“Come to Us First”: Centering Community Organizations in Artificial Intelligence for Social Good Partnerships

Hongjin Lin
ACM CSCW
Published in
5 min readOct 24, 2024
People join hands for a group cheer.
People join hands for a group cheer. Photo credit: Dio Hasbi Saniskoro on Prexel.com

Artificial Intelligence for Social Good (AI4SG) explores the potential of AI technologies for tackling complex social issues, often through interdisciplinary partnerships between teams that build AI technologies and community organizations that are domain experts on social issues. However, the power imbalances in these partnerships could compromise the very social impact AI4SG aims to achieve.

In a recent study, we foreground the crucial, yet often overlooked perspectives and experiences of community organizations in these partnerships. Through 16 interviews with staff from community organizations, this research highlights 1) the disconnection between AI innovation and community needs, 2) the pervasive influence of funding agendas on AI4SG projects, 3) community organizations’ contribution and labor, and 4) their aspirations for future partnerships. Figure 1 provides a brief summary of the relationships among AI4SG stakeholders revealed in the study. We urge all AI4SG stakeholders to center and adequately compensate community organizations for their expertise, assets, and leadership to support effective, ethical, and sustainable AI development.

Links and relationships among stakeholders in AI4SG partnerships. Stakeholders include community organizations, end beneficiaries, technology teams, and funders.
Figure 1: Relationships among stakeholders in AI4SG partnerships. Icon made by GOWI from www.flaticon.com.

The Disconnect Between AI Innovation and Community Needs

For many community organizations engaged in AI4SG collaborations, the hope was to deploy concrete solutions that could reduce manual workload or improve their social services. However, projects often resulted in little more than academic papers; only 2 out of 14 projects we studied reached the deployment stage. For example, a nature conservation expert partnered with an industry team to develop grassland quality estimation software using remote sensing data, but the project was never deployed as technical support ended when funding ran out after a year. Reflecting on the project, they shared that while the partnership was “great for bouncing ideas back and forth,” it was “insufficient for actually building anything of real substance.”

Besides funding constraints, incentive misalignment between community organizations and their partners also contributed to community organizations’ goals not being prioritized. For example, a domain expert in international development noted that “[research labs] don’t deploy the solutions; they finish the paper, and that’s it, because their goal is to research, so when research is done, most probably there is no continuity.” Similarly, frustration arose with private industry partners, where profit motives clashed with the public interest goals of community organizations. As a data science manager in an NGO explained: “it’s very, very difficult to work with private sector partners, because they have the interest of selling the software to us, and we have the interest of tweaking whatever they have for serving our needs.

The Influence of Funding and Power Asymmetries

Funding agendas have a pervasive influence on the scope, timelines, and priorities of AI4SG projects. As a project manager in an international development organization noted: “There’s this perverse incentive of responding to the donor […] we end up building this platform for the funder with all the things the funder wants to see in it.” Importantly, the participant then stressed that the pressure to meet funders’ visions could result in a solution that “never works in the field because no one else wants it in that way.”

Additionally, short-term funding cycles for AI4SG projects often force organizations to focus on quick, achievable outcomes instead of long-term, sustainable solutions. In a three-month summer program, an international organization had to restrict its project to ideas that fit within the timeframe. While the project resulted in a published research paper, it was never implemented because there was no further funding support for deployment.

Community Contributions and Intellectual Labor

Community organizations are essential to the success of AI4SG projects, offering not only domain expertise, but also data and access to local communities that make these projects viable. In addition to data collection, annotation, and management work, our participants emphasized the importance of data privacy, especially when handling sensitive information involving vulnerable populations. This often involved unseen labor, such as meeting reporting requirements. For instance, in a project on estimating child malnutrition, a project manager from a non-profit provided “data [that] not everyone can have,” explaining, “[t]o collect data, you need ethical approval […] we’ve gone through the pain.”

Beyond data work, community organizations also play a critical role in building trust and translating technological solutions into practical tools for their communities. As a project manager explained, “There’s a lot of apprehension among people […] we had to talk to people, get their trust in the program. […] We thought if we just roll [the AI solution] out, people might respond and all, but then that’s not the case.” This illustrates the community engagement efforts required to ensure the practical implementation of AI4SG projects.

The Path Forward: Data Co-liberation and Community-Centered Approaches

To rectify the disconnect between AI projects and the needs of community organizations, we advocate for a shift from a tech-centric approach to a community-centered model for AI4SG. This would mean community organizations are engaged from the earliest stages of project ideation and given co-leadership throughout the process. As one participant expressed, “our dream is that before a research institute decides to do something, they come to us first and ask ‘what do you need?’ Rather than ‘oh I need to use this tool so I need to work with you’” (emphasis added).

Involving community organizations from the start gives AI4SG projects a better chance of meeting real community needs. In the two successful cases that led to deployment, participants had clear technology needs and actively collaborated on problem-solving.

This approach aligns with the concept of data co-liberation, proposed by D’Ignazio and Klein in Data Feminism. Data co-liberation embraces pluralism of knowledge and centers the leadership of community organizations and local communities. By sharing leadership (and power) with community partners, we can shift the focus from the tools to the problems at hand, as well as align goals across stakeholders from the beginning.

To facilitate concrete steps towards data co-liberation in AI4SG initiatives, those who hold the power of shaping funding agendas should invest in projects with meaningful and long-term engagement with communities. Extending funding timelines would also help address the complexity of aligning technological solutions with social challenges. Technology teams, meanwhile, could adopt established community-based research and design guidelines in the Human-Computer Interactions (HCI) literature. While structural constraints in academia and industry may pose challenges, exploring creative, non-traditional funding sources can help support more sustainable partnerships.

Community organizations’ co-leadership is essential not only for fostering more effective, sustainable, and ethical development of technology but also for building meaningful relationships among people.

Citation:

Hongjin Lin, Naveena Karusala, Chinasa T. Okolo, Catherine D’Ignazio, and Krzysztof Z. Gajos. 2024. “Come to us first”: Centering Community Organizations in Artificial Intelligence for Social Good Partnerships. Proc. ACM Hum.-Comput. Interact. 8, CSCW2, Article 470 (November 2024), 28 pages. https://doi.org/10.1145/3687009

--

--

ACM CSCW
ACM CSCW

Published in ACM CSCW

Research from the ACM conference on computer-supported cooperative work and social computing

Hongjin Lin
Hongjin Lin

Written by Hongjin Lin

0 Followers

Ph.D. Candidate in Computer Science at Harvard University. Personal website: https://sites.google.com/g.harvard.edu/hongjinlin

No responses yet