Even though our work cycle on AI is still at an early stage, we can highlight some key insights regarding the role of AI as a tool for supporting political consensus and for advancing democracy in general. Thus, from what we learned during this cycle, we can point out the following:   - AI and polarization: With the help of AI, presidential speeches can be used as raw material for political consensus. Various studies demonstrate the effects that certain AI algorithms have on the dissemination and replication of discourses and content that exacerbate divisions and political polarization (Van Bavel et al., 2021). However, explorations regarding alternative uses aiming for opposite results, that is, emphasizing shared values and consensus-building, are still scarce. Departing from identifying the issues and topics that are common to all mandataries, regardless of any political differences, generative AI can create speech pieces highlighting points of consensus. This opens the opportunity to advance work/investigation agendas that go beyond the view of political speeches as vehicles of inflammatory sentiments and disunity.  - Human-machine interaction: Generative AI is not a silver bullet. To obtain quality outputs, the process of generating speeches with AI required a great deal of human intervention. Most importantly, human decision-making was the essential element behind the process. We decided what speeches were worth using for training the AI, what semantic fields the AI should look at, and when to penalize the AI for its mistakes. Additionally, the process of using AI has required the interplay of the skills of all the members of our team. From political science, data analytics, programming, discourse analysis, political history, all are skills our team have put into practice.   - Not all that new. Technology with a past and learning from other regulatory models: Even though the launch of AI tools such as ChatGPT has created a whole new hype around the topic, the technology is not new, nor are the regulatory efforts. The developments in AI that we are seeing today are the continuation of a path that can even be traced back to the Middle Ages and it has to do with the long-standing interest of using technology for creating augmented intelligence. Moreover, it is not the first time that governments and societies deal with the issue of regulating technology. Thus, learning and adapting mechanisms can be set in relation to regulations in other fields. Regulatory models from industries such as pharmaceuticals, aviation, or nuclear can serve as valuable examples. By studying and adapting these frameworks, we can develop effective approaches to regulating AI.  - Not yet a national and regional ecosystem: The contact with experts has shown us that, in our country, several AI projects are being developed but with little connections between researchers and practitioners at a national level, as well as few instances of collaboration with colleagues within the region. AI in Argentina is not on the public agenda. There are excellent experts, many of whom are internationally recognized, but there is no convergence among them. There is also no cohesive ecosystem that can help define the country's strategic priorities for integrating AI with its development. From the government's perspective, conversation about AI is not a priority. There have been few efforts in terms of strategy and regulation, and what little has been done lacks continuity between one government and the next.  - Geopolitical concerns: There are specific geopolitical concerns in relation to the development of AI; for instance, there are fears that the increasing competition between global powers will trigger a process of technological nationalism. As it happened during the Cold War, barriers to the free flow of hardware, software and know-how can be put into action. This poses a significant risk for developing countries, especially those reliant on foreign technology for their development. 

Building on what we already accomplished, we aim to keep scaling up our AI portfolio. We are working on a knowledge product that further explores how AI can be used for analyzing common values across different presidents and periods of time. Additionally, we are developing others knowledge products related to employment in the era of AI, and regulations on AI.  We hope to engage in conversations with further private partners, and civil society organizations to promote a greater reach of our actions regarding AI. Furthermore, we plan to approach the newly elected national authorities to share our work, hoping to engage them in our challenge.

Programa de Naciones Unidas para el Desarrollo [PNUD] (2023). IA: Inteligencia Argentina. https://www.undp.org/es/argentina/proyectos/ia-inteligencia-argentina    Programa de Naciones Unidas para el Desarrollo [PNUD] (2023). Mucho en Común: Explorando los temas de los 40 Años de Democracia Argentina a través de IA. https://inteligenciaargentina.org/    Moscovich, L., Vallejo Vera, S., Alzú, M.S., Gómez, F., Zapata, M. y Moreno, M. V. (2023). Mucho en común. Los temas y el discurso de la democracia en sus 40 años. Buenos Aires: PNUD. Disponible en: https://www.undp.org/es/argentina/publicaciones/informe-ia-inteligencia-argentina    Moscovich, L.. y De Rosa, A. (2023, October 9). Episode II. What do Argentines talk about when we talk about AI? (an example). UNDP Argentina.  Available at https://www.undp.org/es/argentina/blog/episode-ii-what-do-argentines-talk-about-when-we-talk-about-ai-example    Moscovich, L., y De Rosa, A. (2 de octubre de 2023). Episodio II. ¿De qué hablamos los argentinos cuando hablamos de IA? (un botón de muestra). PNUD Argentina.  Disponible en https://www.undp.org/es/argentina/blog/episodio-ii-de-que-hablamos-los-argentinos-cuando-hablamos-de-ia-un-boton-de-muestra    Moscovich, L. (2023, September 1).AI: Argentine Intelligence. A Conversation about Artificial Intelligence: Episode I: Anyone Can Cook, and Use AI! UNDP Argentina. Available at https://www.undp.org/es/argentina/blog/ai-argentine-intelligence-conversation-about-artificial-intelligence    Moscovich, L. (1 de septiembre de 2023). IA: Inteligencia Argentina. Una conversación sobre la inteligencia artificial: Episodio I. Todos pueden cocinar ¡y usar IA! PNUD Argentina. Disponible en https://www.undp.org/es/argentina/blog/ia-inteligencia-argentina-una-conversacion-sobre-la-inteligencia-artificial 

This work is not only a proof of concept; it is also a prototype for testing the potential of AI to support democratic governance. The action plan is comprehensive, involving stakeholder mapping, interviews with key informants, blogs, and this specific application. The long-term ambition is to leverage collective intelligence to identify past and future plans for AI and to work towards regulating the technology to balance its possibilities and risks.   We developed a specific code to train ChatGPT Plus to analyze texts. To generate the unified speech and analyses of the different topics, we trained an AI model using ChatGPT Plus, which analyzes the shared points and values among the speeches. Additionally, the frequency and associations of 27 topics of public and strategic policy for the country were analyzed, exploring how often they appeared in connection with specific ideas and concepts. The results demonstrate quality and consistency in the outputs at different model specifications, such as temperature (how deterministic or liberal the speech is), frequency penalty (how much repetition is allowed), and various prompts (how much more or less specific the responses are). Therefore, these findings could enable the application of these models to the generation of fictional text in survey experiments (vignettes) and the evaluation of a voter's ability to differentiate between text generated by humans and computers. This is relevant for studies on the proliferation of fake news in political campaigns. We choose them to have a better understating of the state of the art uses of AI for democracy, as well as what the possibilities and limitations are in relation with this technological tool. In regards to the interviews and review of specialized sources, we were able to identify the uses of AI for political consensus as an original field worth exploring. In addition, the data gathered led us to obtain a clearer understanding of the technical nuances of AI that could have an impact on democracy.   The public launch of ChatGPT has intensified the discussions on AI. Open letters, research papers, and opinion pieces are being published daily, so much so that it is hard to keep track of all new developments in the discussion. Thus, we also choose our data as an effort for fulfilling the need of having a moment of convergence regarding what is been discussed in relation to AI, especially in our country

Exploring is a suitable method for building a robust foundation from which start working the relation between AI and democracy. It helps to have a sense of state-of-the-art of the technology, while paying attention to emergent patterns.   On the other hand, testing is a perfect method for designing a novel application of AI and evaluating its strong points and weaknesses. Thus, we hope to create powerful evidence of potential uses of AI for democracy.

The combination of these elements contributes to creating valuable insights on a technology still surrounded by a great deal of uncertainty regarding its impacts. The contact with partners specialized in the topic, together with data gathering from traditional and not traditional sources (podcast, social media, blogs), is allowing us to explore analytics dimensions, variables and indicators that could be helpful to try to make sense of the interaction between AI and democracy. Moreover, by testing an AI tool for politics speech analysis and generation, we hope to create evidence of the possibilities and shortcomings of this technology in relation to democracy. Moreover, our work has a formative effect on the national and regional AI ecosystem. At the same time we are making efforts for studying the AI ecosystem, we are contributing to its formation. By highlighting experts, projects and organizations, and by connecting the dots regarding who is doing what and where, our work is laying the foundations for an emerging ecosystem.