Governing Artificial Intelligence: Public Preferences and Regulatory Options

Global Survey Reveals Citizens Prioritize AI Safety, Public Oversight, and International Regulation

Researchers surveyed over 14,000 people across seven countries to see how they want artificial intelligence to be controlled. Most people want rules that focus on safety rather than just innovation. They also prefer governments and international groups to make the rules instead of private companies.

As AI transforms economies and societies, policymakers are struggling to decide on the best way to govern it. Current global strategies are deeply divided. The European Union emphasizes precaution and public authority. The United States leans toward market-led innovation and national flexibility. China pursues state-led development. While these models dominate the geopolitical conversation, we have lacked data on whether these choices align with what the public wants.

This study identifies a significant disconnect. Instead of a fragmented public, the authors find a consistent global preference for safety, public oversight, and international cooperation. These priorities often contradict the prevailing regulatory models of the world's leading AI powers.

The misalignment of global AI policy

Current AI governance faces a "triple misalignment." There is a horizontal misalignment, as major powers pursue different models. There is a vertical misalignment, where dominant approaches diverge from public demand. Finally, a temporal misalignment exists. Short-term incentives for rapid innovation clash with long-term public concerns regarding risk.

Existing studies have mostly focused on general attitudes toward AI risks. They lack insight into specific regulatory design. This leaves a gap in our understanding of how citizens weigh actual trade-offs. For example, should a rule prioritize a company's ability to innovate or a person's right to safety? As seen in [Figure 1a], while support for regulation is high globally, the specific "how" of that regulation remains a contested territory.

Mapping preferences through conjoint analysis

To move beyond vague opinions, the authors employed a conjoint survey experiment. In a conjoint experiment, respondents evaluate pairs of hypothetical proposals. These proposals vary across three specific dimensions:

1. Policy Objective: Whether the goal is to strengthen technological innovation or ensure public safety.
2. Governance Mode: Whether the rules are set by government authorities or managed through private self-regulation (rules created by companies themselves).
3. Level of Authority: Whether the regulation happens at a national level or through international collaboration.

The researchers applied this framework across three distinct domains—the workplace, policing, and warfare. They wanted to see if preferences shifted when the stakes changed. This tested whether people want a "horizontal" approach (one set of rules for everything) or a "sectoral" approach (tailored rules for different industries). As illustrated in [Figure 2a], this method allowed the authors to calculate "marginal means." These are the average probabilities that a specific feature, like "safety," would be chosen across all possible combinations of the other features.

Consistent signals across borders and domains

The results reveal a striking level of consensus. The authors report that citizens in all seven studied countries strongly support AI regulation. The pooled mean support score was 5.28 on a 7-point scale [Figure 1a].

The preference for safety over innovation is the most dominant trend. The paper reports a 12-percentage-point gap in choice probability favoring safety over innovation [Figure 2a]. This gap means safety-oriented proposals are significantly more likely to be chosen than innovation-focused ones. This preference is remarkably stable. The same hierarchy of priorities holds true whether the AI is discussed in the context of office automation, police surveillance, or battlefield weaponry .

However, the strength of these preferences is not uniform. The preference for safety is most intense among individuals who perceive AI as inherently risky, unpredictable, or personally consequential [Figure 5a-c]. Furthermore, while the preference for international governance is a global trend, it is notably weaker in the two dominant AI powers, the US and China .

In these nations, respondents showed less enthusiasm for moving authority to the international level.

Limits of the experimental design

The study has inherent constraints. Because the researchers used a conjoint design, they had to simplify the regulatory landscape. They reduced it to only three dimensions to keep the survey easy to understand. In reality, AI governance involves more complex variables. These include liability frameworks and specific technical standards.

Additionally, the study represents a single point in time. The authors note that public attitudes are tied to perceived risk and personal experience. Therefore, these preferences are likely to be dynamic. As AI becomes more integrated into daily life, the "unpredictability" and "affectedness" (how much a person feels impacted) of the technology will change. This could shift regulatory demands. Finally, the seven-country sample cannot serve as a perfect proxy for the entire global population.

A verdict on regulatory legitimacy

Is the current trajectory of AI governance sustainable? The study highlights a systemic gap. The "innovation-first" and "self-regulation" models favored by major players like the US diverge from the "safety-first" and "public-oversight" models demanded by the global public.

If governments wish to build regulatory frameworks that command long-term trust, they cannot simply optimize for speed. The data suggests that durable legitimacy will require a pivot toward safety-oriented objectives. It will also require strengthening public oversight mechanisms. For those building the technology, the takeaway is clear. The social license to operate AI will likely depend on how well the industry accommodates the public's demand for safety and oversight.

Figures from the paper