UN Warns That AI Safety Is Lagging Behind AI Progress

While artificial intelligence (AI)’s capabilities are expanding, a new United Nations (U.N.) report says its safety rules are lacking.

The preliminary report by the U.N. Independent International Scientific Panel on Artificial Intelligence published Wednesday (July 1) said that time is running out to develop effective worldwide governance of AI.

“The technology is transformative, but if the world keeps moving along this trajectory, humanity will fail to realize the gains it promises,” Maria Ressa, co-chair of the Independent International Scientific Panel on Artificial Intelligence, said in a news release accompanying the report. “The risks — to societies, to security, and to our species — are too high, and the forces driving AI forward are not the forces that will deliver its benefits.”

The report points to some AI-related successes, such the use of the technology to identify food insecurity, as well as medical and healthcare breakthroughs.

“AI has predicted the structures of more than 200 million proteins, accelerated drug discovery, vaccine development and research into antibiotic resistance,” the report said. “Doctors are using AI to detect diseases such as breast cancer earlier, while health workers in developing countries are using AI tools in local languages to improve patient care.”

At the same time, AI is allowing for the spread of sexual abuse material and sexually explicit deepfakes, with women and children at the greatest risk, the report found.

The U.N. also noted AI’s role in spreading disinformation and exacerbating mental health issues, as well as its use by criminals “to carry out cyberattacks, fraud and social engineering scams.”

The gulf between rapidly advancing AI capabilities and effective methods of risk management could someday bring about “catastrophic outcomes,” the report said.

“Reliable methods for retaining control over highly autonomous AI systems are lacking,” the report said. “There are no scientific guarantees that AI agents will not violate instructions, and evidence is accumulating of cases where they already violate them.”

To illustrate the speed of AI adoption compared to other technologies, the report noted it took decades for electricity to reach most households, while the internet hit the one-billion user mark after 15 years.

“ChatGPT reached 100 million users in two months,” the report said. “Traditional policymaking has not been able to keep pace.”