From Chatbots to Policy Makers: AI’s Role in Democratic Decision-Making
Sung Jun Han, 2024-25 RPW Center Graduate Student Fellow. This year’s group is exploring the theme of Emerging Technologies in Human Context: Past, Present, and Future.
Have you ever wondered how systems, from simple tools to complex technologies, make decisions? One of the most widely used frameworks to describe this process is the Input-Process-Output (IPO) model. As the name suggests, the IPO model breaks down a system’s function into three core components: inputs, processes, and outputs. This model is incredibly versatile and can be applied to various decision-making systems, including the realm of Artificial Intelligence (AI).
Understanding AI Through the IPO Model
Imagine using a chatbot. You ask a question (input), the chatbot sifts through its database to find the most relevant answer (process), and then it gives you a response (output). While the specifics of each component might vary, the IPO model provides a natural framework to understand AI’s decision-making systems.
But what happens when we view AI’s decision-making through the IPO model in a democratic context? Each component of the IPO model—input, process, and output—invites different types of questions:
1. Inputs: Who Makes Them?
The “who” question is critical here. In a democracy, inputs can come from citizens, making the process democratic. For example, if citizens collectively provide their preferences for presidential candidates to an AI system, the inputs are democratic. However, if the inputs come from a disconnected, oligarchic group, the process lacks democratic integrity.
2. Processes: Is It Truly AI?
The crux of AI decision-making lies in the algorithmic process. The complexity of these algorithms, especially the extent of human interference, determines whether the decision-making system is genuinely AI. If the process lacks a complex algorithm, it may not qualify as an AI decision-making system.
3. Outputs: Binding or Non-binding?
Outputs can either be binding or non-binding. For instance, asking ChatGPT for fashion advice doesn’t bind you to its suggestion. However, if a group decides to follow whatever the AI suggests, the output becomes binding. In political contexts, this could mean AI enacting laws or policies versus merely providing advice.
Categorizing AI Decision-Making Systems
Based on these distinctions, we can categorize AI systems into four categories using the IPO model:
1. Democratic Inputs:
- Binding Output: Democratic Decision-maker (e.g., Enlightened Preference Voting*)
- Non-binding Output: Democratic Non-decision-maker (e.g., ChatGPT, Gemini, Claude)
2. Non-democratic Inputs:
- Binding Output: Non-democratic Decision-maker (e.g., Multivac**)
- Non-binding Output: Non-democratic Non-decision-maker (e.g., Fitbit, YouTube algorithm, search engine algorithm)
The Democratic Justification
For my project, I plan to argue that only democratic non-decision-makers and certain limited models of democratic decision-makers align with democratic ideals like self-government, political equality, and non-domination. While my project will touch on the regulation of algorithms, it will not delve deeply into current debates on fairness, transparency, and accountability. Instead, it aims to establish a moderate idealization, laying the groundwork for discussing AI in democracy once acceptable criteria for algorithm construction, regulation, and monitoring are in place.
In essence, by understanding AI through the IPO model, we can better navigate the complexities of integrating AI into democratic systems, ensuring that the core values of democracy are upheld.
* Enlightened Preference Voting is a system where all citizens provide inputs on election day, including their preferences, answers to political knowledge tests, and background information. This predetermined decision-making procedure uses these inputs to generate an outcome that reflects what citizens would ideally want if they were fully informed about the issues.
** In Isaac Asimov’s story “Franchise,” Multivac is a supercomputer that determines the outcome of elections. It does this by asking questions to the most representative American voter, whose responses are used to make the final decision.
Sung Jun Han is a doctoral candidate in the Philosophy Department at Vanderbilt University, specializing in political philosophy and the intersection of AI and democracy. He is deeply engaged in researching what he calls “Madisonian Lottocracy,” a topic central to his dissertation. Jun has presented a short version of his work at the Philosophy, Politics and Economics (PPE) Society meeting, as well as at the APA 2024 Eastern Division meeting.