00:00I'm honored to be here and to be able to speak with you all at this important time.
00:04We've been talking about artificial intelligence for years, but it feels different in recent weeks and months.
00:10Rapid model progress has made the timeline feel more compressed, the upside more tangible, but also the stakes and the
00:16risks more immediate.
00:18We have a choice in front of us.
00:20AI can either be more like a new renaissance of creativity and discovery, or more like a new industrial revolution
00:26of upheaval and disarray.
00:29The renaissance was a time when new tools, new ideas, and new institutions expanded what people believed they could do.
00:36AI can play that role for our time.
00:39I believe that people are much more capable than they imagine themselves to be,
00:43but have been held back by limited technology that was meant to empower, but instead distracts and often steals attention.
00:50The best version of AI is not about making people cogs in a giant machine, or optimizing every part of
00:56life until the human parts disappear.
00:58It's about giving people more agency in an ever more complex world, more ability to learn, to create, discover, build,
01:05participate.
01:06There are many things that AI cannot and should not automate.
01:10AI should be built for people, not simply to turn the crank of the machines faster for their own sake.
01:17Some of the things people working to build AI have said are so dystopian that they sound like the plot
01:21of bad science fiction movies.
01:24Anything this powerful is also intimidating.
01:27This moment is complicated as we feel both tremendous potential and very understandable anxiety at the same time.
01:35These models can help people discover new knowledge, new medicines, better healthcare, stronger education, and build more productive economies.
01:41They can be new tools for people to pursue dreams and ambition that today feel out of reach.
01:46The complexity of life has increased so dramatically for the average person.
01:51AI can be a defensive tool against this complexity as much as it can be this tool of unprecedented discovery
01:57and innovation.
01:59On the other hand, as AI systems become more capable and more autonomous, they can move faster than our institutions,
02:06concentrate power in too few hands, or make decisions that people no longer understand or control.
02:14That concern becomes especially important as we approach systems that can improve themselves and future versions of themselves,
02:21often called recursive self-improvement.
02:24As the process of building AI becomes more automated, the pace of AI progress could accelerate rapidly.
02:30This moment calls for extreme care.
02:33I largely agree with Professor Bengio, although I'll say two instead of three.
02:37I believe there are two ways that AI progress could go very badly that we have to avoid.
02:43First, we could lose control of the future to AI.
02:47The risk is that it moves so fast that people can no longer follow what's happening or intervene when needed.
02:52This would obviously be terrible.
02:54The industry must not accept too much technological risk just because the benefits are too great
02:59and that they feel too important to slow down.
03:01Beating companies in a competitive race is not a reason to make rash decisions,
03:05nor do we believe we are locked in a race where we are unable to do that.
03:08We have unilaterally slowed down in the past.
03:10We will do so in the future.
03:11We do not want to build systems.
03:12We think it's important that everyone avoids building systems
03:15where we cannot make the alignment, monitorability, and safety guarantees that we must.
03:20We need to understand what these systems are doing
03:22and have strong evidence that they will do what people intend,
03:25even as they get very, very smart.
03:27Actually, especially as they get very, very smart.
03:30It doesn't matter whether people put their risk of catastrophe at 10% or 1% or 12% or
03:340.1%.
03:35None of these levels are remotely acceptable,
03:37and we should not train models that we cannot make an extremely strong case
03:41that we'll be able to keep under human control.
03:44So that's one way things can go wrong.
03:46In the other direction, these systems could concentrate too much power in too few hands.
03:51No one person or company or country should be able to use the most powerful AI models
03:56to impose their worldview on everyone else.
03:59A company or country that believes only it can be trusted with this technology
04:03can use that belief to justify almost anything else.
04:06We have to reject that logic, even when it's convenient for any one actor.
04:10I believe there is a pragmatic path through both of these.
04:14At OpenAI, we approach these questions with a clear goal.
04:18AI should give people more power over their own lives,
04:21more freedom to live lives the way they want,
04:23more ability to solve problems, big or small, and to improve their communities.
04:27We don't want to fall into the trap of blind optimism.
04:30We also don't want to fall into the trap of doomerism either.
04:33Getting pulled into either extreme will not help us successfully figure out the challenges in front of us,
04:38especially at a time when the world and its people face such enormous challenges
04:42that this technology can help solve.
04:45So we try to walk the middle path between these two poles.
04:48For us, that means three principles.
04:51First, as AI becomes more capable,
04:54people must remain at the center of AI decision-making.
04:57Alignment is not an abstract research question.
05:00It is the work of ensuring that these systems reliably remain under human control,
05:04reflect human values,
05:05and help people guide the next stages of development,
05:08we are not trying to and must not automate human judgment or human values.
05:13Second, the benefits of scientific progress and economic growth must be by people and for people.
05:19AI should help researchers make discoveries, doctors treat patients,
05:22teachers teach students, entrepreneurs start companies,
05:25communities solve problems that fell out of reach, and much more.
05:28The continuous story of human progress is not the tools replace human agency,
05:32it is that the right tools make people more capable,
05:35and I believe this is going to go much further than we believe possible.
05:39Third, this technology must empower people individually.
05:42A great future will come from people realizing their potential
05:45and building value for themselves, their communities, their countries, and the broader world.
05:50We might be the builders, one of the builders of this technology,
05:53but we are not the heroes of this story.
05:55Our role is to enable people all over the world
05:57and to trust in the magic of humanity's skill and faculties.
06:02We now have models capable of discovering knowledge that we have never had before.
06:07Consider the progress we've seen in math.
06:09Three summers ago, our models were okay at grade school math.
06:13Two summers ago, they were pretty good at high school math competitions.
06:17Last summer, we reached a gold-level medal
06:19in the most prestigious international math competition.
06:22And just a few weeks ago this summer,
06:24one of our models solved one of the Millennium Prize problems,
06:27the Navier-Stokes equations.
06:29These equations are used for aircraft design,
06:32weather forecasting, and the study of blood flow.
06:34We are not solving math problems for their own sake,
06:36but to empower people to discover new knowledge
06:39and to use it to improve health care,
06:41raise standards of living,
06:42and expand everyone's potential around the world.
06:45There remains so much to be discovered.
06:48But it's very important that companies not substitute
06:50for the democratic process.
06:52If AI is to be democratic,
06:54the most important decisions cannot be made
06:56by labs in San Francisco alone.
06:58They must be shaped through democratic processes
07:00and by governments accountable to the people that they serve.
07:04At the international level,
07:05this will require cooperation.
07:08In our history,
07:09there have been times where countries who compete
07:11and don't always like each other very much
07:13still come together for shared interests
07:14and the collective good
07:16in the face of a powerful new technology.
07:18We believe this must be one of those times.
07:22We need a mechanism for complementary national
07:24and international frontier AI standards.
07:27Standards for measuring capabilities,
07:29assessing risks,
07:30determining whether safeguards are sufficient,
07:32and preserving meaningful human oversight
07:34as systems become more autonomous.
07:37We need common standards
07:38so countries can compare evidence,
07:40verify compliance,
07:41and have a shared language and understanding
07:43about what is happening.
07:45We need accurate and speedy incident reporting,
07:48classification and reporting protocols
07:50so the world can learn from failures
07:52before they become catastrophes.
07:54And we need secure channels among governments,
07:57critical infrastructure operators,
07:58and technical experts
07:59to share emerging vulnerabilities and new threats.
08:03Obviously, these standards should not lock in incumbents
08:06or favor one business model over another.
08:08They must support open and closed model developers,
08:11new entrants in established labs,
08:13national authority,
08:14and international cooperation.
08:16Each government should decide
08:17how to incorporate standards into its own legal system,
08:19but we will all be better off
08:21if we can agree on what good evidence,
08:23good safeguards,
08:24and good oversight look like on the global stage.
08:27We do not have all the answers,
08:29and we're very open to any better ideas.
08:31We're ready to work with you all
08:33and try our hardest to figure out the path forward.
08:36We are at a crossroads.
08:38One path leads to a world where power is concentrated,
08:41decisions are opaque,
08:42and people feel that the most important technology
08:45of their lifetime is something being done to them.
08:48The other path leads to a world where AI works for people
08:51because people, through democratic institutions,
08:54help decide how it should develop and guide it,
08:56where scientific progress improves real lives,
08:59where more people can participate in the future,
09:01and where this technology continues to give individuals
09:04and communities greater power to live life as they choose.
09:07This is a future of realized potential,
09:09excitement, and wonder,
09:10and it is the future we believe is worth building.
09:12Thank you very much.
09:15Subscribe to One India and never miss an update.
09:20Download the One India app now.
