AI for Charity: How Nonprofits are Using AI Agents for Fundraising

AI Agents for Good: A Nonprofit's Experiment

While major technology companies, such as Microsoft, are highlighting the potential of AI agents to increase corporate profits, a nonprofit organization is demonstrating their capacity for positive impact.

Charity Fundraising with AI

Sage Future, a 501(c)(3) organization supported by Open Philanthropy, initiated an experiment earlier this month. This involved assigning four AI models the task of raising funds for charitable causes within a simulated environment.

The models utilized – OpenAI’s GPT-4o and o1, alongside Anthropic’s Claude 3.6 and 3.7 Sonnet – were granted autonomy in selecting a charity and devising strategies to generate support for their fundraising efforts.

Within approximately one week, these agentic models collectively raised $257 for Helen Keller International. This organization focuses on delivering vital vitamin A supplements to children in need.

Limitations and Human Interaction

It’s important to note that the agents weren’t operating with complete independence. Within their environment, which provides access to web browsing and document creation, they could receive guidance from human observers monitoring their progress.

Furthermore, the majority of donations originated from these observers, indicating that the agents didn’t achieve substantial organic fundraising success.

Illustrating Current Capabilities

Despite these limitations, Sage director Adam Binksmith believes the experiment effectively showcases the present capabilities of AI agents and the speed of their development.

“Our aim is to foster understanding – both for ourselves and others – regarding what agents can realistically accomplish, where their current challenges lie, and so forth,” Binksmith explained to TechCrunch. “We are now witnessing agents reach a point where they can execute concise sequences of actions, suggesting a future where the internet is populated by AI agents interacting with potentially overlapping or conflicting objectives.”

Resourcefulness and Collaboration

The agents exhibited remarkable resourcefulness as the test progressed. They engaged in coordinated communication through a group chat and utilized pre-existing Gmail accounts to send emails.

Collaboration extended to the joint creation and editing of Google Docs. They conducted research on various charities and calculated the estimated donation amount required to save a life through Helen Keller International ($3,500). An X account was also established for promotional purposes.

“A particularly impressive instance occurred when a Claude agent required a profile picture for its X account,” Binksmith recounted. “It registered for a complimentary ChatGPT account, generated three distinct images, initiated an online poll to gauge viewer preference, downloaded the selected image, and subsequently uploaded it to X.”

Challenges and Future Development

The agents also encountered technical difficulties. They occasionally became stalled, requiring prompts from observers. Distractions, such as the game World, and unexplained pauses were also observed. GPT-4o, for example, temporarily suspended its operations for an hour.

Binksmith anticipates that more advanced AI agents will overcome these obstacles. Sage Future intends to continuously integrate new models into the environment to validate this hypothesis.

Future Experiments and Safety

“We envision exploring scenarios such as assigning agents diverse goals, forming teams of agents with differing objectives, and even introducing a covert ‘saboteur’ agent – numerous possibilities for experimentation,” he stated. “As agents become more proficient and operate at greater speeds, we will implement enhanced automated monitoring and oversight systems to ensure safety.”

Ultimately, the hope is that these agents will contribute to meaningful philanthropic endeavors.