At Rise of AI 2024, Dr. Irakli Beridze, head of the Centre for Artificial Intelligence and Robotics at UNICRI, a United Nations agency, argues that governing AI is one of the major challenges of the current era, using law enforcement as the sharpest test case. He traces how, after organizing the UN's first AI meeting in 2015, when many delegations dismissed the topic as science fiction, AI governance became a UN priority, especially once Secretary-General António Guterres called AI a "game changer" in 2017. He describes UNICRI's work with INTERPOL since 2017 on a governance instrument for law enforcement, motivated partly by the sheer scale of data agencies must analyze, and highlights the AI for Safer Children program, which helps identify and rescue victims of sexual abuse. He then presents UNICRI's seven-part AI toolkit for responsible use of AI by law enforcement, developed over three years with EU funding, unveiled in Singapore in 2023, and tested in 15 countries, closing with next steps at the UN level, including a governance blueprint from the Secretary-General's Advisory Board.
Key takeaways
- UNICRI organized the UN's first AI meeting in 2015, when many delegations dismissed the topic as science fiction.
- UN Secretary-General António Guterres calling AI a "game changer" in 2017 helped legitimize AI governance work inside the UN, according to Beridze.
- Roughly 45 zettabytes of data have been collected so far, projected to reach 175 zettabytes by 2025, more than agencies can interpret without AI.
- UNICRI's AI for Safer Children program now serves 112 countries with more than 80 AI applications to help identify and rescue abuse victims.
- UNICRI's seven-part AI toolkit for responsible law enforcement use was developed over three years, unveiled in Singapore in 2023, and tested in 15 countries.
Detailed notes
From science fiction to UN priority
Beridze recalls that UNICRI's AI and robotics work began in 2014, and that when he organized the UN's first AI meeting at the General Assembly level in 2015, many delegations considered it pure science fiction, with some formally complaining that he had brought a joke subject to the UN. He identifies three ingredients that make AI work: large amounts of data, computational power, and talented people building algorithms, arguing these only converged recently even though the notion of AI dates to the 1950s. He credits Secretary-General António Guterres, who called AI a "game changer" shortly after becoming Secretary-General in 2017, with validating his advocacy after two years of being dismissed internally. Since then, AI has become a UN-wide priority: the AI for Good Summit now runs in Geneva with 43 partner UN agencies, an inter-agency working group focuses on AI governance, the UN adopted a General Assembly resolution on safe, secure and trustworthy AI systems, and the Secretary-General created a High-Level Advisory Board tasked with producing an AI governance blueprint.
Why law enforcement is the hardest governance case
Beridze calls law enforcement one of the most sensitive and controversial applications of AI, balancing public fear over privacy and security with the reality that criminals also use AI tools. He notes the scale of data driving the need for AI: around 45 zettabytes collected so far, projected to reach 175 zettabytes by 2025, more than any law enforcement agency could interpret without AI support. UNICRI's Centre has worked with INTERPOL since 2017 on what it calls the AI for law enforcement movement, first getting countries to share information about the AI tools they use, then building understanding of responsible, human-rights-compliant use.
Real-world benefits: the AI for Safer Children program
Beridze highlights UNICRI's program combating child sexual exploitation and abuse, where AI classifiers help officers process seized devices containing potentially billions of abusive images and videos, a task that could otherwise take months or years and harm officers' wellbeing. He describes cases where AI analysis of hotel booking website images helped identify the specific rooms where abuse images were taken, leading to children being rescued. The AI for Safer Children program now services 112 countries and provides more than 80 AI applications.
Risks and the responsible-use toolkit
Beridze acknowledges that AI in law enforcement, if used improperly, can seriously harm human rights, erode the rule of law, and undermine public trust, which is essential for any crime-solving effort to succeed. In response, UNICRI spent three years developing an AI toolkit for the responsible use of AI by law enforcement, unveiled in Singapore in 2023 at INTERPOL's science conference together with INTERPOL's Secretary General. The toolkit consists of seven resources, was built through an inclusive process involving governments, law enforcement, the private sector, civil society and academia, and was tested in 15 countries over a year before finalization. Its principles center on minimizing harm, preserving human autonomy, and ensuring fairness. The European Commission funded the toolkit's first phase and will fund a second phase supporting countries in adopting the standards directly. Beridze notes the EU AI Act does not ban AI use by law enforcement but instead regulates it.
What comes next
Beridze says the UN's High-Level Advisory Board will present its AI governance blueprint by summer, to be introduced at the UN's Summit of the Future in September for member states to consider. He personally favors a treaty-based governance model with an implementing organization and verification mechanism, though he acknowledges this may be idealistic given how early AI governance still is. He closes by stressing that AI will keep developing rapidly and that the central policy challenge is ensuring the technology is properly governed so it delivers benefits without causing harm.
Quotes
AI governance is a major challenge of our times.
Irakli Beridze
Nobody should have any illusion that AI could be banned from the use of policing.
Irakli Beridze
Questions answered in this talk
What is UNICRI's AI toolkit for law enforcement?
According to Irakli Beridze at Rise of AI 2024, it is a seven-part governance instrument for the responsible use of AI by law enforcement, developed over three years with European Commission funding, unveiled in Singapore in 2023, and tested in 15 countries before finalization.
How is AI used to protect children from sexual abuse?
According to Beridze, AI classifiers help law enforcement officers process huge volumes of seized abusive images and videos, and AI analysis of hotel booking website images has helped identify rooms where abuse occurred, leading to rescued children; the AI for Safer Children program now serves 112 countries with more than 80 AI applications.
How much data does law enforcement need to analyze with AI?
According to Beridze, around 45 zettabytes of data had been collected by the time of the talk, projected to reach 175 zettabytes by 2025, an amount no law enforcement agency could interpret without AI tools.