Due to the swiftly evolving landscape of artificial intelligence, the American Mathematical Society (AMS) deemed it necessary to form a research and policy development Advisory Group on Artificial Intelligence and the Mathematical Community. This initiative reflects a strategic commitment to harnessing AI in ways that profoundly impact the mathematical disciplines.

In its deliberations, the advisory group considered ways to integrate generative AI into their analytical processes. Generative AI’s ability to detect gaps in works such as this makes it capable of offering perspectives perhaps previously unconsidered.

Here, we’ll explore the application of multi-persona prompting, a prompt engineering technique to optimize the output of conversational chatbots like ChatGPT, to refine and enrich the discourse within specialized advisory groups. We employed this technique on ChatGPT by asking it to analyze a series of white papers released by the AMS Advisory Group.

We looked at how this approach can address complex problems by synthesizing varied expert insights into a cohesive and strategic narrative. The results not only highlight the capabilities of generative AI but also set the stage for its future applications as an assistive tool in policy formulation and organizational strategy.

Emerging Importance of Multi-Persona Prompting

Multi-persona prompting has the potential to become a helpful tool in the strategic toolkit of many organizations. By prompting a generative AI model to simulate the viewpoints of various experts, this approach enables a multifaceted exploration of complex issues, drawing on the distinct perspectives of diverse personas to enhance decision-making processes.

It has the potential to facilitate a deeper understanding of the challenges at hand, leading to more robust and innovative solutions. In the context of the AMS, the application of this technique aligns well with their mission to anticipate and shape the future of mathematics in relation to AI.

Application of Multi-Persona Prompting

The AMS Advisory Group made an important decision to explore the intersection of AI and mathematics through their series of white papers. These documents address the contribution of AI to mathematics, the impact of machine learning on diversity, equity, and inclusion within the field, and the role of generative AI in educational research and publishing.

The multi-persona prompting technique involves asking a generative AI chatbot, in our case ChatGPT, to simulate the insights of various expert personas, each contributing unique perspectives that enrich the group’s understanding and strategic output. By examining the AMS Advisory Group’s white papers through the lens of generative AI, we can potentially find gaps in the group’s narrative.

The practical application of this prompting method involves several key steps.

• Individual analysis. Each persona, representing a committee chair or expert, analyzes the issues using their specialized knowledge, ensuring a comprehensive review tailored to specific areas of concern.
• Collaborative exploration. Simulated panel discussions allow these personas to exchange insights, fostering a richer understanding of the issues and their broader implications.
• Strategic synthesis. The final step involves integrating these diverse perspectives into a cohesive narrative, which can serve as the foundation for robust policy recommendations. This is the critical step where humans play a role in providing a reality check to the generative AI work that has been done.

Analysis and Findings

During the simulation phase of the exercise, ChatGPT simulated members engaging in a dynamic exchange of ideas, focusing on the implications of their white papers and building a comprehensive understanding of AI’s impact across different aspects of mathematics. Next, we had the initial simulation pass its findings on to a secondary simulation: This time the chairs of the society’s five policy committees. Here’s how a snippet of that synthetic conversation unfolded:

• Development and oversight. Carla (science policy) and Sarah (equity, diversion, and inclusion) proposed creating a joint task force to oversee the development and deployment of AI tools within the AMS, ensuring they adhere to ethical guidelines and promote equity.
• Educational applications. Christine (education) advocated for pilot projects that integrate AI tools into mathematics curricula, with input from Laura (publications) on how these tools are referenced and cited in educational materials.
• Professional development. Raphael (profession) suggested workshops and training sessions that prepare mathematicians for a landscape increasingly influenced by AI, potentially supported by policy initiatives that Carla (science policy) would coordinate.

While ChatGPT’s simulation of multi-persona discussions offers valuable perspectives, analyzing its output remains crucial. This analysis helps identify any biases or oversights that may persist in the AI-generated dialogue, ensuring that the recommendations are both balanced and comprehensive.

Ultimately, it is the human insight that prevails, as the nuanced understanding and decision-making capabilities of the committee members are irreplaceable. This critical step ensures that AI’s contributions are effectively integrated into human-led processes, rather than overshadowing or dictating them.

Future Implications and Conclusion

The application of multi-persona prompting showcases not only the adaptability of AI technologies in complex decision-making scenarios but also sets a blueprint for its future use in organizational strategy and policy development. This method exemplifies how generative AI can assist policymakers in discovering comprehensive solutions, ultimately fostering well-aligned and effective policies.

Generative AI should be viewed as a partner that enhances human capacities, providing insights and efficiencies without displacing the critical judgment and ethical considerations that only humans can offer. By augmenting rather than replacing human effort, it serves to elevate the effectiveness of human-led processes, ensuring that AI’s capabilities are harnessed to complement and extend our own skills and judgments.

Author’s Disclaimer

The perspectives expressed herein are derived from an analytical exercise and do not necessarily represent the official stance of the AMS or any associated entity.