Next Steps for Human-Centered Generative AI: A Technical Perspective

Through iterative, cross-disciplinary discussions, we define and propose next-steps for Human-centered Generative AI (HGAI). We contribute a comprehensive research agenda that lays out future directions of Generative AI spanning three levels: aligning with human values; assimilating human intents; and augmenting human abilities. By identifying these next-steps, we intend to draw interdisciplinary research teams to pursue a coherent set of emergent ideas in HGAI, focusing on their interested topics while maintaining a coherent big picture of the future work landscape.

Introduction. The recent development of Generative AI—ranging from large language models [26, 148] to visual generation techniques [100, 115, 157]—promises to revolutionize how humans work in a wide range of tasks [109]. Meanwhile, various research communities, will soon, if not already, be working on topics related to Generative AI. Historically, when new technological breakthroughs emerged, there was a tendency for researchers in adjacent fields to pursue “low-hanging fruits” and rapidly produce results. While such an approach does accelerate our knowledge of the new technology in the short-term, it somewhat limits researchers’ vision from seeing a holistic picture and how certain problems can and should be tackled with cross-disciplinary efforts. To establish a unified framework that ties various emergent research across disciplines, this paper proposes Human-centered Generative AI (HGAI, pronounced ‘H’-/gaI/) as an overarching topic and lays out specific next steps for achieving HGAI.

Discussion / Conclusion. Currently, Generative AI is one of the fastest growing fields; yet, we argue that focusing too much or pursuing some immediate research ideas might lose sight of a holistic picture that can connect multiple disciplines towards some long-term, shared missions. Unifying a wide range of ongoing topics as well as less-explored ideas, this paper contributes a research agenda that lays out the landscape of next-steps for HGAI. Specifically, One current limitation is a lack of direct involvement with diverse stakeholders who are impacted by Generative AI (e.g., designers and artists). Mitigating harm caused by Generative AI itself is another grand challenge, as indicated in this comprehensive report [9] by epic.org. We plan to address specific stakeholders in the future as we pursue specific topics identified in the proposed agenda. We hope these next-steps can serve as starting points for researchers across disciplines to collaborate and pursue specific ideas while staying informed of the big picture. As Generative AI continues to develop at unprecedented speed and scale, we believe that taking a human-centered approach early on can have a significant long-term impact on the future of human-AI symbiosis [89].