What we’re learning from using generative AI in our innovation work
NEW TECHNOLOGIES, PORTFOLIO DEVELOPMENT
Arianna Friedman, Innovation Intern, with contributions from Aditi Soni, Service Designer, UNDP IRH, Andreas Pawelke, Learning, Systems Thinking & Digital Specialist, UNDP IRH, and Shreya Krishnan, Portfolio Designer, UNDP IRH
For the Innovation Team at UNDP Europe and Central Asia, much of our work revolves around creating, storing, managing, and sharing knowledge. Approaching the latest AI developments with a blend of openness and discernment, we have started test driving generative AI tools and technologies to expedite ideation processes and streamline data analysis to see how it enhances our work.
We recently organized an internal hands-on learning session to explore various use cases for generative AI. During the session, we discussed how tools like ChatGPT can be set up as a reasoning or conversational partner to help us think through ideas for project activities. Furthermore, rapid improvements to platforms like Midjourney also position generative AI as a valuable tool for improving our storytelling capacity and attracting new audiences, partners, and donors.
Beyond automating some of our daily tasks, we are exploring how generative AI supports our new ways of working. Our shift from conventional project formats to a portfolio approach is transforming how we perceive, understand, and respond to various issues – from the future of work to depopulation. In the process, we are finding new methods to improve our readiness to address obstacles as we push the boundaries of development work in ever changing contexts – from adapting our methodologies to harnessing and embracing new technologies and partners.
Synthesis software for amplifying analytical capabilities
Combating societal distrust through social listening, also called community or deep listening, is at the core of our portfolio approach. Social listening exercises can take many forms, but the overall intent remains the same: driving transformative change at the local level.
Social listening exercises involve compiling a diverse collection of narratives in a city or region, extracting patterns of people’s lived experiences, and analyzing how these encounters are shaped by underlying values and behaviors, as well as evolving needs, challenges, and opportunities. These insights are integral to understanding what is and is not important to citizens. As a compass for action, they pave the way for aligning initiatives and government decision-making with the collective pulse of residents. This coordination, established through an intimate understanding of citizens’ perceptions and experiences, becomes the foundation for generating meaningful outcomes that embrace inclusivity as their guiding principle.
To explore the application of generative AI in social listening, we integrated Notably and ChatGPT into our workflow to test their effectiveness in accelerating data sorting and insight generation based on inputs from social listening sessions. For example, to analyze a wide range of challenges and experiences, we generated a conversation script using Notably’s qualitative tool. While this initial step eased our workload, manual intervention was necessary to highlight and code “want” statements – an area where AI fell short. But after tagging the script, the AI was able to generate concise summaries, offer thematic insights, and even generate a persona based on the applied tags.
However, Notably’s AI treated all statements as equal and could not prioritize which ones would be more important than others based on the context of the interview. This resulted in generic responses. On the other hand, ChatGPT’s text-prompt interface proved more versatile since it responds to detailed instructions, like counting the number of statements, giving a thematic analysis, and ranking things in importance. Nevertheless, ChatGPT’s analytical capabilities still lacked the granularity required to distill precise and comprehensive thematic insights.
To test other use cases for generative AI, we used ChatGPT to craft different citizen profiles for a mock interview activity during our five-day, in-person portfolio design and management bootcamp. The purpose of the simulated listening exercise was to immerse participants in the intricacies of social listening through role-playing scenarios. Using ChatGPT to generate profiles was an invaluable time-saving measure. Preparation that would typically demand a week’s effort was streamlined to just two days. However, without specific guidance, the platform produced profiles that predominantly skewed towards a singular male perspective, reflecting a narrow segment of society. Intrigued, we opted to retain these homogenous profiles with the intent of gauging participants’ discernment of potential biases. Those involved were quick to point out the lack of female perspectives, but it took more prompting to identify the lack of socioeconomic representation.
A new dimension to co-creation exercises
On a national level, UNDP country offices in Europe and Central Asia are also exploring ways to integrate AI in their work processes. Already, two country offices have tested and piloted a new partnership with Urbanist AI, a participatory platform for urban planning and co-design that integrates multiple generative AI technologies.
To build capacity for co-creation workshops and enhance social listening activities as part of our portfolio approach, UNDP North Macedonia’s Accelerator Lab worked with Urbanist AI to customize a half-day community workshop on redesigning parking lots, busy streets, and other public spaces in the context of cooling heat islands to combat climate change in Skopje. In a similar exercise, the UNDP Kosovo* led a workshop to collaboratively envision and develop child-friendly open spaces like parks for kids and teens in Pristina. Both meet-ups are linked to programmes under the City Experiment Fund, an initiative funded by the Slovak Ministry of Finance that engages cities in systems thinking-based processes to address urban challenges.
In both scenarios, participants were first divided into groups and sent out to document urban issues, and then were tasked with redesigning the areas in mini design charrettes. Participants applied Urbanist AI’s text-to-picture function to rapidly render lush landscapes and social infrastructure on top of existing urban spaces and used those images to engage in lively, constructive discussions on reimagining how the space could be optimally utilized now and in the future. From young children to seasoned bureaucrats, the workshop captured the attention of a wide range of participants.
From a vacant landscape…
…to a potential park
From a parking lot…
…to a shaded open-air market
Images: UNDP North Macedonia
Throughout these sessions, Urbanist AI emerged not only as a tool, but also as a mode for getting people together. Attendees enjoyed engaging in the exercises and had positive learning experiences, which we hope will magnify the impact of UNDP’s participatory design workshops. As we fine-tune our approach, our next step is figuring out how to extend the benefits of platforms like Urbanist AI beyond collaborative co-creation exercises to synergize other aspects of the portfolio process. Additionally, we’re exploring potential partnerships that could be unlocked through innovative applications of such platforms.
From an empty lawn…
…to an activated outdoor space
From minimal play amenities
…to a colorful playground
Images: UNDP Kosovo
Final takeaways
We find ourselves in the early stages of the AI revolution, eagerly observing the development of this technology and its adaptive applications. Its current constraints and biases are not inherently fixed; rather, they mirror the collective shortcomings of humanity. As we assess the outcomes of these learning experiences and refine our approach to continuous innovation, our decisions about how we move forward and create systems that reduce bias carry even greater significance.
Our commitment stands firm in addressing the downsides associated with generative AI. These challenges include the displacement of jobs due to automation, the threat of misinformation and deep fakes undermining democracy, biases present in training data, privacy and ethics concerns, and the ongoing issue of the digital divide. But we are also optimistic about a new horizon of possibilities. Our trials taught us that process and methodology can be enhanced by generative AI. While there is a learning curve to crafting the right queries, it is possible to guide the output in the right direction, especially when using tools that allow for step-by-step instruction input.
We foresee generative AI as being a crucial resource in supporting our new way of working with cities and communities. Amid resource constraints and the demands of our portfolio design and management initiatives, such as the City Experiment Fund and Mayors 4 Economic Growth, generative AI could be especially promising for practitioners in this space. As AI technology improves, these facilities will become indispensable accelerants and aides in the work of practitioners and partners who are stretched thin, navigating a complex landscape of demands.
* All references to Kosovo shall be understood to be in the context of the Security Council Resolution 1244 (1999).
Everyone has the right to live in cities that are comfortable, accessible, and environmentally clean, where public spaces prioritize human well-being and inclusivity. Indeed, many […]
The United Nations Development Programme (UNDP) participated in the Cardano Summit 2024, held on October 23-24, building on our strategic partnership with the Cardano Foundation, […]
The United Nations Development Programme (UNDP) in Europe and Central Asia is co-organizing the Blockchain4Good virtual summit on October 30-31, 2024. This event aims to […]