Critically Considering the Role of AI in Human Storytelling and Character Creation

Anna Kasunic, Geoff KaufmanNAACL 2018

TL;DR

This paper argues for a reorientation of AI research in storytelling, moving beyond AI as a sole storyteller to AI as a collaborative partner that amplifies human voices, especially marginalized ones. Through interviews with character creators, the authors highlight pain points in authentic representation and propose AI-assisted approaches that focus on "listening" to human storytellers and supporting their creative processes.

What Problem Does This Paper Address?

The paper addresses the current trajectory of AI research in storytelling, which largely focuses on AI generating narratives and characters. While this is an intriguing area, the authors argue that it overlooks the inherent value and complexity of human storytelling. They highlight that existing narratives often suffer from issues of representation, bias, and authenticity, which AI, if trained on these corpora, could inadvertently amplify. Furthermore, the paper points out that the focus on "agentic" (interactive) narratives might be overshadowing the potential of "listening" (passive consumption) narratives and the crucial role of humans as storytellers themselves. The authors aim to spark a dialogue about how AI can better serve human storytelling endeavors, rather than simply automating them.

Methods

The paper is primarily an opinion piece and a qualitative exploration, rather than a traditional empirical study with novel algorithms. The authors:

  • Literature Review: They synthesize existing research on AI in narrative generation, character creation, and the psychology and sociology of storytelling.
  • Semi-structured Interviews: They conducted interviews with 14 individuals deeply involved in character creation (novelists, screenwriters, game designers, etc.) to understand their challenges and pain points.
  • Conceptual Framework: They propose a framework that distinguishes between "listening" and "agentic" narrative forms and advocate for a shift towards AI as a co-creator and supporter of human storytellers.

Datasets

Not discussed in the paper. The paper relies on existing literature and qualitative interview data.

Key Insights and Ideas

  • AI as a Listener, Not Just a Speaker: The core argument is that AI should be reoriented to "listen" to human storytellers, assisting them rather than replacing them. This involves amplifying marginalized voices and supporting the creative process.
  • The Value of "Listening" Narratives: While "agentic" (interactive) narratives are popular, the paper argues that "listening" narratives (like films and books) can be equally, if not more, engaging and transformative. AI could innovate in creating new forms of "listening" experiences.
  • Character Creation is Hard: Interviews revealed significant anxieties and challenges for creators in authentically representing characters different from themselves, particularly concerning issues of bias, stereotyping, and cultural appropriation.
  • AI Can Help Identify and Mitigate Bias: AI could be a valuable tool for flagging potential issues of inauthenticity, bias, and under-representation in narratives, acting as a critical assistant rather than a perpetuator of existing problems.
  • Collaboration is Key: The paper highlights the benefits of collaborative creation processes (e.g., in theater and game development) and suggests AI could simulate these collaborative dynamics for creators working in isolation.

Key Results

As an opinion piece, the paper does not present quantitative results. Its "results" are qualitative insights derived from interviews and literature synthesis:

  • Pain Points in Character Creation: Interviewees expressed discomfort and anxiety around representing diverse identities, fearing misrepresentation or cultural appropriation.
  • Need for Authentic Representation: There's a clear demand for AI to help identify and address issues of stereotyping, under-representation, and misrepresentation in storytelling.
  • Potential for AI as a Co-Creator: The interviews suggest that AI could support human creators by flagging potential issues, offering prompts for reflection, and even simulating collaborative feedback from characters.
  • "Listening" Narratives Offer Untapped Potential: The authors argue that AI research should explore innovative "listening" narrative forms that are not necessarily agentic but still engaging and transformative.

Limitations and Future Work

The authors acknowledge:

  • Wizard of Oz Approaches: Many of their proposed ideas, especially those involving sophisticated AI interaction or style transfer, would currently require "Wizard of Oz" (human-in-the-loop) approaches.
  • Not Exhaustive: The paper is an opinion piece and does not provide an exhaustive review of all AI storytelling research.
  • Future Work: The authors suggest several promising directions:
  • AI for Human Storytellers: Developing AI systems that assist humans in identifying biases, stereotypes, and authenticity concerns in their narratives.
  • Crowd-Powered Feedback: Leveraging crowdsourcing to improve AI narrative generation and provide feedback mechanisms for human creators, especially from underrepresented groups.
  • Simulating Collaboration: Using intelligent agents to simulate collaborative creative processes for isolated creators.
  • Innovating "Listening" Narratives: Exploring new forms of AI-powered "listening" narratives that can soothe, heal, and transport, potentially using multi-modal sensing and style transfer.
  • Focus on Empathy and Transformation: Designing AI systems that foster empathy, perspective-taking, and positive psychological or behavioral outcomes.

Summary updated

BibTeX
@inproceedings{kasunic2018learning,
  author = {Anna Kasunic and Geoff Kaufman},
  title = {{Learning to Listen: Critically Considering the Role of AI in Human Storytelling and Character Creation}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2018},
}
Figure