Why AI Is an Even Bigger Deal Than You Think | Reed Hastings | TED

00:04So Reed education which a lot of people I don't I don't know how many know how active you are in education education philanthropy charter schools etc. What are you doing in that space? uh on the board of Khan Academy um and generally I've been working for the last 25 years uh to try to find ways to make in particular US schools but some international uh work better um and have better outcomes for kids and I would say after a billion dollars in 25 years um we're down below where I started so uh it's a hard problem and what do you you know we heard about the the risks of edtech um but you are

00:43investing in in edtech you're giving to edtech. What are your thoughts? What what's the trade-off there? Why why do you continue to to potentially believe there? You know, in the 18th century, most factories were steam engine driven. So, they have a big steam engine, rotating uh rods and pulleys and belts and drove the factory, very efficient.

01:01And then electricity came in and they replaced the steam engine with an electric engine um thinking that it was going to help a lot. And productivity didn't increase at all. And they were puzzled. And then they realized, hey, the limiting factor is the power distribution system.

01:17All these spinning rods and mechanical power. And they ripped that out and just put in small electric motors for each device. Then they could move things around, fit them in better because it wasn't aligned to the spinning. Uh they could have variable speed, turn different motors on and off.

01:32Uh and then um productivity increased dramatically. So that's a classic uh kind of economic, you know, surprise lesson. And it's always stuck with me because I think that's what we're seeing, which is we keep doing things to improve classroom education, but the fundamental power distribution is 25 kids stuck at the same level.

01:56And that the friction that that creates which is you know roughly a third of the kids are behind a third of the kids are bored and above and a third you're teaching to is the fundamental friction in uh our mass education system. Uh and the theory is if each of us had an individual human tutor.

02:14Um so imagine you go to school um you have your normal social activities but when it comes to learning uh you get an individual who's going to sit down with you and um and then they could do con academy or they could do colorbook or whatever is appropriate uh that learning would be massively increased um and there was a a famous study uh 40 years ago Bloom uh that documented this uh and now a friend of mine uh Ben Summers is redoing that study but at much bigger scale.

02:47And I think what we're going to see is the key to much more learning. Um where middle school kids know all of the high school curriculum, high school knows all of the college, much better outcomes will be individualized education. How do you square that with what Jonathan Height said?

03:03Look, these screens, I mean, at least it's a correlation. We don't know causal yet, but it seems to be distracting. It seems to to correlate with with some test scores going down. Is it for you a little bit of it? Is it an all or nothing type of thing?

03:14I'm a huge fan of Jonathan. Totally agree with uh all his zip up the um phones and don't use phones and and I think you guys clarified uh he likes uh offline tablets. That's fine. It's the internet that's the the problem, not the physical device.

03:31And so I think there's lots of ways to uh cater to the concerns that he correctly expresses, which is uh letting kids go wild on the internet under 16 is is not great. But you don't have to do that to be able to do individualized tutoring. So the individualized tutoring we're doing is with humans.

03:49Okay? Now they can use some technology if they want. Uh but then obviously that's cost prohibitive because it's about $100,000 uh per kid per year. Um so then the the the hard challenge becomes how do we use AI to approximate that human and provide everyone an individualized education as AI gets better.

04:11current AI is not good enough to do that. But, you know, in three years, we've gone from, you know, chat GPD35 and barely um being able to do high school math to, you know, just incredible intelligence and um you know, that's likely to just continue to double double double um and get better and better and better.

04:33So there is a world where the AI I think will be able to match and beat the human individual tutor. And what does that world look like? Let's just say it's in 10 years. Are you imagining that you're just socializing and then you go to this AI tutor that even maybe looks embodied in some way but there are you imagine there's no human teacher to what do you think happens to that role that profession?

04:56What about the adult humans in the classroom? Well, let's think about schools. So uh three big purposes. One is create good citizens. Another is give economic opportunity to the kids and then the other is socialization um social emotional learning uh how to work with other people adults outside of your family.

05:15So only in the first part is really where uh online is really good and what we want to do is have teachers be able to focus on social emotional learning. um they become uh you know really helping maturity interpersonal skills uh values clarification all those kind of higher level things um and then we've got to figure out in the AI age um you know how do we enhance that role of creating good citizens okay because one of our uh ways to come together is to have you know a tighter idea of who we are as a country and you know the K12 systems been able to take that for

05:53granted for the last 100 or 200 years maybe post civil war you know because society was working well but if we're going to go into a period of stress it's really important for that that mission to get attention also and and I just want to double click on that and make sure maybe we have a common vision or maybe it's divergent you you still see a major role for the human teacher and the human classroom you just see that role shifting potentially going sort of okay so uh most teachers today um their pride center is teaching and connecting and you know understanding the material.

06:28Okay, some part of that is really connecting on a personal level with the student. So that part is the social emotional but in terms of transferring information what educators uh cynically call sage on a stage it's eliminating sage on a stage as a teaching modality.

06:45Okay. And so it's really just focused on the individual. What would education be if there was no mass teaching? And to be fair, you know, if you go to an ed school, if you went to an ed school 20 years ago, this is what they were preaching.

06:59Differentiated instruction, active learning, don't be staged on the stage, have a Socratic discussion, do a simulation, have so it's really potentially, and this is what I say because I get this question a lot. The teacher, I think, moves up the value chain and is able to facilitate and drive a lot of that active learning, which is better better for everyone.

07:14I think it's more fun for the teacher, right? The positive side of it. And the other part is once you can do a lot of this in software, you can do it globally. So it's really hard to scale up the teacher force. If you have incredible software, it's a pretty inexpensive to make it globally available.

07:32No, that's right. I I mean, you know, we we talk a lot about the Khan Academy board that the technology can raise the the safety net, raise the floor. We have seen stories of young women of Afghanistan using Khan Academy, one of them's at MIT now.

07:43I mean, it's amazing things. But we see also in the classroom, most students need that human element. Arguably all of them do ideally if they have it. Switching gears a little bit because your your other board you're on is in obviously very related to this anthropic.

07:58I actually I'm just curious what you know you could do anything. What made you join that board? What's it like at those board meetings when I'm assuming y'all talk about um pressures from the White House, how your new model might break all software?

08:12um what tell us what you can. Yeah, it's a lot like your board meetings, you know, talking about uh the software and what it can do and how it needs to get better. So, the mission of the company is very clear. It's not maximizing profits. Um it's that we're successful, the hum humanity, how do we get into the age of AI successfully crossing through sort of this uh portal and they recognize it's going to be very challenging.

08:42Um, and they're very dedicated to having that happen in rolling out AI and having the incredible beneficial outcomes, whether that's the Whimo self-driving, whether that's gene editing, um, whether that's curing cancer, you know, 10, 20 years from now, uh, it's very likely we'll have pretty abundant energy.

09:00Um, we will have, uh, amazing health outcomes. I mean, so much positive outcome from the AI infusion into science. Um, and I would say Anthropic is very serious about helping us manage or avoid most of the downside. And and how have you all pulled that off in closed doors?

09:20And I I've been in some of those closed doors where pe people are genuinely afraid more than 10% chance that this could be an existential threat to humanity. It does seem that anthropic somehow is is proving it to be very responsible um or that that's what we appears to be and at the same time moving very fast hyper speed.

09:39It feels like almost every few weeks there's something new and it and it's very tangible in what it might do for work. How are y'all balancing that at anthropic instead of just saying go go go? You know, I think all of the big models are improving rapidly and you know, you're probably going to see them go, you know, certain ones on the lead in certain areas over time and, you know, frankly, it's good for the country if, uh, you know, we have three really successful models to to choose from.

10:02Um, and then how do they balance it? Um, you know, case by case. uh so uh I think each one they have to see you know how accelerated is learning how powerful is it uh what are the downside scenarios what is the new possibilities it can do and I'm curious about anthropic itself I I had a chance to visit there a couple of weeks ago and you know I I take pride that you know Khan Academy we're super nimble and we're innovating etc and we're obviously trying to leverage AI for for social good as much as possible when I visited there I tangibly felt that they were pioneering completely new

10:38ways of running an organization, new ways of developing product. I think it was something that co-work was what was it a week or two that that it was essentially built primarily by the AI itself. What will an organization look like uh in the future?

10:49I think you'll have a pretty good crystal ball there. Yeah, I don't know that most companies will come to look like anthropic. I think it really depends on your industry. If you happen to be a pure software company, then might be relevant for that classic company.

11:01But broadly across the economy, I think everyone is figuring out, you know, it's a a bigger version of the internet wave where all companies had to, you know, do things and we used to talk about our AOL keyword, you know, and crazy stuff like that, right?

11:16Which was the phasing in. Um, this is a lot bigger and more intense, but it's sort of a larger version of that same thing, which is all companies around the world, organizations, governments, militaries are scrambling to figure out uh, you know, how to use AI.

11:30Well, I guess related to that, people are talking about it with software engineering. People are talking about call centers. I have a friend who has one of his startups has a call center in the Philippines. They're going to lay off 80%. That's 7% of that country's GDP.

11:43How are you thinking about jobs? How just just as a thinker, how is anthropic thinking about it? Well, look, if you look over the last 200 years, there've been a bunch of dislocations. Um, but they were in, you know, happened slowly in over one part of the economy.

11:58Um and so the danger is you know are there multiple that happen in multiple fields. If these happen slowly then people are able to find other roles. Um so it depends on how fast this all comes. Um and uh again there's I would say the biggest uncertainty for everybody is how fast is the AI getting better and how fast will it be getting better going forward and then that um changes your views and assumptions.

12:25So let's look at self-driving cars. I mean, you know, we thought 20 years ago it was going to be pretty vast and it's 20 years later from when it started 2007 and the DARPA Grand Challenge and we're at like what.1% of all miles maybe 0.001%.

12:43I mean that's really pretty tiny okay 20 years later. So these things take a long time to actually mature and diffuse. Another example is Jeffrey Hinton won the Nobel Prize for uh his work on neural nets and really the father of AI and in 2016 so 10 years ago he said stop training radiologists now because in 5 years 2021 there would be no need for them.

13:10So what's happened instead is um as uh radiology got AI boosted the cost came down. You can walk into an MRI center in the US for $300 and get an MRI now. And so docs started ordering them more and using them more and insurance covered them more.

13:25And so the number of scans has gone way up. And guess what? We have a shortage and radiologists. We have 35,000. We need about 40,000. Wages have climbed to close to $500,000 a year. Um so even the best intention people in the field, okay, can prophecy disaster in radiology and have it be not accurate.

13:47So again, and it's just a timing thing because in 20 years, I'm confident Hinton will be right. Okay? So just think there's two examples there, which is a lot of the stuff may not happen right away, but it's still probably going to happen in a long time, 20 years.

14:04Some of it might happen in a short time. So you want to be ready for it. And that makes sense. Although something does feel different about this time. And all the people leading these these uh AI labs are talking about not 20 years, they're talking about next year, they're talking about 2 years you're going to have a data center of, you know, super intelligent geniuses that can do our work.

14:27Do you do you think they're wrong or do you think it's a probability and even if it's a even if it's a 10% chance that they're right, are you worried that this can lead to political polarization? What happened in globalization could now happen what happened in 30 years could happen in two or three.

14:42Is that is that not a concern or should we start doing something about that? Um if the AI really gets incredible in a very short amount of time like starts writing itself and self-improving all of that isn't it writing 90% of itself right now.

14:55Um you know again lines of code is a tricky measure um you know when it's invented a new type so of learning you know so there's you know reinforcement learning um you know when it invents something completely new as the then you can say that um but so there are cases where it's pretty fast and so I think it's important to say we should be ready now here's the thing we talk a bunch about unemployment what it will do every other time we've had big unemployment it's been a recession and so the stock market's down and government tax revenues are down.

15:28Um, this time if AI is successful in the way we think it will, I think we're going to see high productivity, high GDP growth, um, high stock market, and high unemployment. So, we'll have money to do things. And so, think of it like the Alaska Fund, which is oil, or the Norwegian sovereign fund.

15:50We need some ideas like that. What are we going to do with all these huge tax revenues that are going to be able to come in with big growth? And if we've got a fund which is for the benefit of all citizens, then we may in fact be able to have a path to a, you know, a glorious and harmonious society.

16:03And I don't mean, you know, UBI, that's got a a bad tink to it, but it's uh a partial sharing of the rewards, which is again what happened with the Alaska fund and and the Norwegian fund. Makes sense. Well, Reed Hastings, thank you so much and thanks everyone.

16:20Accelerating possibilities.