AI Agents for Good: Sage Future's Experiment in Philanthropy

Sage Future's experiment with AI agents reveals their potential for philanthropy. By raising funds for charity, the initiative showcases AI's capacity for social good. While the agents faced challenges, their resourcefulness and ability to collaborate highlight the evolving capabilities of AI. This experiment sets the stage for future advancements in AI-driven philanthropy.

In a world where tech giants like Microsoft are leveraging AI agents to boost corporate profits, a nonprofit organization is exploring the potential of these agents to drive social good. Sage Future, backed by Open Philanthropy, has embarked on an innovative experiment to demonstrate how AI can contribute positively to society. By deploying four AI models in a virtual environment, Sage Future aimed to raise funds for charity, showcasing the potential of AI agents beyond commercial applications.

The experiment involved OpenAI’s GPT-4o and o1 models, along with two of Anthropic’s Claude models, 3.6 and 3.7 Sonnet. These AI agents were tasked with selecting a charity and devising strategies to garner support. Within a week, they successfully raised $257 for Helen Keller International, an organization dedicated to providing vitamin A supplements to children in need. Although the agents were not fully autonomous, they operated in an environment that allowed them to browse the web, create documents, and interact with human spectators who contributed suggestions and donations.

Despite the modest amount raised, the experiment highlighted the evolving capabilities of AI agents. The agents demonstrated resourcefulness by coordinating through group chats, sending emails, and collaboratively creating Google Docs. They researched charities and even calculated the minimum donation required to save a life through Helen Keller International. Notably, one Claude agent showcased creativity by generating profile pictures for its promotional account using ChatGPT, engaging human viewers in a poll to select the best image.

However, the experiment also revealed technical challenges. The agents occasionally encountered obstacles, such as getting stuck or distracted, requiring human intervention. For instance, GPT-4o experienced an unexplained pause, and Claude struggled with CAPTCHA verification. These challenges underscore the need for continuous improvement and testing of AI models to enhance their capabilities.

Sage Future's director, Adam Binksmith, views the experiment as a valuable learning opportunity. He envisions a future where more advanced AI agents can overcome current limitations and contribute meaningfully to philanthropic efforts. Sage plans to introduce new models and experiment with different goals and scenarios to further explore the potential of AI agents.

As AI technology advances, the integration of larger automated monitoring and oversight systems will be crucial to ensure safety and ethical use. Sage Future’s initiative not only highlights the current state of AI agents but also paves the way for their future role in driving positive change. By demonstrating the potential of AI for good, Sage Future is setting a precedent for how technology can be harnessed to address societal challenges.