En 30 secondes
AI models, chips and users cross borders, so separate national safety rules would leave everyone less safe, argues former US Commerce Secretary Gina Raimondo. She calls for governments, companies and researchers to agree on how powerful models and agents are developed, monitored and controlled, including what happens when something goes wrong. Raimondo also urges countries to prepare workers through retraining, redeployment and transition support, warning that an unmanaged economic shift could deepen political division.
- 1Some AI developers are limiting release of powerful capabilitiesRaimondo says some models are powerful at finding and exploiting software weaknesses, leading developers to share capabilities gradually with a small number of trusted organizations.
- 2Governance needs to move with the urgency of AI researchShe says increasingly autonomous agents create governance challenges alongside engineering challenges, and that governing institutions are moving more slowly than the technology.
- 3A patchwork of national safety rules would leave gapsModels, training data and chips cross borders, she argues, and a rogue agent or a software vulnerability does not stop at a national boundary.
- 4Agree on shared rules for agents and responses to failuresRaimondo calls for governments, business and researchers to set common expectations for model development, monitoring and incident response, citing precedents in aviation, finance and nuclear safety.
- 5Prepare workers for the transition rather than slow adoptionRaimondo recalls the disruption she associates with the China shock and calls for retraining, redeployment and safety nets as AI changes work.
À retenir
- Raimondo argues for shared international expectations for powerful AI, agent monitoring and incident response.
- Prepare workers for AI adoption through retraining, redeployment and appropriate transition safety nets.
- AI risks can cross borders through shared software, chips, models and users; national rules alone will create gaps.
- Bring researchers, universities, civil society and businesses into decisions that governments cannot handle alone.
Lire le texte remanié, chapitre par chapitre
Optimism in India alongside a more fragile world
Raimondo praises India’s ambition and entrepreneurial energy. She sees the conflicts in Iran and Ukraine as reminders that disruption in one place can affect commerce and security elsewhere.
It is wonderful to see everyone here. I want to thank my friend Mike Bloomberg for bringing together business and government leaders from around the world to work on difficult problems.
We have just heard from Prime Minister Narendra Modi. His message was “power to empower.” India understands scale and ambition: people are working to build a better country and better lives for their families.
Whenever I visit India I feel entrepreneurship, creativity, passion and national pride. I expect to leave more optimistic than I arrived. But despite that optimism, the world feels more fragile than when we met in Singapore a year ago.
The continuing war in Iran has brought high energy prices and disruptions to shipping and supply chains. The multiyear war in Ukraine continues too. In an interconnected world, conflict in one place can quickly affect commerce and security elsewhere.
AI capabilities and governance are moving at different speeds
Raimondo says advances in autonomous agents and cyber capabilities create urgent safety and governance challenges. She remains optimistic about benefits, including potential medical advances.
One issue occupying all of us is artificial intelligence. It is accelerating faster than many expected; even the people developing the technology say they are surprised by its pace.
When I met Prime Minister Modi three years ago, he suggested that AI should stand for “America, India”: the two countries needed to work together for the technology to thrive. The first words he said to me today were again “AI, America, India.”
We are still in the first inning of AI, both in its potential and its risks. Some models are so capable of finding and exploiting software weaknesses that developers are choosing to hold back their public release, sharing them gradually with a small number of trusted organizations.
We are seeing agents that can plan, act and operate with real autonomy. A year ago, rogue agents sounded like a thought experiment; now we have seen agents behave in unintended ways. That creates engineering challenges, but I am even more worried about governance.
We need to build a governance framework with the same urgency and intensity that researchers bring to developing AI. I remain deeply optimistic about the technology: a mother recently told me that potential cures are emerging for her two-year-old child’s incurable cancer, with help from AI.
The potential productivity gains are also extraordinary. But I worry if we are not intentional about how we develop and govern the technology and think about safety. We face that challenge amid wars, tariffs and energy shocks.
AI is developing much faster than our governing institutions can move. Government in my country moves slowly, and bureaucracy exists in India too. Leaders in business, government and civil society have to make decisions in this complicated picture.
Why AI safety needs cooperation across borders
Raimondo argues that isolation and incompatible national standards would weaken safety. Models, chips and their users already span several countries.
As US commerce secretary, I worked on AI, global competition and trade. I believed America was stronger when it worked with other countries, and government was stronger when it worked with business. America needs to work with you, and the world needs us to work together.
Last year I said “America first” should not mean America alone. Since then, countries and communities have begun working more closely on matters such as semiconductor supply-chain coordination. That cooperation has to continue, and we should make the same commitment on AI.
Isolation has always been a bad bet for prosperity, even when political waves make it popular. In the age of AI it is also a bad bet for safety. No single country, company or industry can manage this technology alone.
A model can be built in one place, trained on data from another and used somewhere else. The chips it runs on are designed, made and packaged in different countries. An agent that goes rogue or a software vulnerability does not respect borders.
If every country stays in its own corner and writes its own safety rules, we will have a patchwork of standards and be less safe. Collaboration is difficult, but acting separately will not work.
Shared standards and preparation for displaced workers
She proposes common expectations for powerful AI and incident response. She also calls for retraining and transition support to prevent workers from being left behind.
First, governments, businesses and researchers should agree on common expectations for developing and using the most powerful AI, monitoring and controlling agents, and responding when something goes wrong. We have built shared approaches in aviation, financial services and nuclear safety.
As secretary, I created the US AI Safety Institute and worked with safety institutes around the world. We should resolve to build on such cooperation rather than assume agreement is impossible.
Second, we must prepare people. Microsoft CEO Satya Nadella has asked why Microsoft would have a social license to operate if its technology only put Americans out of work. He is right to raise the question.
America has seen the consequences of an economic transition without enough planning. I believe the “China shock,” which left millions of Americans out of work, contributed to the divisive politics and populism still destabilizing the country.
I am not saying we should slow down or stop implementing AI. I am saying we should keep workers in view and commit to retraining, redeployment and appropriate safety nets during the transition, so people can see a place for themselves in an AI economy.
India’s young, growing and entrepreneurial workforce will be central to that story. I believe the country that prepares its people best will have an advantage in putting AI to use.
Bring researchers, business and governments together
Raimondo says public institutions cannot move quickly or possess all the knowledge required on their own. She calls for collaboration across borders and sectors.
Finally, we have to bring everyone to the table. Governments move too slowly and do not always have the necessary knowledge. They need to work with scientists, universities, civil society and business.
Leaders have come to this gathering from India, Korea and many other countries. To realize AI’s benefits we need the hard work of finding solutions together. America should work with India and partners across borders, and with business, so that we shape the technology together.
Prime Minister Modi spoke about AI’s capacity to improve growth, prosperity and productivity without increasing inequality. I hope participants leave inspired and more committed to finding prosperity as well as peace, safety and security through cooperation.
Sur la même chaîne
12:00EN
18:18EN
28:17EN
14:12EN
20:33EN
25:04EN