The AI Political Party

Modelling AI political party: Operating system for better democracy

Author: SEON GYU GO (Fukushimagakuin University, Korea)

Abstract

The article discusses the democratic potential of the adoption of AI technologies that overcomes the intrinsic challenges in the present-day representative democracy. As one of the pioneering works pondering on the possibilities of AI parties and AI agents in the Korean political landscape, the article provides theoretical frameworks of delegation and AI agents in augmented reality. Drawn on the theoretical understanding, the article proposes a future of democracy in coexistence with AI agents and humans in AI political parties, and examines current institutional hurdles towards the vision. AI agents and AI political parties can streamline democratic processes and invite more citizens, who are marginalised in traditional representative democracy, to political decision-making process. The article contributes to achieving better democracy by discussing the possibilities and the risks of doing politics with AI.

Keywords: hyper democracy, augmented democracy, AI agent, AI political parties, AI politics

[국문초록] AI 정당의 이론적 · 제도적 설계와 민주주의적 함의

기술발달은 AI가 정치과정이나 통치 수단으로 활용될 가능성을 높이고 있다. 이미 AI 정치가, AI 정당이 등장하고, 선거에 출마하고 있다. AI가 정책결정 과정에도 참가하여 의사결정의 주요한 수단으로 활용되고 있다. 최근에는 세계 여러 지역에서 전개되고 있는 전쟁에서도 AI 무기가 작전을 수행한다. 이렇듯 AI는 정치 분야에서 도입되고 있다. 지금까지 AI는 다양한 분야에서 자동화 수단으로 활용되고 있다. 그러나 정치는 시민의 직접적, 집합적 참여를 기본적인 전제로 성립된다. 정치의 자동화는 주권자, 의사결정의 주체로서 인간이 그 권리로부터 소외, 배제되는 것을 의미한다. 이러한 문제는 AI 정치, AI 정당의 제도화에서 해결해야 할 핵심적 과제이다.

이 논문에서는 액체 민주주의, 확장 민주주의, 하이퍼 민주주의라는 AI 기술, AI 에이전트에 기반한 이론적 틀을 응용하여 AI 정치의 현실적 가능성을 설명한다. 그리고 AI 에이전트의 역할, 기능 등을 정리하고, AI 정당 시스템을 디자인하였다. AI 에이전트에 기반한 AI 정당이 현실적으로 제도화하기 위한, 법적, 제도적 조건을 한국적 정치 현실에 주목하여 구체화하였다. 또한, AI 정당이 현실정치에 주는 영향과 문제점을 도출하였다. 마지막으로 AI 정당 도입이 민주주의 진화에 어떻게 기여하는지를 도구적 관점, 평등주의적 관점에서 제시하였다.

키워드: AI 정치, AI 정당, AI 에이전트, 확장민주주의, 하이퍼 민주주의, 지시와 위임

Table of Contents

  1. Abstract
  2. [국문초록] AI 정당의 이론적 · 제도적 설계와 민주주의적 함의
  3. Ⅰ. Introduction
  4. Ⅱ. Theoretical frameworks: Delegation and sovereignty in the AI environment
  5. Ⅲ. Operating system for AI political parties
  6. Ⅳ. The nexus between political reality and AI Political Parties
  7. Ⅴ. Conclusion: doing politics with AI for a better democracy
  8. VI. About the Author: Seon Gyu Go (고선규)

Ⅰ. Introduction

Representative democracy, which is “the locus of the dynamics that keep modern democracy in motion,” has electoral representation at the core (Urbinati 2011). People periodically elect representatives to whom they delegate power and authority. The election of representatives is done in electoral districts determined by population size or region. Despite being the locus of democracy, the representative system has intrinsic limits (Manin 1997; Rosanvallon 2011). To name a few, power holders can manipulate the system by controlling representative institutions or the election process. Low political participation and voter apathy in representative democracy is a growing challenge across the democratic communities threatening the political legitimacy of elected representatives. For instance, less than a quarter of electorates cast ballots at the election for New York mayoral in 2021. The connection between global financial capitalism and the concentration of power is concerning as well (Virginia Eubanks 2018). This concentration of wealth and political power has had an undue influence on parliamentary politics and individual voting behavior (Wolin 2008). Furthermore, political polarization within societies of representative democracy escalates and political tribalism linked to political parties increasingly grows, creating a more divided and hostile society as observed in the elections in the US and Korea (Scholtz, 2024). Hence, the crisis of representative democracy observed in various countries is seen as a threat not only to political, economic, and social stability but also to humanity’s survival itself (Intergovernmental Panel on Climate Change 2021).

To address the problems of representative democracy, various alternative forms of democracy have been proposed. With the advent of internet technology and the rise of online political spaces, direct democracy, in which voters bypass representatives and make decisions directly, has gained more attention. However, it is practically difficult to implement direct democracy (Mark Baldassare 2007). Voters, let alone professional politicians, cannot simply afford the time. For instance, during the 114th session of the U.S. Congress from 2015 to 2016, a total of 6,536 bills were deliberated in the House of Representatives, an average of nine per day. In the Senate, 3,548 bills were deliberated, an average of five per day, and 329 bills were passed, an average of three bills passed per week (Hidalgo, 2018). The vast number of bills processed exceeds voters’ physical time and cognitive ability to understand the content of these bills.

Therefore, discussions have emerged about using AI to enhance voters’ cognitive processing capabilities. Since 2022, with the popularization of ChatGPT, these discussions have become a reality. ChatGPT 4.0 ranked in the top 10% of test-takers on the U.S. bar exam, and scored 1401 out of 1600 on the U.S. SAT exam, demonstrating logical reasoning, mathematical thinking, reading comprehension, and explanatory skills (Nishimi, 2023). Since ChatGPT, generative AI has evolved into AI agents and digital AI. The emergence of AI agents makes augmented democracy and new forms of AI political parties possible. Augmented democracy involves AI avatars, authorized by citizens, participating in the political process on their behalf. Additionally, a rise of AI parties can create a new democratic system. The AI agents in a democratic system and AI parties are emerging as the notion that only humans can serve as representatives of voters is changing.

As AI technology continues to develop, the potential for its application in political processes and governance increases. Already, AI politicians and AI political parties are emerging in various countries. In some Japanese local governments, AI-based policy decision-making has already been implemented and utilized (Go 2019). In the 2018 local government mayoral election in Japan, an AI robot ran as a candidate. An AI candidate ran in New Zealand’s general election as well. In the 2023 Danish parliamentary elections, an AI political party, Det Syntetiske Parti, participated in the proportional representation vote. Since it is an AI party, the party leader was also AI. This Danish AI party communicates with voters through a chatbot and formulates policies based on data collected and learned from the platforms of minority parties that emerged in Danish politics since the 1970s. In Norway, AI evaluates the policies proposed by parties and politicians. The question of AI political parties is an urgent issue that can no longer be ignored. AI’s involvement as a tool for governance and as an actor in political processes raises issues that are closely tied to the fundamental political issue of power. AI political parties are inherently connected to the essence of politics and have a significant impact on democracy.

Furthermore, there are concerns about bias in internet data, the lack of transparency in policy-making processes, and the difficulty of knowing whose opinions are being reflected in these policies. Democracy is premised on the idea that citizens directly participate in political decision-making, but in an AI party, AI is perceived as making decisions, not the citizens. Despite this, AI parties are expected to become a tool for radical changes in the political process.

The paper is organized as follows: Chapter 2 describes Liquid Democracy and Augmented Democracy as theoretical frameworks and explains the concept of AI political parties. Chapter 3 describes the relationship between AI agents and AI political parties and the working structure of AI political parties. Chapter 4 describes the conditions under which AI political parties are possible in practice and suggests legal and institutional improvements. It analyzes the effectiveness of AI political parties. Finally, Chapter 5 concludes and summarizes the impact of AI parties on democracy.

Ⅱ. Theoretical frameworks: Delegation and sovereignty in the AI environment

Since the French Revolution, democracy has been founded on principles of self-governance, collective decision-making, and individual participation in political decisions affecting the community, with voting as the primary mechanism. However, facing the intrinsic errors of representative democracy, various alternatives have been proposed to address the shortcomings of representative democracy. Discussions have emerged about participatory democracy, deliberative democracy, and liquid democracy (Hardt and Negri 2012). Recently, augmented democracy and hyper democracy have appeared.

The ideas behind liquid democracy and augmented democracy are predicated on the use of digital technology. The impact of AI on democracy can be examined through the instrumental and egalitarian theoretic perspectives of democracy (Inoue Akira, 2022). First, the instrumental frameworks can be divided into political processes and procedural legitimacy (Nwokeafor 2017). Here, AI political parties can contribute to strengthening the latter in particular. Second, from an egalitarian democracy perspective, AI political parties help individuals establish equal relationships with one another as AI political parties promote equal participation in decision-making processes. AI agents assist youth, women, the elderly, and people with disabilities, enabling them to participate equitably. Thus, AI political parties can offer a useful solution to overcoming bureaucratic decision-making in representative politics.

This part of the article focuses on liquid democracy and augmented democracy as they are the two most influential frameworks as alternatives to representative democracy. First, liquid democracy became possible with the development of an information society and social media-based political communication. This political system enables voter participation and voting through the use of social media. The first implementation of liquid democracy through a delegate system was Demoex, a regional political party in Stockholm, Sweden, in 2000, and the German Bundestag used liquid democracy as a method of citizen participation in 2011 (Gonoi, 2018). It is also known as delegative democracy (O’Donnell 1994). In liquid democracy, voters can choose to participate directly in certain issues or delegate their votes to others for other issues. It combines elements of both direct and representative democracy. Furthermore, a delegate can transfer the authority they receive from one person to another, and this delegation process can be repeated indefinitely (Dhillon et al. 2023). The distinguishing feature of liquid democracy is that voters can choose to participate directly in specific issues, unlike traditional representative democracy.

Methodologically, the widespread adoption of blockchain technology, open-source systems like Liquid Feedback and Agora Voting, which use decentralized decision-making processes, enables deliberation and discussion while protecting the privacy of the delegator within the network (Nakamoto 2009). The decentralized nature of blockchain makes it highly compatible with democracy, and its encrypted and tamper-resistant decision-making processes have garnered attention. Furthermore, as big data becomes more integrated into everyday life, all national data is being consolidated. Systems are being established to manage, analyze, and utilize this data at the government level, and data is increasingly being used in political processes (Macnish and Galliott 2020).

However, there has been considerable criticism of the practice of allowing a delegated representative to further delegate to another, ultimately leading to the persistence of minority representation (Paulin 2014). Additionally, liquid democracy relies on internet access, and thus, the issue of the digital divide can limit voter participation. Furthermore, verifying voters’ identity and intent when they participate in decision-making online is a challenge. When participating through the internet, it is impossible to verify whether the person sitting at the terminal is truly the voter. This issue has been a subject of criticism, as seen in the case of Germany’s Pirate Party, where the party’s network devolved into a bureaucratic structure.

To overcome the fundamental limitations of liquid democracy, diversifying the functions of representation is essential, as is automating political functions. The automation of politics can be understood by drawing parallels to self-driving cars or factory automation. A self-driving car operates without a driver, and in factory automation, robots carry out all tasks, including labor, management, and inspection, without human involvement. By automating democratic processes and functions, greater efficiency can be achieved.

Second, augmented democracy can be an alternative to present-day representative democracy. Direct democracy depends on robust political communication. Although AI can enhance political communication, there still remains a problem in direct democracy, which is voters’ cognitive limitations. Augmented democracy is capable of handling large volumes of legislative bills and policy reviews, which are typically dealt with by representative bodies. Political communication on a scale beyond human cognitive capacity becomes possible and voters can make more informed decisions, as the system relies on digital technologies that assist in processing political matters (Hidalgo, 2018). It combines representative democracy with direct democracy and AI agent technology. In hyper democracy, social consensus is achieved with the help of AI agents supporting human political communication.

Augmented democracy is a political participation system that leverages rapidly advancing AI technologies, using digital agents like digital twins, avatars, software agents, and AI agents. Instead of electing and delegating representatives as in liquid democracy, augmented democracy utilizes AI agents that learn the political inclinations of voters. A voter creates an AI agent that shares their political views. They can be aligned with Democratic, Republican, or other political ideologies, and it is possible to combine different political ideologies in appropriate proportions (Hidalgo, 2018). AI agents can also learn about political ideologies such as progressivism and conservatism, and express positions on issues like gender, human rights, climate change, child labor, fair trade, war, and AI ethics.

From a user interface (UI) perspective, the existing system of representative democracy is overly inefficient and inconvenient. To elect representatives, electoral districts are assigned based on geographical and population considerations. While geographical divisions were an inevitable choice to aggregate voter opinions and efficiently convey the public will, this system often results in inefficiencies. In South Korea, for example, a member of the National Assembly represents over 300,000 voters on average. In the augmented democracy system, each voter can have their own representative. There can be as many representatives as there are voters. Each voter’s AI agent can check bills during the legislative process and express approval or disapproval. The AI agents break down thousands of policy packages presented in manifestos during elections and express agreement or disagreement with each individual policy. Voters can also participate directly in the legislative process via their AI agents, and there is no need for representatives to always be human. In fact, AI agents may be more efficient in performing these tasks.

In the augmented democracy system, creating an AI agent is simple. Voters can log into the system and provide their AI agent with information about their political inclinations and relevant data. The AI agent learns from everyday information, shopping habits, reading preferences, political party support, policy interests, and social media activity to understand the voter’s political orientation. However, there are challenges associated with this augmented democracy system. It is necessary to determine how algorithms are trained, ensure data security, and verify that the system operates democratically. Moreover, AI literacy must be improved to enable participation by people of all ages and backgrounds. It is important that the system be easy to use, even for the elderly. Furthermore, if AI takes over the roles humans traditionally play in the political process, it could undermine the essence of humans as political beings. Therefore, AI political parties should delegate rights to AI agents, while voters retain the power to make the final decisions directly. When delegating authority to AI, it is the voter who ultimately holds the decision-making power. Delegation or representation means that the voter remains the principal decision-maker.

In conclusion, the AI political party aims to address the limitations of traditional representative democracy by enhancing political innovation, improving political efficiency, and ensuring the legitimacy of policy decisions — all while keeping voters in the role of sovereign decision-makers. Additionally, the AI political party seeks to elevate the quality of voter participation. Typically, participating in politics or voting in elections requires substantial time and information processing, and the available data for making decisions is often limited. However, if AI can handle information gathering, analysis, and drafting alternatives, voters can participate more efficiently. Discussions about democracy have long focused on reducing the burden of voter participation while increasing the degree to which public opinion is reflected. From participatory democracy to deliberative democracy, liquid democracy, augmented democracy, and hyper democracy, direct democracy elements are being strengthened. At the same time, technological advancements are shifting realistic democratic alternatives toward digital democracy. Therefore, the expansion of an augmented democracy system, where AI agents act as virtual representatives of voters, makes the concept of an AI political party feasible. What follows explains the definition of AI agents, their functions, and the structure and role of an AI political party.

Ⅲ. Operating system for AI political parties

1. Roles and Components of AI agents

While ChatGPT has introduced various possibilities, it has also presented users with certain inconveniences. Efficient use of ChatGPT requires appropriate prompt inputs, which often involve complex and detailed instructions requiring advanced expertise. To address this, the “Custom Instruction” function was introduced to allow users to preset common prompt information. However, ChatGPT is still a passive AI, responding only to user prompts. To overcome the limitations of passive AIs like ChatGPT, AI agent technology has rapidly developed.

AI agents operate based on large language models and gather necessary information from the internet and databases. They can generate program code to produce outcomes. AI agents can also reference their own files to generate documents and collaborate with other AI agents to divide tasks and perform various functions. AI agents create a “task execution plan” to achieve the goals or results assigned by users. They can interact with users about the plan and workflow while incorporating feedback into their tasks. Based on the task plan, AI agents think and make decisions autonomously until they achieve their goals (Ito et al. 2020).

As generative AI technology advances, various forms of AI systems are emerging. With the increasing amount of learning data and improvements addressing the limitations of ChatGPT, more “capable AIs” called Artificial Capable Intelligence (ACI) are being developed. These intelligent AIs go beyond simply assisting workers and play roles as producers and managers. They get involved with the whole process of crafting economic strategies as well as managing election strategies and social infrastructure, including the AI agents. AI agents consist of four main elements which interact to achieve goals: profile, memory, planning, and action (Lei 2023).

  1. Profile refers to the socio-economic attributes of the AI agent, such as age, gender, occupation, and residence, as well as subjective factors like identity, and social or political tendencies. Adding further nuance to their roles, AI agents could be programmed with specific personality traits, such as those defined by the MBTI telling how extravertive or introvertive the voter is and whether the voter’s decisions are mainly based on sensing or intuition. The AI agent builds up the profile by interacting with everyday information the voter produces: social media, search data, movies, media, reading preferences, hobbies, travel, etc. The profile also includes roles and positions within political parties or organizations. Assigning a socio-economic profile to an AI agent helps clarify its role in political representation, policy-making, and legislative processes. Furthermore, AI agents operate within social structures, influenced by broader societal, economic, and political frameworks. These structures shape the perceptions, thinking, and actions of both humans and AI agents, meaning that the output of AI agents is also affected by the societal context in which they operate (Nishimi 2023).

  2. Memory in AI agents stores past interactions, decisions, and relevant data, enabling them to draw on this information when making decisions. Memory in AI agents functions similarly to human memory, storing learned data and text. Just as humans accumulate experiences and knowledge from childhood, AI agents search, store, and utilize data necessary for task execution. Memory is essential for AI agents to overcome the limitations of large language models, like the token limits found in systems such as ChatGPT. For instance, ChatGPT 3.5 can process 4,096 tokens equivalent to around 3,000 Korean characters at a time, while ChatGPT 4.0 can handle 8,192 tokens or approximately 6,000 characters (OpenAI 2023). To work around these data capacity limits, AI agents need a memory function. AI agents also differentiate between short-term and long-term memory. Short-term memory temporarily stores information to contextualize current events, while long-term memory accumulates abstract insights from past experiences. Long-term memory helps improve the quality of the agent’s decisions and outcomes. AI agents recall memories based on recency (the latest memories), importance (weighted memories), and relevance (the most relevant memories to the current situation).

  3. Planning refers to how the AI agent develops strategies and plans based on its assigned tasks, interacting with the user to ensure the plan aligns with the desired outcomes. Effective planning is crucial to ensuring high-quality outcomes.

  4. Action refers to the execution of the planned tasks. It involves the AI agent autonomously making decisions and taking actions to achieve its goals. The action function is essential for task execution, similar to how humans use physical actions to complete tasks.

In augmented democracy, planning in AI agents is necessary to achieve results that align with voter preferences in political processes. This involves creating a roadmap or task plan similar to human thought processes. Additionally, AI agents execute actions based on large language models, gather external information, analyze it, establish policies, draft laws, and communicate policies to the public through media and social media platforms. The planning and actions allow AI agents to actively participate in the political process.

The four components of AI agents interact to achieve objectives (Kimihiro 2023). For example, socio-economic profiles and memory influence priorities. Conservative AI agents might prioritize conservative articles, events, and knowledge. Memory also interacts with planning, and the past experiences of the effectiveness of policies or the success of legislation informs AI agents’ planning processes. Finally, planning and action influence each other with detailed plans leading to more efficient actions. The outcomes of these actions feed back into the agent’s socio-economic profile, resulting in a more personalized AI agent.

2. Structure of the operating system for AI agents

An AI agent is akin to the AI assistant JARVIS from the movie Iron Man. JARVIS informs Iron Man about the condition of the suit and flight, and provides tactical analysis and attack options during battles. It also manages daily schedules and provides advice and information on various situations. AI agents are defined as systems that adapt to and operate within their context to achieve their goals (Franklin, 1996). Unlike ChatGPT, AI agents are plug-in based. Examples include self-driving cars, smart grids that autonomously adjust power supply and demand, smart home systems that regulate temperature and lighting, autonomous vacuum cleaners, and automated stock trading systems.

In AI political parties, AI agents learn from specific databases like local governments, companies, organizations, or personal data. AI agents can identify areas for improvement from the processes and results of their work and accumulate those improvements through learning. In this sense, AI agents research, execute tasks, and propose policy alternatives in a manner similar to humans.

We already use various AI agents in platforms like Netflix, Amazon, Meta, Google, Naver, YouTube, and music streaming sites, which predict content based on user behavior. And a number of systems are already using AI agents. GPT Researcher specializes in creating research reports. Users can input titles, objectives, page lengths, and formats, and the AI agent will generate a table of contents for the report. This table of contents can be adjusted through communication with the AI agent, reflecting the user’s specific requirements. It is also possible to delegate the entire task to the AI agent. Once the AI agent completes the report, it reflects on the improvements identified during the process, allowing it to perform better in future tasks, which is a distinguishing feature of AI agents.

Cognosys is another AI agent platform. Launching in April 2023, Cognosys automates daily tasks for users. It collects internet data for research and generates reports based on user needs. This system aims to solve problems in a structured way by automating tasks. Besides those general functions, AI agents for specialized services, such as analyzing customer information or competitor strategies, have appeared. As AI agent technologies and services evolve, the establishment and operation of AI political parties become feasible. Based on the development of large language models (LLMs), ChatGPT, transformer technologies, and AI agents, the operation of AI political parties seems a natural progression. AI agents can formulate and execute strategies related to policy creation, public opinion formation, and campaign planning. They can also gather information on opposition parties, analyze strategies, and develop response plans. AI agents can work with real-time fact-checking systems to address challenges such as fake news, deep fakes, ideological polarization, and digital gerrymandering during the political process.

In augmented democracy, voters can instruct AI agents to draft political alternatives or align themselves with policies of certain parties. Voters can delegate voting preferences, policy choices, and rights to an AI agent. AI agents can even negotiate with other AI agents or disseminate persuasive information to shape public opinion (Jung et al. 2023). The delegation relationship between individuals and AI agents retains the framework discussed in Liquid Democracy. In this system, individual voters can autonomously decide which issues to participate in directly and which to delegate. However, the role of an AI agent may not differ significantly from that of a human proxy. The key difference is that the AI agent provides the necessary information and alternatives for voters to decide whether to participate directly or delegate. In this way, in a society where AI politics and AI political parties function, most political and party participation could take the form of direct democracy. Ultimately, the final approval for political decisions lies with the human voter, ensuring the preservation of democratic principles. In this structure, AI agents respect human sovereignty by involving voters in final decisions, maintaining the core tenets of democracy.

Ⅳ. The nexus between political reality and AI Political Parties

1. Institutional limits for the adoption of AI parties in Korea

The AI agents can play an efficient and central role in AI political parties. An AI political party differs conceptually from AI democracy or automated democracy in which decisions bypass human involvement. Instead, in AI political parties, individuals and AI agents simultaneously participate in political activities, gather opinions, and encourage political participation. The generative AI technology is used to draft policies and campaign promises, which are later explained, debated, negotiated, and formulated into legislation with the help of AI agents. They aim to expand elements of direct democracy by leveraging AI technology while maintaining the core principles of representative democracy. The organizational structure and operational principles of an AI political party are not significantly different from traditional offline parties, except for the crucial distinction that participation is handled by AI agents delegated by humans. The AI agents act on behalf of the individuals but must seek final approval from the voters at critical decision-making stages. This approval process can be set up at the outset.

In fact, there are already companies where both humans and their AI agents work together as AI political parties. ALT, a company developing personal AI in Tokyo, equips all its employees, including the CEO, with AI agents. In official meetings, employees can send their AI agents instead of attending in person if it’s deemed more efficient, such as when they are traveling. The CEO’s AI agent even handles certain stages of interaction with external visitors. Additionally, the company’s payroll reflects the work of both employees and their AI agents. The CEO’s wage, for instance, increased by about 18% due to their AI agent’s work (Hadfi and Ito 2022). This real-world example suggests that operating an AI-based political party is not far-fetched.

Digital technology is rapidly transforming the landscape of politics. The impact of AI on the 2024 U.S. presidential election makes it clear that changes are needed. The rise of AI political parties may require revisions to various laws and institutions, including the Political Parties Act, Public Official Election Act, Political Funds Act, the role of the Election Commission, the Information and Communications Network Act, and the Criminal Code. What follows is a close look at the laws and regulations that need to be amended in preparation for the emergence of the AI political parties in Korea.

First, the definition of party composition in the Political Parties Act (Article 3) must be updated. Currently, party composition involves a central party and local offices in special cities, metropolitan cities, and provinces. The online presence should become legal as well. Similarly, the conditions for party formation (Article 4) must be amended. These laws were written before the existence of the internet and digital technologies, and they do not reflect the digital era of today. Provisions regarding the number of local offices (Article 17) and party members (Article 18) are also outdated, requiring revision. Admiral Yi Sun-sin, who fought valiantly in the 16th century wars, was with no doubt one of the greatest commanders in the world’s history, but he would not win the battle against killer robots in space. First and foremost, the revision of the laws written before the emergence of the internet, let alone AI, is urgently required.

The normalization of AI political parties requires comprehensive amendments to the entire Political Parties Act, including the chapters on party formation (Chapter 2), party mergers (Chapter 3), party membership (Chapter 4), and restrictions on certain individuals from becoming party members (Article 22). These chapters need to be abolished. If individual AI agents can participate in AI political parties, the relationship between voters and their AI agents needs to be fundamentally redefined. This also ties into a broader discussion on the extent to which AI agents should be allowed to participate in politics. Similarly, regarding party operations, changes are necessary for party dues (Article 31), party committees (Article 37-3), policy discussions (Article 39), party dissolution (Chapter 7), and penalties (Chapter 8). These provisions implicitly assume that political organizations exist in physical, geographical spaces and are composed solely of humans. However, AI political parties operate by delegating participation and decision-making to AI agents. While party decisions can be made either through delegation or direct participation, much of the routine party activities will be conducted by AI agents.

Changes to the Public Official Election Act will be required as well. Currently, Article 3 defines an elector as “a person eligible to vote, or someone listed on the overseas voter roll.” However, AI agents could be authorized to participate in elections and decision-making on behalf of voters through delegation. Provisions like Article 6-3, which guarantees voting rights to infectious disease patients (amended in 2022), may become obsolete, or AI agents could take over voting responsibilities. The revisions are needed to generate AI engaged to ensure a fair election in an effective and efficient manner. The duty of neutrality for public officials (Article 9) is subject to debate in the evolving digital landscape and technological advancements. With the revision of regulations, AI agents could play a major role, making the Fair Election Support Group (Article 10) more efficient. Tasks like election broadcasting review (Article 8-2), election reporting review (Article 8-3), right to rebut election reports (Article 8-4), internet election reporting review (Article 8-5), and election polling review (Article 8-8) could be more efficiently handled by election management AI agents.

The emergence of AI political parties would require fresh definitions of electors (Article 15) and eligibility for candidacy (Article 16). With the rapid advancement in brain research, technologies supporting communication for people with brain disorders are emerging, and candidates with serious conditions like ALS and cerebral palsy attended parliamentary sessions in the House of Councillors election in Japan in 2019. In 2024, Neuralink, a brain implant start-up led by Elon Musk, successfully implanted a computer chip into a human brain. The implanted chip serves as an interface that connects the human brain to a computer, analyzing brain waves that occur when a person thinks or performs an action. Such technology allows individuals with brain disorders to communicate their thoughts to others.

These advancements in technology and digital tools will have the effect of expanding political participation. Additionally, candidacy (Chapter 6), which is limited to human candidates, should include “delegated AI agents” to run for office. Election expenses (Chapter 8) would see the most significant changes. Campaigning and election management costs for AI agent candidates, AI agent-led political parties, voter AI agents, civic group AI agents, and election management organizations could be greatly reduced. The current methods for candidate nominations, registration, and deposits may need to be abolished or transitioned to public funding as the inclusion of AI agents and AI parties would decrease election costs. While polling stations may continue to exist, their usage would decline. Overseas elections might still exist as a system but would become less distinct from domestic elections. The cost of voting could be drastically reduced as well, and the expenses incurred in overseas elections, domicile and onboard voting, would be significantly slashed, thanks to the creation, viewing, and objection processes for the voter registry automatically managed by AI agents in the digital space. Furthermore, if AI agents participate in elections through delegation, the vote-counting process could be completed much faster.

The rules for election campaigns (Chapter 7) would also need to be updated to allow AI agent participation. Campaign posters and manifestos could primarily be distributed digitally. Changes would be unavoidable in how newspaper ads, broadcast ads, online ads, public speeches, and debates are conducted. The crackdown on illegal election campaigns would also need to shift from human-driven enforcement to digital and automated systems. Additionally, the eligible age for participating in voting would be another issue requiring collective thoughts and discussion, as those who are ineligible to vote (Article 18) or run for office (Article 19) could exercise their sovereignty through delegation.

In conclusion, automation and digital transformation (DX) of election management tasks would be necessary, demanding innovation in election management organizations. Moving from paper-based administrative tasks and face-to-face interactions with stakeholders to a fully digitized, automated system would be crucial. Innovating election management would inherently lead to organizational changes within election management bodies.

2. AI Political parties: possibilities and risks

A wide acceptance of AI political parties would address the political and institutional deficiencies in traditional representative democracy. The most promising improvement would be the expanded inclusiveness. AI political parties can work with AI agents to deliver minority issues overlooked by traditional representative democracy and to support minority and socially disadvantaged groups by expanding their opportunities for decision-making and political engagement. AI agents can be employed to bring minority issues to the forefront. This possibility echoes the theory of change which highlights the political impact of IT technology and the internet in particular (Lilleker 2014). Information asymmetry in democracy can pose a political threat, but with AI parties, a specialized party in a particular issue is possible. In fact, in Japan and Denmark, where AI political parties already operate, the focus has been placed on advocating agendas favoring marginalized groups. In preparation for Denmark’s general election in 2023, the AI political party collected and analyzed policies from minority parties since the 1970s and proposed policies that prioritize marginalized communities. The Japanese AI party is prioritizing budgets benefiting disadvantaged groups. Furthermore, AI political parties can encourage marginalized groups to proactively participate in the policy-making process as well.

AI political parties can induce more participation of the marginalized by lowering the financial barrier of political engagement. Korean politics has been criticized for being high-cost and low-efficiency. Running for local councils or the National Assembly requires high campaign costs, but AI political parties can operate at a fraction of that cost, making IT literacy the most critical factor. Crowdfunding can efficiently organize and mobilize supporters without large financial investments, and anyone can create an AI political party with minimal costs, providing greater opportunities for underrepresented groups and contributing to the development of democracy. In a country like South Korea, where internet technology is highly advanced and social and economic activities take place online, AI political parties could offer a useful means for marginalized groups. One of the groups at the political margin is the youth. AI political parties and AI agents could also serve as an efficient method for political participation for the youth, who are highly engaged in the digital world.

However, it is not all rosy with the AI parties. There are a number of critical issues we need to address for the adoption of AI political parties. First, decisions and policies can be made based on bias caused by confusion between what is considered objective and what is fair. Objectivity and fairness are different values, but bias can be created by AI linked to the naturalistic fallacy, a problem of confusing facts with norms. For example, male dominance in society is a social phenomenon, but it may be wrongly perceived as natural. Similarly, while AI bias is a man-made and socially structured issue, it may be seen as natural or inevitable. AI bias is a societal issue that must be addressed through institutional reforms and adjustments in learning processes. In fact, Open AI makes deliberate efforts to correct errors that arise during the development of ChatGPT. However, changes in societal institutions and perceptions are also required to fully address social biases.

Second, fake news and deep fakes. AI political parties and other political groups could create AI agents to spread slander and negative campaigns online. Although fact-checking could mitigate fake news and deep fakes, the gap between technological advancements and legislative or administrative responses might allow AI agents to be corrupted by distorted information. In a society like South Korea, where divisions between political camps are intense, this issue could become even more problematic because the political landscape of the online space mirrors and easily exacerbates divisions in the real world.

Third, a potential for political advantage or disadvantage needs to be watched carefully. President Obama achieved a landslide victory through micro-targeted campaigns in the US presidential election in 2008. However, in the 2016 election, Donald Trump leveraged AI technology to win the presidency. During Trump’s election, as well as the Brexit decision in the UK, prediction models based on vast amounts of data played a key role. Developing and utilizing advanced AI requires massive financial resources. As we enter the AI era, there is a growing trend where financial power, rather than a desire for innovation, determines technological dominance. The development of ChatGPT by OpenAI was largely driven by significant investments from Microsoft. The current AI competition between the U.S. and China, characterized by a hostile rivalry, is dominated by these two countries due to their superior manpower and financial capabilities. The capacity for financial mobilization, which decides a victor in the era of AI politics, can be an advantage to a particular political party in Korea.

With the emergence of the internet, progressives were more active in utilizing the new technology, as was observed in the election of President Roh Moo-hyun, the world’s first “internet president.” However, financial mobilization capabilities are generally considered advantageous to conservative circles in Korea. Additionally, the conservative circles can also be more advantageous due to the data produced by their supporters, a large share of which is the elderly. The potential for advantage or disadvantage in AI-related technology stems from social data. As the country transitions into a hyper-aged society, the proportion of elderly individuals continues to rise. Despite advancements in internet technology and the normalization of communication via social media, more conservative information may become prevalent due to the increased online presence of the older generation. The political culture shaped by authoritarian periods, government-led media control, and the dominance of conservative media also significantly influence the data available in society. The data utilized by AI political parties are based on what South Korean society produces, potentially giving the conservative camp an edge in terms of financial resources and learning data. In this sense, it would not be utterly impossible for such AI agents to support a martial law declared in the 21st century in Korea. Yet, the early bird effect offers a counter argument against such a scenario. As Everett M. Rogers explained in his 1962 book titled Diffusion of Innovations, the role of “innovators” and “early adopters” remains highly relevant, aligning closely with the spirit of innovation and creativity. The early adopter effect could also serve as a valuable tool for encouraging political participation and the reflection of youth opinions in South Korean politics.

In conclusion, the emergence of AI political parties marks not only an evolution in democracy but also a fundamental shift in the role of humans as political entities. Simultaneously, the opportunities for discussing political issues related to social welfare, human rights, gender, and redistribution will expand, benefiting marginalized groups such as youth, the elderly, migrants, and people with disabilities. Ultimately, AI political parties can strengthen the procedural legitimacy of democracy and enhance equality of opportunity. The evolution of AI technology, through AI political parties, has the potential to contribute to the ongoing evolution of democracy, but the potential comes with a degree of concerns we need to work on.

Ⅴ. Conclusion: doing politics with AI for a better democracy

Existing theories of modern democracy struggle to explain the political changes brought by AI, and discussions on new political realities such as the emergence of AI political parties are limited. Democracy as we know it was conceptualized in an era before the internet, let alone AI, which means there is a need to redefine and redesign democracy to incorporate this new technology that is impacting every part of our daily lives. The introduction of AI political parties raises important questions about how to balance the ultra-efficiency and rationality of algorithms with respect for human agency as political subjects and how to ensure political and social necessity. The balance of these factors will differ in every country.

In AI political parties, the role of humans will fundamentally change. In traditional representative democracy, humans are responsible for gathering and analyzing information, assessing party policies, and making decisions. However, in AI political parties, humans will primarily evaluate AI agents, make decisions as principal actors, and exercise their rights. Humans will also assume responsibility for the political outcomes generated by the AI agents to which they delegate tasks. In other words, while judgment may be delegated to AI agents, humans remain accountable for the results. Instead of providing a definitive argument, the article explores the potential of the AI political parties.

Although this article ponders on the legal revisions in preparation for the AI political parties, in reality, the adoption may take considerable time, influenced by political advantages and disadvantages, much like the adoption of internet voting. Additionally, challenges such as AI bias, AI ethics, security concerns, and the verification of AI agents’ identities must be addressed. However, the innovation that AI political parties can bring to political processes, such as enhanced efficiency, increased representation, and broader participation, makes them indispensable for the progress of democracy. Initially, parties could test the concept by integrating AI agents into the youth committee or candidate primary processes. AI agents could be used to facilitate discussions on specific issues between parties, supporters, and voters. Pilot programs in private organizations or party leadership elections could help ensure stability before broader implementation. With AI technology advancing rapidly, politics should not hinder its development.

VI. About the Author: Seon Gyu Go (고선규)

Born in Yeongwol County, Gangwon Province, Seon Gyu Go completed his undergraduate degree in Political Science and Diplomacy at Dankook University, Korea, before pursuing advanced study in Japan. After research at the University of Tsukuba, he earned an M.A. and Ph.D. in Information Science from Tohoku University, focusing on the social-scientific study of AI, robotics, IT, and social media (SNS).

Following his doctorate, he held positions at Seoul e-Government Research Center, Sejong Institute’s Japan Research Center, and served as a full-time professor at the Election Training Institute of the National Election Commission. Since 2018, he has been involved in IT and AI-related matching projects between Korea and Japan. Currently, he is a Research Fellow at Waseda University’s Institute for System Competitiveness and a lead Research Fellow at the Global Research Network (GRN) in Korea.

His first-hand experience of the 2011 Tōhoku earthquake and the subsequent Fukushima nuclear accident prompted him to focus on the decommissioning of nuclear power plants using robots. This sparked further research into expanding the utility of AI and robotics for broader societal problem-solving. His current projects explore how AI can address challenges in a super-aged society and in the socio-economic transformations wrought by the Fourth Industrial Revolution.

With over 50 published academic papers (e.g. AI·Robot은 인간과 공생 가능한 천사인가), he is also the author of more than ten Japanese-language books, including ネット選挙が変える政治と社会 (Internet Elections and the Transformation of Politics and Society, Keio University Press). In addition, he runs the Facebook group Human+AI Coevolution Platform. His scholarship intersects with Syntheticism’s vision of a “techno-social sculpture,” particularly in exploring AI-driven participatory models of governance.

Contact: tohokugosg@gmail.com

Other Political AI Literature by Seon Gyu Go

  1. “AI Policy Making and Solving Local Issues in Japan” – Civic Politics Research (Konkuk Univ. Institute for Civic Politics Research), Issue 7: 3–26 (Dec 2023), by Seon Gyu Go and Yoko Sakurada
    GRAFIATI.COM
    A scholarly study on how AI and robotics can assist policy decisions in local governments. It examines case studies in Nagano and Hyogo Prefectures (using AI to generate thousands of policy scenarios for sustainable communities) and in Fukushima (using AI robots in education).
    (Published in Korean with English abstract; accessible via DOI: 10.54968/civicpol.2023.7.3)

  2. **「<세종정책브리프> AI는 한국의 통일문제 현안들을 해결하는 대안일 수 있는가?」 (고선규, 2020) – SPN Seoul Pyongyang News column, Feb 17, 2020** [SPNEWS.CO.KR](https://www.spnews.co.kr/news/articleView.html?idxno=26141) In this Korean policy brief, Go explores whether AI could be a solution to Korean reunification issues. He discusses the dawn of “AI politics,” citing the 2018 Tama City mayoral election in Japan where an AI candidate (the AI Mayor project led by Michihito Matsuda) ran for office, noting that an AI lacks self-interest and organizational ties, enabling more neutral policymaking. He also examines implications of AI-driven policy proposals in Japan and the advent of AI weapons, calling for careful governance of AI in political and military spheres. (Full text in Korean available on SPN Seoul & Pyongyang News.)

  3. 「인공지능과 어떻게 공존할 것인가: 인간+AI를 위한 새로운 플랫폼을 생각한다」 (고선규, 2019)
    How to Coexist with Artificial Intelligence: Thinking of a New Platform for Human+AI, 타커스(Takus) Publishing, 2019
    Google Books | Kyobo eBook Preview (English translated)
    This Korean-language book provides context for Go’s perspective on AI, including AI’s growing presence “from the robots we meet in daily life to AI politicians, AI artists, and AI doctors.” It considers how humans can adapt to and collaborate with AI entities and suggests that improving “AI literacy” will be crucial for the future—an emphasis consistent with Go’s academic focus on AI’s impact on society and governance. Additionally, the book presents Go’s vision for rethinking social and institutional structures to integrate AI in ways that advance democratic engagement.