Last month, China’s internet regulator released proposals to regulate generative AI systems such as chatbots. The proposal includes a number of clauses on AI’s wish list that support democratic values, from anti-discrimination to legal liability for developers. But the regulation is clearly aimed at strengthening China’s authoritarian regime. By examining the contradictions inherent in this proposal, developers, analysts, and governments can draw lessons for building a truly democratic framework for AI.
Protect individuals while strengthening state control
Certain parts of the draft regulation, if enforced in a uniform and meaningful way, would make real progress in protecting millions from potential harm from AI. Privacy requirements prohibit profiling based on user activity and retention of data that could lead to actions such as re-identification. Transparency requirements make it easier to identify potential or actual infringements caused by AI systems. Liability and remedies allow individuals to notify companies when content is generated that violates their likeness or privacy. All these requirements are reminders of principles often touted as supporting democratic values.
At the same time, the draft demonstrates a clear intention to strengthen government control over China’s technology and information ecosystem. Under this rule, generated content will be required to “reflect core socialist values.” Content that contributes to “subversion of state power” is prohibited. The draft’s vague language gives regulators greater leverage to impose their will on tech companies. The requirements are focused on private sector entities that develop and deploy generative AI systems. There is no mention of the use of AI by the government.
The draft regulation makes strong positions on two of the most contentious issues in the AI governance debate. A strong advocate of protecting society (and governments) from the risks posed by AI systems and technology companies, rather than applying a laissez-faire governance approach that gives the private sector greater latitude in developing new AI products. doing. And when it comes to the question of whether the developer or the government should take the lead in defining AI rules, this regulation is really about transferring power to the government.
Democratic values transcend high-level principles
On paper, there is actually little international disagreement about the high-level AI governance principles set out in China’s regulatory drafts. Indeed, many of the measures proposed in the regulation are often marketed as affirming democratic values such as transparency, respect for individual rights, fairness and privacy, according to a consensus international AI policy document. It corresponds neatly to the AI principles described in .
Of course, the devil is in the details. At just 2,300 characters, the draft regulation leaves little room for precise guidance on vague phrases like “false information” and “damage to.” . . ‘Physical and mental health’ actually means. Engaging largely at the level of broad principles leaves considerable leeway in interpretation and enforcement.
China’s draft regulations outline how high-level policy guidance and principles obscure the implementation decisions and trade-offs that determine whether AI is developed and used in ways that affirm democratic values. indicates whether or not For example, the US AI Bill of Rights blueprint contains similar high-level principles describing transparency, remediation, and fairness. But US government agencies enforcing these principles through different governance structures and oversight mechanisms are likely to produce different results than China’s enforcement agencies. Similarly, the first discussion draft of the Council of Europe’s proposed treaty on artificial intelligence, human rights, democracy and the rule of law contains few high-level principles that are themselves controversial. However, the Convention’s proposed follow-up mechanism would allow signatories to criticize each other when specific operational procedures do not faithfully reflect the values they were intended to codify, and the Convention would We may be able to do a better job of reinforcing democratic values.
China’s approach to AI geopolitical competition
Many in the United States argue that AI is so essential to economic and military power that efforts to curb its development or rebuild it would be a major strategic failure. However, China’s recently proposed regulations suggest it does not share this view.
Two of the draft’s most notable features are its lack of strength and nuance in many provisions. Articles on bias prohibit providers from “engaging.”[ing] Generating discriminatory content based on a user’s race, nationality, gender, etc. This is an impossibly high technical demand, even if the developers understood how the word “discriminatory” should be interpreted. Another article requires that generated content be “true and accurate.” It lacks precise definitions of vague, self-explanatory concepts such as “discriminatory” and “true and accurate,” with obvious political and cultural overtones.
In defining hard bans on certain types of content, the proposed requirements are at the very heart of the modern machine learning paradigms that power systems like ChatGPT. Under this paradigm, the model learns to replicate statistical patterns in very large datasets, so engineers can identify sexist, racist, or other There is little hope of guaranteeing that the nasty patterns of are not learned.Technological advances that allow engineers to ensure that generative AI systems work reliably I never have Undesirable results may be years away, even if they do materialize.
This will force China to make a difficult trade-off between accelerating the adoption of generative AI systems and maintaining a robust technology and information ecosystem. A literal interpretation of the proposal’s stringent requirements would almost prohibit the introduction of technology that could create significant economic value if boosters were correct. However, by only allowing developers to comply loosely with regulatory mandates, governments are not subject to regulatory sanctions for content that could harm individuals or undermine the government’s control over the information environment. will slip through.
Some of the regulations suggest a more flexible approach to regulation enforcement than most provisions suggest. For example, in one article, he gives developers three months to improve their rule-breaking content. But China’s publication of these rules, even in draft form, underscores China’s skepticism of the theory that broader geopolitical imperatives call for unrestricted development of generative AI systems. showing.
Implementing democratic values in AI
What lessons should a nation that truly wants to promote democratic values take from this proposal?
First, the draft regulation shows that many of the consensus AI principles often touted as affirming democratic values can be just as easily used to protect authoritarian states. Ensuring AI systems meet broader democratic goals requires not only technical proposals but also strong organizational and oversight mechanisms.
Lawmakers should take transparency measures to help civil society groups and the public understand how regulators are exercising their enforcement discretion. Regulators and accountability bodies need additional resources to double down on their mission to protect consumers, individuals and communities. To build public trust in institutions and the AI itself, applications must demonstrate that they follow high rhetorical principles such as “fair”, “”. “Accountability” and “Human Centricity”. A follow-up mechanism for international agreements could provide a forum for democratic countries to criticize each other’s practices, similar to the treaty bodies of human rights instruments. Governments and businesses need to create broad and inclusive processes for gathering input from diverse publics about what they want from new technologies.
Second, many of the most important uses of AI systems are within government, from providing critical services to law enforcement and surveillance. Regulations that apply only to the private sector, such as China’s, completely ignore some of the most powerful potential negative impacts on liberal and egalitarian democratic ideals. Recent U.S. Efforts to Increase Government Transparency and Oversight of AI Use (Currently Poorly Implemented)) Measures applicable to high-risk government use of AI (such as the EU’s proposed AI law)) This is a welcome development in this regard. However, debates about how AI rules should apply to law enforcement (e.g. blueprints for the US AI Bill of Rights and proposed EU AI law) will clearly address these issues. It shows that there is very little
Finally, the draft regulation points to a potential flaw in the geopolitical competition narrative over AI. The introduction of AI systems that erode individual privacy, perpetuate discrimination, and widen inequality may only undermine public trust in industry and government, rather than enhance America’s technological and strategic advantage. Promoting AI systems that spread hate speech and misinformation will undermine a democratic society’s ability to reason, respectfully and deliberately.
Domestic action is needed to shape AI developments so that they do not undermine these fundamental pillars of long-term strategic competitiveness. For example, taking steps to ensure that the opacity and technical complexity of AI do not undermine the effectiveness of existing laws, such as liability and discrimination, could reduce the externalization of AI harm. be. Funding for sectoral regulators to address the impact of algorithms in their areas of responsibility could increase governments’ ability to develop flexible and effective regulation when needed. A combination of government R&D funding and regulatory action has the potential to encourage the kind of AI research that benefits society at large.
Success in this broad AI strategic competition concept requires close attention to implementation details, government oversight of algorithmic use, and intelligent domestic regulation. Democracies should avoid participating in a short-term race to advance the technological development of fragile and unregulated AI systems that will only lead to Pyrrhic victory.
