[0:00:03] Artificial intelligence is widely known and there's a lot of news about it. Everybody can see it and feel it throughout their day with gadgets and technology. But now we're looking at artificial general intelligence and a lot of people don't actually understand or know what this is. What specifically is artificial general intelligence? At OpenAI, we define artificial general intelligence as highly autonomous systems which outperform humans at most economically valuable work. And so one analogy for how to think about it.
[0:00:31] Think of it as an organization of many of the smartest humans that have ever existed working together on a single goal without breaks, without ego, without distractions, and operating at computer speeds rather than human speeds. And so if you think about the impact of something like this, you could imagine something that would go and read the entire scientific literature and then think about how to cure cancer and how to combine insights from different disciplines and how to have great...
[0:01:00] I think that technology
[0:01:27] It's fundamentally about enhancing humans, right? It's fundamentally about, you think about our abilities when you give a human a smartphone, right? We're suddenly able to text anyone on the planet, right? It's almost psychic communication. And I think that that is the point of technology. The technology for its own sake isn't what we should be pursuing. I think that we have a wide range of technological options available for us of what do we want to build, right? Where we sit today is that we're starting to see the shape of technologies that could really bring us to AGI. And the question of,
[0:01:57] How do we want to shape that technology? Do we really want to build systems that are going to be fully autonomous, that are going to look a lot like the robots in movies? Or do we want a system that's much more like a Google that just works way better? You can ask it all sorts of questions of, what should I do today? And it tells you what you should go do, right? And so today, when we're sitting before this very powerful technology exists, before it's deployed, we have the option of thinking about, how can we make sure that this is something?
[0:02:22] That really enhances everyone. It certainly is a very powerful technology and it could possibly solve some amazing problems that we face in the world today. How do you go about preventing the misuse of this technology? So I think there are three main risks. So risk number one is that you build systems that are not aligned with human values, right? So that you have a system that pursues goals that are not what the operator wanted. Second problem is humans subverting.
[0:02:49] And this is something that we see all the time with hackers. And a third problem is that we actually could build systems that do what their operators want, are unable to be hacked, but somehow don't result in human lives generally getting better. And so you have this economy that's growing and flourishing, but somehow only a few people are benefited. And I think that all three of those things are things we need to worry about. Something that we can all influence and all have a stake in is the conversation around
[0:03:17] What should the values of these systems be? And who benefits? Who's in control? And what should the deployment look like? And as you advance in the space, what kind of developments and technologies and algorithms and applications are you currently working on? And where do you see artificial general intelligence going within your space at OpenAI?
[0:03:39] So in 2018, we had three major breakthroughs and a number of other smaller advances. So we became the first company to be able to build AIs that can play esports at world competitive levels that we were able to program a robotic hand, which is something that has existed for many decades. No one's been able to program before. And this was using the exact same technology for both professional esports and for controlling this robotic hand. We also were able to have language breakthrough.
[0:04:08] And we're able to have a single model, so a single artificial brain that's able to solve this huge variety of different natural language processing tasks. And so that's the flavor for us, solve tasks that couldn't be solved before in increasingly general ways. And that our strategy at the end of the day on the technical side is to continue to advance AI until you reach human level AI.
[0:04:32] How advanced will AGI get? Will it become better in all aspects than humans? Will humans end up not being able to compete? So I think that the question of how humans relate to this technology is really open. And it is the case that I think there will always be a place for humans in the same way that even though cars are a lot faster, people still run track. Humans are a lot more multi-dimensional than any particular thing. And so I think that the real question is,
[0:05:02] If we have these really powerful technologies that can go and do science and can start companies and be creative and create great works of art, how do we want to relate to that? How do we want to spend our time and how do we make sure that we can merge with this technology, at least at a social level?