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

“What do we mean by “Human” in HGAI? There are various stakeholders involved in the ecosystem of Generative AI, including

(1) people whose data is used for model training (e.g., artists’ and designers’ work),

(2) people who label and moderate the data (e.g., to filter out toxic contents [5]),

(3) people who develop Generative AI models (e.g., academic professors/students and employees in tech companies who build LLMs),

(4) people who develop systems that use Generative AI models (e.g., game development platforms that use LLMs to enable conversational characters),

(5) end-users of Generative AI and its applications,

(6) and finally, people who are impacted (in)directly by various (un)intended consequences of Generative AI (e.g., teachers grading AI-generated essays).

Our discussion of HGAI primarily focuses on the betterment of individuals who own the data and the end-users of Generative AI, although some of our next-step ideas and calls-for-actions do speak to the other stakeholders as well.”