AI for Human Agency
Why and how we are investing in LLMs for the social sector
The current debate over AI is often characterized as a battle between “doomers” and “boomers” – pitting those who see AI as an existential threat against those who believe it will usher in an era of rapid human progress. Meanwhile, Large Language Models (LLMs) like OpenAI’s ChatGPT are already transforming life and business as we know it, while raising concerns over their potential to manipulate, mislead, or be used to nefarious ends.
Central to the boomers’ optimism is whether and to what extent AI can be harnessed for social good. One dimension of this issue is often overlooked, however: how can AI amplify and advance the organizations already helping communities thrive – the countless nonprofits, social enterprises, charities, and other social ventures around the world working in areas like education, healthcare, mental health, and jobs? This sector is particularly critical right now, as countries’ systems of social care buckle under demographic and other pressures.
At the Agency Fund, we are keenly interested in this question. Our mission is to enhance human agency: to ensure that people have the resources they need – including psycho-social support like coaching, mentorship, and counseling – to navigate toward a better future. Technology and data are already at the heart of our approach. But we think AI may have a major role to play, too.
This is why we’re starting to experiment – carefully – with the use of LLMs and other forms of generative AI. Here’s our hypothesis: If used safely and effectively, these technologies could dramatically increase the scale and reach of core social services. The need for these services, like mental health or counseling, completely outstrips the sector’s delivery capacity. But with LLMs, providing customized and interactive support to very large audiences is now both technically feasible and affordable. LLMs won’t just expand and deepen the social sector’s impact – we believe they can exponentially increase it.
To test this hypothesis, we’ve put out a call for proposals, with an initial funding round that allocates $2-4 million to projects that explore and evaluate these possibilities. Applications are open through December 22.
Here is how we are thinking about the case for AI in the social sector.
Why we are investing in AI
Many of the world’s deepest problems defy one-size-fits-all solutions, but many social sector organizations still pursue “magic bullets.” At the Agency Fund, we believe in the power of highly customized, context-aware solutions driven by our grantees and their beneficiaries. Our goal is not to deliver top-down answers. Rather, we work with our grantees to help people build their own muscle for agency, feel what it means to exercise it, and get stronger at expressing it. (See how our funder-doer model works.)
LLMs align with this approach by helping us better navigate impact-rich contexts that defy one-size-fits-all thinking. We already draw inspiration from the tech sector’s commitment to continuous learning (by combining research, action, learning, and adaptation) and building for scale (by finding what is working and expanding it quickly, wherever it is effective). Through funding and hands-on engagement, we seek to prepare our grantees to apply these principles for social good at scale.
Generative AI’s evolving abilities align with this mission in three ways:
1. Democratizing access to contextualized information
Our partners are deploying LLMs to provide vetted guidance to low-income households in domains like parenting, education, career development, and beyond. Rather than overloading people with dense information on websites, manuals, and informational leaflets, LLMs can engage users in conversation to help them make sense of complex processes.
Many of our partners are building these capabilities right now. CareerVillage is integrating AI into its career counseling platform. The Apprentice Project is exploring AI-based career advice and creative exploration tools for school children. In Kenya, UC Berkeley and the Busara Center are evaluating LLM-based decision support tools to deliver business management and marketing advice for rural micro-entrepreneurs. Kabakoo has launched Mentor AI, a chatbot supporting learners who are seeking to upskill themselves into digital careers.
2. Scaling care solutions
Overworked nurses, teachers, and other front-line workers are often unable to provide customized care and feedback for their clients. But AI support can empower these professionals. For example, an LLM can generate text responses to patient WhatsApp queries, with oversight from a nurse or therapist. Eventually, these services could become so good that they can be run independently.
This is the thesis of several of our grantees. Myna Mahila Foundation, for example, is experimenting with an AI bot to assist young women in India with sexual and reproductive health queries – while Rocket Learning is developing an AI coach for Indian parents and caregivers to augment their early childhood education programs. Family Empowerment Media is enhancing its nurse-staffed helpline with AI capabilities, and SameSame focuses on customizing mental health support for LGBT+ youth in challenging environments.
3. Deepening operational efficiency
Nonprofits and social sector agencies rarely have the funds to resource an efficient back office. LLMs provide a practical way to automate routine tasks, from scheduling to administrative data cleaning and analysis. This frees up valuable staff time (and budgets) to focus on delivering services to those in need. LLM-augmented operations are also a relatively low-risk way for the sector to gain experience with AI tooling.
Such applications can benefit a wide range of social sector organizations. We helped our East African grantee Shujaaz, for instance, use ChatGPT to run batch language analysis (i.e., language detection, translation, and sentiment analysis) on hundreds of thousands of social media and SMS messages – analyzing language data in Swahili, English, and the hybrid language Sheng to assess the Shujaaz network’s programs to inspire, entertain, and mobilize more than 7.5 million 15-24-year olds in Kenya and Tanzania.
How we are investing in AI
Across all our experiments with generative AI, we’re investing in learning, benchmarking, and rigorous evaluation to understand the impact of these technologies. How will access to LLMs change the outcomes for our grantees’ end users – in terms of their livelihoods, health, or education? Will generative AI really help our grantees serve the neediest, or will it exacerbate inequalities?
Bringing AI into the social sector must also be balanced against the risks of these technologies. As many of us have experienced, LLMs can “hallucinate,” or produce highly plausible but completely incorrect answers to factual questions. Unless or until these problems can be resolved, LLMs will not be suitable for applications requiring high accuracy, specialized domain knowledge, or strong reasoning skills (e.g., medical diagnosis, legal advice, or national security). And when utilizing LLMs in any sector, it’s critically important for humans to be involved – a principle the Agency Fund and its grantees are taking to heart as we explore these new horizons.
The history of technology is rife with innovations that sparked intense debate but ultimately led to unprecedented gains in productivity and human development. While today’s boomers and doomers are unlikely to agree anytime soon, recent leaps in AI have undeniably unlocked new and far-reaching capabilities. As we begin experimenting with LLMs, we’re cautiously optimistic about AI’s transformative power for the social sector – by making social ventures more effective, supercharging their ability to reach people in need, and advancing technologies and platforms that may actually promote human agency, rather than stifle it.