The first independent scientific expert panel on artificial intelligence established by the United Nations released a preliminary assessment report stating that the rapid development of artificial intelligence (AI) has brought huge potential benefits to countries and people around the world, but it also comes with major risks. This report will be submitted to governments at the first United Nations Global Artificial Intelligence Governance Dialogue in Geneva from July 6 to 7. It is the first independent scientific assessment of artificial intelligence on a global scale. A complete and comprehensive version is expected to be released next year.
The report was co-authored by 40 leading scientists and experts from all regions of the world, who participated in an individual capacity for a three-year term and were not directly bound by any government, agency or company, and aimed to provide independent scientific evidence to decision-makers. The report points out that in the context of the rapid evolution of artificial intelligence capabilities, policymakers urgently need reliable scientific basis to formulate regulatory frameworks. However, the current understanding of some highly autonomous systems by the scientific community is still lagging behind, and the adaptability of society and government is obviously insufficient. Effective tools for restraining this type of highly autonomous artificial intelligence are still limited.
Expert panel co-chair Yoshua Bengio stressed in the report that scientific evidence of "deceptive behavior" in artificial intelligence is growing. He pointed out that as system capabilities further improve, scientific research cannot guarantee that artificial intelligence will not bring catastrophic harm—such harm may originate from the behavior of the system itself or from users with malicious motives. The report concludes that "the potential benefits of artificial intelligence are huge," but warns that the negative effects of the technology cannot be ignored if there is a lack of adequate constraints and careful planning during large-scale rapid deployment.
The report points out that the rapid promotion of artificial intelligence is profoundly changing social, economic and environmental systems, and also poses challenges to users' mental health. On the positive side, artificial intelligence applications have demonstrated the ability to improve efficiency and create new value in many fields such as education, medical care, scientific research, and industrial production. But on the negative side, the report warns that AI could be used as a destructive tool or to undermine public trust, social cohesion and democratic deliberation (deliberative discussion) by manipulating information and emotions.
According to this preliminary assessment, more than one billion people around the world use conversational AI every week, but the level of application is highly uneven across countries and industries, with developing countries generally lagging behind in terms of access and usage. In terms of computing power concentration, data cited in the report stated that among the top 500 artificial intelligence supercomputers in the world, the United States accounts for about 75% of the computing power and China accounts for about 15%, highlighting that the current artificial intelligence infrastructure is highly concentrated in a few economies.
The report also paid attention to the serious imbalance at the language level. There are more than 7,000 languages in the world, but current mainstream artificial intelligence models only cover a very small number of them in training. For many languages, machine translation quality problems are prominent, and errors not only affect daily communication, but may also cause serious consequences in scenarios that are highly sensitive to personal safety, such as medical diagnosis and treatment decision-making.
The expert panel pointed out that the risks brought by the proliferation of artificial intelligence have begun to become apparent in many fields. These include potential negative impacts on human rights, social systems, and the natural environment, such as the increasing frequency of child sexual abuse material generated by artificial intelligence and sexual violence content implemented using deepfake technology on the Internet. The report believes that related phenomena have exposed obvious shortcomings in content supervision, criminal justice and transnational cooperation, and there is an urgent need to strengthen responses at the technical and institutional levels simultaneously.
In terms of the information environment, the report warns that artificial intelligence has significantly lowered the threshold for large-scale production and precise targeted dissemination of "persuasive content." This trend could lead to a progressive erosion of information integrity, thereby eroding public trust in media and institutions, undermining social cohesion, and interfering with the deliberative mechanisms of democratic processes that rely on facts and reasoned debate. Experts believe that in the face of such changes, the traditional information security and media regulatory frameworks have become insufficient, and it is necessary to update governance ideas and strengthen international coordination.
The report also pointed out that most countries - including some advanced economies - lack the ability to conduct technical assessments of the latest and most powerful artificial intelligence models, and it is difficult to have substantial technical participation to influence the global governance of these systems. This gap in technical capabilities and resources has further exacerbated international inequality in the formulation of artificial intelligence safety standards and risk prevention and control. The expert group called for helping a wider range of countries improve their assessment and regulatory capabilities through multilateral cooperation mechanisms and capacity-building projects, thereby forming a more representative voice in global governance.
This preliminary report will serve as a core reference document at the United Nations Global Artificial Intelligence Governance Dialogue for governments and relevant international organizations to discuss future regulatory frameworks and cooperation paths. The expert group stated that it plans to continue to collect and evaluate the latest scientific research results and practical application cases in the next year to provide a more detailed and systematic evidence base for the proposed comprehensive report to support all parties in seeking a more stable balance between risk prevention and control and technological innovation.