Democratic AI Principles
The Center for Democratic AI is focused on increasing public engagement in AI policy and promoting democratic values in AI development.
Democracy is not simply one form of government. Democracy is a living, breathing system that relies on a delicate balance of power. This balance of power exists between different socioeconomic classes, between government officials and their constituents, between employees and employers, buyers and sellers, information producers and consumers, the individual and the collective, and between rights and concessions.
Democracy relies on balance, reciprocity, equality, and transparency. AI threatens to disrupt this balance in multiple ways. AI can concentrate wealth and power into the hands of a few powerful technology developers. AI can obscure information, prioritizing one type of information while ignoring all others. AI built without transparency and accountability can bring harm, harms that go unaddressed. AI can disrupt the workforce and diminish paths to financial stability for millions of people. The algorithms and data collection mechanisms that power AI can erode autonomy, privacy, and self-determination by surveilling people and classifying them into different groups. The environmental impacts of the AI industry can pollute the healthful environment we know and love.
The Center for Democratic AI aims to support and bolster the principles that uphold democracy, and ensure that AI development not only abides by these principles, but enhances them. The principles below are guidelines, a roadmap, to help ensure that AI is created safely and justly. These principles are published so that the public is informed on what democratic AI can and should look like, and so that policymakers can have a guideline of what their policies should aim to accomplish.
Key Takeaways:
We need an independent, nonpartisan oversight board to whom AI developers must report. This board must represent public interests and have the power to enforce AI safety and responsibility standards.
Information sharing is essential. Developers must publicly disclose their models' capabilities, weaknesses, and accumulated insights, so knowledge isn't hoarded by companies alone.
AI can generate enormous wealth, wealth that should not remain concentrated. The wealth created by AI must be distributed broadly across society through taxation, dividends, or public funds.
Autonomy and Self Determination
AI models are built on algorithms that possess and process vast amounts of information. This information goes beyond just describing human behavior, it can begin to alter it. Profiles built on individual data violate privacy and can be used to understand and reshape human thinking and behavior. Autonomy is the foundation of democracy, for self-government requires citizens who form beliefs and choices through their own reasoning. When decisions are quietly steered by systems optimized for someone else's benefit, people stop acting as free participants and become instruments of external goals. Because a population nudged rather than persuaded is not self-governing, protecting autonomy is a necessary part of protecting democracy.
What this looks like applied to AI:
AI should assist in human decision-making as opposed to modifying what users think or want. Behavioral data should not be used to maximize clicks, purchases, or votes.
Algorithmic influence should be visible and reversible. People cannot exercise self-determination if they are unaware of behavioral nudging. Personalized systems must disclose when content or pricing has been tailored, and must give users the option to not participate in these systems.
Policies should consider enforcing data collection as an opt-in, rather than opt-out system. The default mode of data collection should be “off” until a user toggles it “on”.
Financial Opportunity and Equity
Democracies rely on equality in many forms, particularly when it comes to equal representation in government and equal financial power within the economy. Philosophers from Aristotle to Alexis de Tocqueville have long stressed that wealth equality underpins stable, functional societies. Firstly, when wealth pools too narrowly, it inevitably converts into unequal mismatched political influence. Secondly, a citizenry with no meaningful economic security cannot participate in democratic life; people become dependent, disillusioned, or disengaged. Financial equity is therefore not a separate concern from democracy, it is one of its load-bearing conditions.
What this looks like in AI:
We must put in place protections for workers made redundant due to AI. As AI automates tasks and entire roles, displaced workers need structured support so technological progress doesn't translate directly into personal financial ruin.
Universal basic income should be considered in AI policy. As AI increasingly decouples productivity from human labor, a guaranteed income floor ensures that people retain economic security and consumer power even as traditional wage-based livelihoods shift or disappear.
Implementing a tax scheme that taxes AI systems that replace human labor. When AI performs work once done by paid employees, taxing that displaced labor value ensures AI's productivity benefits society broadly, not just the companies deploying it.
Representation
Representation is a core tenet of democracy, and it is equally crucial when it comes to AI. This extends beyond fair representation in algorithms and datasets to reduce biased outputs, it demands representation of the public in the actual creation of AI itself. AI should not be built solely on the will of a small group of technologists, implemented wherever and however they choose. It should be built to serve what the public wants and needs, and its creators should abide by the standards and regulations the public sets. The priorities, interests, and safeguards of ordinary people, including protections against job loss, must be embedded in AI's development. Democracy cannot function if the technology reshaping every aspect of society is designed and governed by a small group of people, and is not accountable to the larger majority.
What this looks like in AI:
The creation of an independent oversight board that is made up of nonpartisan public representatives.
This independent board must have binding enforcement authority. An oversight board with no power to halt harmful deployments, mandate changes, or penalize violations is only symbolic.
Transparent reporting from AI companies to the public. AI companies must regularly disclose their developments, AI assessments, and decisions in accessible terms, so citizens can participate readily in AI development.
Accountability and Reciprocity
Representation is step one of democracy, but representation without accountability and reciprocity is meaningless. It does nothing for citizens to have their voices heard if that input generates no response, no consequence, and no change. Democracy is fundamentally a two-way relationship: citizens grant legitimacy to institutions, and in exchange, those institutions must answer to the people they affect. When companies building and deploying AI can hear public concerns and simply proceed unchanged, representation becomes a hollow ritual rather than a functioning democratic check. This is where CDAI shines: we work to build genuine two-way engagement between the public, companies, and policymakers, ensuring input translates into actual response.
What this looks like in AI:
The creation of an independent oversight board that is made up of nonpartisan public representatives. This board would assess AI companies' behavior, require regular reporting, and enforce fines or halts of operations when companies fail to meet public standards. Accountability requires consequences.
Antitrust laws must be enforced. Companies evade accountability far more easily when they hold monopoly power or represent the only viable option available to consumers. Robust antitrust protections create a competitive AI landscape, a market where companies must remain responsive to the public, because users and regulators alike have genuine alternatives and leverage.
Publication of AI liability frameworks. People need a clear, reliable legal pathway to hold AI systems and their creators accountable when harm occurs. Without defined liability, companies can deflect responsibility indefinitely, leaving affected individuals with no recourse and no reciprocity.
Transparency and Information
Advancements in technology have historically expanded democratic values by democratizing access to information. Both the printing press and the internet broke down barriers between knowledge and the public, empowering people to make informed decisions and exert real control over their own lives. AI holds the same potential to make specialized knowledge accessible to far more people than ever before. However, it carries an equal and opposite risk: AI can tailor information to individual users in ways that fragment shared understanding, and it can generate and spread misinformation at unprecedented scale and speed. For AI to strengthen rather than undermine democracy, information must be verified and shared openly, and the systems producing that information must themselves be understood by the public they serve.
What this looks like in AI:
Regular transparency and safety testing must be mandatory. AI developers must be transparent about what their systems can and cannot do. This requires rigorous, independent safety and impact testing before and after deployment, with results made publicly available.
Public institutions must have access to the data and insights garnered by technology companies. Companies must share the data, methodologies, and insights behind their AI systems, allowing researchers, journalists, and citizens to independently verify claims, identify risks, and understand how these systems actually function.
Safeguards must be set in place to prevent the spread of misinformation. Because AI can tailor content to individual users, systems must be designed and audited to prevent the spread of harmful, individually-targeted misinformation, ensuring people receive accurate information rather than a version of reality engineered for engagement or influence.
Equality
Equality is the thread running through every principle stated above. Democracy cannot survive concentrated power or concentrated information. Rather, true equality demands both equal decision-making power and equal financial standing. This matters urgently with AI, because these systems are built on the contributions of the public: our work, our data, our CAPTCHA solutions, our books, our photographs. If AI is constructed from collective human contribution, its benefits cannot justly belong to a select few. The public has a direct claim on the value the it has already created.
What this looks like in AI:
Member of the public must have equal access to data and information. Insights and analytics derived from collected data should not remain locked in corporate holdings. They must be shared publicly so that ordinary people benefit from the same knowledge and analysis companies use internally. This would prevent a growing knowledge gap where corporations understand the world, and the public, far better than the public understands itself.
AI-generated wealth should be distributed. The global AI market, valued at over $189 billion in 2023, is projected to reach $4.8 trillion by 2033. Yet despite being built on public data, this wealth remains concentrated among a handful of companies. Distribution mechanisms, such as the targeted taxation of AI-driven profits, sovereign wealth funds, or direct dividends paid to consumers for their data, would ensure the public shares in wealth it helped create.
Decision-making power should not be concentrated only amongst corporate leaders. No single company or developer should hold unilateral authority over systems that affect the entire public. Decision-making about AI's rules, deployment, and consequences must be distributed across regulators, workers, and citizens.