Insights from a first Roundtable Discussion on AI and Behavioural Science

Written with the help of Claude 3 (Opus)

The rapid advancements in artificial intelligence (AI) have sparked discussions across various fields, including behavioural science. As AI tools become more sophisticated and accessible, behavioural scientists are exploring ways to leverage these technologies to enhance their research, improve decision-making, and tackle complex societal challenges. To delve into the intersection of AI and behavioural science, a roundtable discussion was organized, bringing together leading experts in the field.

The participants of the roundtable included Elina Halonen, Sarah Osman , Abigail Emery , Colin Strong (FMRS) , and Laura de Molière , each bringing unique perspectives and experiences to the conversation. The discussion revolved around several key themes, including the practical applications of AI in behavioural science, the role of behavioural scientists in AI adoption and understanding, the use of AI as a problem-solving tool, and the potential for AI to elevate the field of behavioural science.

AI in practice: Building knowledge retrieval engines and improving behavioural science application

Elina shared her experiences with building knowledge retrieval engines using AI tools. She encountered challenges in ensuring that the AI uses the provided documents accurately and consistently. Elina noted that the complexity of the topic significantly influences the difficulty of building a reliable knowledge retrieval system. For instance, while creating a system for veterinary information was relatively straightforward, building one for nudge theory proved more challenging due to the varied conceptualizations and schools of thought within the field.

Elina emphasized the importance of curation when building AI tools, particularly for topics with diverse viewpoints or epistemological differences. She also highlighted the significance of clearly defining the use case, as the purpose of the tool determines its design and functionality. For example, a knowledge retrieval system for question-answering requires a different approach than a decision support tool, which necessitates careful consideration of the user's decision-making process, potential inferences, and communication of information.

Sarah discussed how AI can help improve the application of behavioural science in international development. She noted that AI could be a valuable tool in translating complex theoretical concepts into practical guidance for professionals working on projects in various contexts. By providing real-time, language-specific, and context-aware information, AI can help bridge the gap between theory and practice, enabling community health workers and other professionals to better understand and address the barriers affecting decision-making among their target populations.

Sarah also highlighted the potential of AI in analyzing social media data during crises, such as the war in Sudan, to inform humanitarian work and service delivery. By processing and interpreting user-generated content, AI can provide valuable insights into the situation on the ground, helping organizations respond more effectively to the needs of affected communities.

Both Elina and Sarah stressed the importance of inclusive design and strategic partnerships when developing and deploying AI products and services. They emphasized the need for diverse perspectives and collaboration to ensure that AI tools are accessible, culturally sensitive, and aligned with the needs of their intended users.

The role of behavioural scientists in AI adoption and understanding

The participants also discussed the role behavioural scientists can play in helping people adopt AI and overcoming initial disillusionment, noting that while AI tools may be prevalent in certain circles, many individuals, even those in technical roles, may not see the immediate benefits or may become discouraged after initial attempts. Behavioural scientists can contribute by facilitating the adoption of AI, addressing concerns, and highlighting the potential advantages of these tools.

Abigail explained her efforts to help teams across the government use large language models to improve their work. She has been working with various teams, including behavioural science, insight, research, and policy teams, to identify and implement use cases for AI in their respective fields. By providing relevant and accessible case studies and helping people build useful mental models of how the technology works, Abigail has been able to help colleagues overcome their fears and reservations about using AI.

Abigail also discussed the importance of building useful mental models to understand AI's strengths and weaknesses. She noted that human expectations about AI's capabilities often do not align with reality, leading to either missed opportunities or potential dangers. By creating effective mental models, such as the "autocomplete" analogy or the "enthusiastic intern" perspective, behavioural scientists can help users better understand and interact with AI systems.

Furthermore, Abigail explored the significance of natural language input in AI and how it can increase accessibility and creativity. She highlighted how natural language input allows users to accomplish tasks by simply describing them, rather than needing to learn specific software or coding languages. This advancement has the potential to make AI more accessible to a wider range of users, including older individuals, those with less technical experience, and people with accessibility needs.

The role of behavioural scientists in AI adoption and understanding is crucial, as they can help bridge the gap between the technology and its users. By facilitating adoption, building useful mental models, and promoting the benefits of natural language input, behavioural scientists can contribute to the effective and responsible integration of AI in various domains.

AI as a tool for problem-solving and elevating behavioural science

Colin discussed his perspective on using AI as a tool to solve specific problems. He emphasized the importance of starting with the problem itself and then choosing the appropriate AI tool, rather than beginning with the tool and trying to find a problem it can solve. Colin likened this approach to his experience growing up in a household of engineers and skilled craftspeople, where the focus was always on selecting the right tool for the job at hand.

Colin also noted that AI seems to be more effective when dealing with codified or structured information. He observed that when using AI to generate behavioural diagnoses and interventions based on a pre-defined framework, the results were quite promising. However, he cautioned that AI should not be viewed as an all-encompassing solution, but rather as a tool that can be applied to specific problems.

Laura explored how AI can help elevate the field of behavioural science to better deal with complex problems. She discussed the potential applications of AI in behavioural science, such as agent-based modeling, which could help researchers understand complex systems and emergent properties before designing interventions. Laura also mentioned the possibility of using AI to enhance qualitative data analysis, enabling researchers to draw more insights from qualitative data by understanding belief systems and adjusting the direction of inquiry in real-time.

Laura emphasized the inherent benefits of engaging with AI, such as fostering discipline and deliberation in problem analysis. She noted that interacting with AI forces users to be more definite and disciplined in their problem analysis, as they need to clearly articulate their thoughts and break down the steps involved in the process.

Both Colin and Laura recognized the potential of AI as a tool for problem-solving and elevating behavioural science. By starting with the problem and selecting the appropriate AI tool, researchers and practitioners can more effectively address the challenges they face. Furthermore, by exploring the applications of AI in areas such as agent-based modeling and qualitative data analysis, behavioural scientists can expand the boundaries of their field and tackle complex societal issues.

Challenges and considerations

Throughout the roundtable discussion, the participants raised several challenges and considerations regarding the use of AI in behavioural science. One of the primary concerns was the need for inclusive design when developing and deploying AI products and services. Sarah emphasized the importance of working with diverse stakeholders, such as disability-inclusive organizations, to ensure that AI tools are accessible and meet the needs of all users. She also highlighted the potential risk of perpetuating existing inequalities if AI is not designed with inclusivity in mind.

Another challenge discussed was the potential for reactance in the workplace when introducing AI tools. Sarah noted that some individuals might be skeptical or resistant to using AI, viewing it as a threat to their jobs or expressing a lack of trust in the technology. To address this issue, the participants emphasized the need for careful introduction and integration of AI in the workplace, focusing on the benefits it can provide and how it can augment, rather than replace, human expertise.

The importance of curation when building AI tools was another key consideration raised by Elina. She stressed that the quality and compatibility of the data fed into AI systems significantly impact the reliability and usefulness of the outputs. When dealing with complex topics or diverse viewpoints, careful curation is essential to ensure that the AI tool provides accurate and coherent information.

Additionally, the participants discussed the need for diverse perspectives and collaboration in the development and decision-making processes surrounding AI. They expressed concern about the concentration of power among a few influential individuals in the tech industry and emphasized the importance of involving a wider range of stakeholders, including behavioural scientists, in shaping the future of AI.

Addressing these challenges and considerations is crucial for the responsible and effective integration of AI in behavioural science. By prioritizing inclusive design, managing potential reactance, ensuring proper curation of data, and fostering diverse perspectives and collaboration, behavioural scientists can navigate the complexities of AI and harness its potential to advance their field.

Future directions and opportunities

The roundtable participants shared their thoughts on the future directions and opportunities for AI in behavioural science. They expressed excitement about the potential for AI to democratize access to powerful tools and insights, enabling a wider range of individuals and organizations to benefit from behavioural science knowledge. As AI continues to evolve and become more user-friendly, it has the potential to empower practitioners, policymakers, and researchers to make more informed decisions and develop more effective interventions.

The participants also discussed the role of AI in fostering creativity and innovation within the field of behavioural science. By leveraging AI's ability to process vast amounts of data, generate novel ideas, and identify patterns and connections, behavioural scientists can explore new avenues of research and develop fresh approaches to complex problems. AI can serve as a catalyst for interdisciplinary collaboration, bringing together experts from various fields to tackle societal challenges from multiple perspectives.

Looking ahead, the participants envisioned a future in which AI and behavioural science work together seamlessly to address some of the most pressing issues facing society. From improving public health outcomes and promoting sustainable behaviors to enhancing education and supporting social justice initiatives, the potential applications of AI in behavioural science are vast and promising. As the field continues to evolve and mature, it is essential for behavioural scientists to stay engaged with the development of AI, shaping its trajectory and ensuring that it is used in an ethical and responsible manner.

Conclusion

The roundtable discussion on AI and behavioural science provided valuable insights into the current state and future potential of this rapidly evolving field. The participants shared their expertise and perspectives on the practical applications of AI, the role of behavioural scientists in AI adoption and understanding, and the use of AI as a tool for problem-solving and elevating behavioural science.

Key takeaways from the discussion include:

1. AI has the potential to revolutionize the way behavioural science is practiced, enabling researchers and practitioners to tackle complex problems more effectively and efficiently.
2. Behavioural scientists play a crucial role in facilitating AI adoption, building useful mental models, and ensuring that AI tools are designed and used in an inclusive and ethical manner.
3. AI should be viewed as a tool to be applied to specific problems, rather than an all-encompassing solution.
4. The future of AI in behavioural science is bright, with numerous opportunities for innovation, collaboration, and positive societal impact.

As the field of behavioural science continues to evolve and integrate with AI, it is essential to acknowledge and address the challenges and considerations that arise, such as prioritizing inclusive design, managing potential reactance, ensuring proper data curation, and fostering diverse perspectives and collaboration.

The roundtable discussion highlighted the importance of ongoing dialogue, knowledge-sharing, and collaboration within the behavioural science community. Moving forward, maintaining an open and proactive approach to AI while remaining mindful of its limitations and risks is crucial. By working together and leveraging the power of AI responsibly, behavioural scientists can unlock new frontiers of understanding and make significant strides in improving the well-being of individuals and society as a whole.