Thank you to all our participants, co-organizers, student volunteers, funders, and partners who made this possible. And to Joy Ming for the beautiful visual summaries.

Our conversations spanned:
🔷 Meaningful use cases of AI in high-stakes global settings
🔷 Interdisciplinary methods across computing and humanities
🔷 Partnerships between academia, industry, and civil society
🔷 The value of local knowledge, lived experiences, and participatory design

Over three days, we explored what it means to design and govern pluralistic and humanistic AI technologies — ones that serve diverse communities, respect cultural contexts, and center social well-being. The summit was part of the Global AI Initiative at Cornell.

This was the week of reflection, new ideas, and a renewed sense of urgency to design AI systems that serve marginalized communities globally. Can't wait for what's next.

Pragnya Ramjee presented work (with Mohit Jain at MSR India) on deploying LLM tools for community health workers in India. In collaboration with Khushi Baby, we show how thoughtful AI design can (and cannot) bridge critical informational gaps in low-resource settings.
dl.acm.org/doi/10.1145/...

Ian René Solano-Kamaiko presented our study on how algorithmic tools are already shaping home care work—often invisibly. These systems threaten workers’ autonomy and safety, underscoring the need for stronger protections and democratic AI governance.
dl.acm.org/doi/10.1145/...

Joy Ming presented our award-winning paper on designing advocacy tools for home care workers. In this work, we unpack tensions between individual and collective goals and highlight how to use data responsibly in frontline labor organizing.
dl.acm.org/doi/10.1145/...

Dhruv presented our cross-cultural study on AI writing tools and their Western-centric biases. We found that AI suggestions disproportionately benefit American users and subtly nudge Indian users toward Western writing norms—raising concerns about cultural homogenization.
dl.acm.org/doi/10.1145/...

Sharon Heung presented our work on personalizing moderation tools to help disabled users manage ableist content online. We showed how users want control over filtering and framing—while also expressing deep skepticism toward AI-based moderation.
dl.acm.org/doi/10.1145/...

www.fastcompany.com/91324551/cha...

Kudos to Dhruv Agarwal for leading this work and such fun collaboration with @informor.bsky.social!

As these tools become more common, it’s critical to ask: Whose voice is being amplified—and whose is being erased? www.theatlantic.com/technology/a...

Huge congratulations to Mahika Phutane for leading this work, and Ananya Seelam for her contributions!

We’re thrilled to share this at ACM FAccT 2025.

Read the full paper: lnkd.in/eCsAupvK

Our findings make a clear case: AI moderation systems must center disabled people’s expertise, especially when defining harm and safety.

This isn’t just a technical problem—it’s about power, voice, and representation.

Disabled participants frequently described these AI explanations as “condescending” or “dehumanizing.”

The models reflect a clinical, outsider gaze—rather than lived experience or structural understanding.

AI systems often underestimate ableism—even in clear-cut cases of discrimination or microaggressions.

And when they do explain their decisions? The explanations are vague, euphemistic, or moralizing.

We studied how AI systems detect and explain ableist content—and how that compares to judgments from 130 disabled participants.

We also analyzed explanations from 7 major LLMs and toxicity classifiers. The gaps are stark.