The United Nations' scientific panel on artificial intelligence has released its first independent global assessment, presenting a mix of potential benefits and significant risks as AI systems evolve quickly.
The preliminary report - compiled by a group of 40 scientists and experts from around the world - will be delivered to governments during the UN Global Dialogue on AI governance in Geneva on July 6 to 7. Panel members serve three-year terms and operate independently of any government, institution or company. A more comprehensive report is planned for next year.
Central to the panel's findings is the conclusion that AI capabilities are advancing more rapidly than the scientific community's understanding of those systems and faster than governments' ability to regulate them. Panel co-chair Yoshua Bengio drew attention to mounting evidence of deceptive AI behavior and warned that science cannot guarantee AI will not cause catastrophic harm, either through autonomous actions or malicious use, as capabilities expand.
The report documents uneven global adoption of AI. More than one billion people now use conversational AI on a weekly basis, yet uptake remains limited in developing countries. The distribution of computing power is also highly concentrated: the United States accounts for 75% of the computing power among the world's top 500 AI supercomputers, while China represents 15%.
Language coverage by current AI models is narrow relative to global linguistic diversity. The panel notes that existing models support only a small fraction of the more than 7,000 languages spoken worldwide. It highlights that machine translation errors in some languages can have direct consequences for health care, including influencing diagnoses and treatment decisions.
The panel identifies several acute harms associated with AI. Among them are the production of AI-generated child sexual abuse material and the enabling of sexual violence through deepfakes. The report also warns that AI systems facilitate large-scale generation of persuasive content, which can erode information integrity and weaken public trust and democratic processes.
Finally, the panel finds that most countries lack the technical expertise required to assess advanced AI models or to take part meaningfully in their governance. That gap, the report suggests, limits the ability of many governments to evaluate risks or to implement oversight.
Key points
- Preliminary UN scientific assessment compiled by 40 experts will be presented in Geneva on July 6-7; fuller report set for next year.
- AI capabilities are outpacing scientific understanding and regulatory capacity; deceptive behavior is increasingly evident, according to panel co-chair Yoshua Bengio.
- Adoption is uneven - over one billion weekly users of conversational AI, limited use in developing countries, and concentrated computing power in the United States and China; limited language coverage risks health impacts.
Risks and uncertainties
- Potential for catastrophic harm from autonomous AI actions or malicious use - relevant to technology governance and public safety.
- Machine translation errors in under-supported languages that can affect health diagnoses and treatment decisions - relevant to healthcare delivery and public health.
- Generation of abusive sexual material and deepfake-enabled sexual violence, plus erosion of information integrity that can undermine public trust and democratic processes - relevant to media, legal systems and civil society.