A Natural Language Processing Model for building consensus
What it does.
This tool uses ChatGPT to build a Natural Language Processing (NLP) model useful for analyzing and highlighting common elements between different political speeches. For example, it can identify the frequency and correlation of themes across various political ideological options. Furthermore, the tool can generate artificial speeches that emphasize the factors of consensus within a society’s political landscape.
Value proposition for the government/other partner.
This tool empowers governments and other stakeholders to generate evidence in support of initiatives aimed at achieving political consensus. Thus, the tool has potential for decreasing expressions of political violence within societies and for strengthening good governance.
Why and when to use it.
This tool is suitable for organizations concerned about the levels of increasing polarization in their societies and wishes to explore how technology can be an ally in tackling social and political divisions. Moreover, this tool would be most beneficial when the organization is interested in discovering vectors of consensus across different political views. Additionally, this tool is most useful when there is good-quality, easily available, and trustworthy data regarding political speeches. These conditions not only guarantee the technical success of the model but can also enhance its legitimacy. In addition, the tool is most useful when you can gather a multidisciplinary team.
Known issues and troubleshooting.
The tool won’t be the best option if there is a lack of sufficient and reliable data. If the data is not easily available, you may end up consuming much of the time and resources of the project in collecting the required data and even fail to obtain the necessary data. Furthermore, if there is no reliable data from official or socially accepted sources you may end up losing the legitimacy and credibility of the project and being targeted of bias accusations.
Also, if your team lacks the necessary skills and you can’t bring in an external expert, then this tool is also not your best fit. Building and implementing the tool requires computing skills, as well as knowledge in discourse analysis, political science, and history, among others. The richer the skill set of your team, the better the model will turn out to be.
Finally, keep in mind that ChatGPT offers an AI-based model with limitations in its explicability regarding the outputs obtained. This means that you will reach points where you will not be able to explain how the algorithms have operated. In political contexts marked by a high level of distrust between parties, the limitations regarding the explicability of the model can become a major issue.
Context.
The development of this tool has been part of the Argentina Lab portfolio of actions regarding AI, which we have named ‘AI: Argentine Intelligence.’ All these actions are aimed at promoting public debate on AI in our country. In relation to this tool, we utilized ChatGPT 3.5 to train an AI model for analyzing speeches delivered during presidential inaugurations and legislative session openings. We studied a total of 52 speeches and conducted three analyses. The first involved a descriptive analysis of the speeches, evaluating their length and the frequency of keywords. The second analysis aimed to determine if the speeches addressed some specific public policy issues and how they approached these topics. Lastly, we concentrated on generating a unified discourse by processing all the speeches.
Cost to implement.
The cost of hiring the required team to plan, coordinate and implement the project ranges between USD 15,000 and USD 25,000.
Time
Between planning, building, implementing, and evaluating the tool, you can expect to spend a minimum of two to three months.
People.
To implement a Natural Language Processing Model to build political consensus, you will need a team of 4 members with the following roles:
- One person responsible for the general direction of the project.
- One computing specialist to set up the algorithm.
- One project assistant to support on data collection.
- One designer for managing all the visual aspects of the project.
Focal point.
Country, year, and language.
Argentina, 2023-ongoing; Spanish and English.
Resources.
- Blog (English) – A Conversation about Artificial Intelligence: What do Argentines talk about when we talk about AI?
- Blog (Spanish) – Una conversación sobre la inteligencia artificial:¿De qué hablamos los argentinos cuando hablamos de IA?
- Blog (English) – A Conversation about Artificial Intelligence: Anyone Can Cook and Use AI!
- Blog (Spanish) – Una conversación sobre la inteligencia artificial: Todos pueden cocinar ¡y usar IA!