Werner Vogels predicts the future | Amazon CTO on robot companionship, quantum-safe computing, more!

Coolish. Yeah, engineering has come a long way since. I think the first NoSQL tools were really the first tools. Now if you look at DynamoDB, if you look at these, they are super robust, highly scalable tools where everybody can build a business on. Come a long way.

So first of all, it's 21 years now, so that's a bit. And there is a predecessor to DynamoDB, it's Dynamo. One of the things when I joined Amazon, I joined as what was called a Director of Systems Research. The idea was to bring, let me go a little bit back. Think about 1994. When Jeff Bezos starts thinking about the internet and he starts a bookshop. He doesn't really want to start a bookshop. He's just fascinated by the internet. What are the things that you can do online that you will never be able to do in real life? And he just picks a bookshop. A good bookshop has 40,000 titles in stock. Yet there's millions of books out there. So he thinks you could do that online. But there is no software that you can buy. There is no book that says 'here is an e-commerce operation.' The word e-commerce doesn't exist yet. So everything that the Amazon engineers had to do to build Amazon was to invent everything themselves. Because the kind of technology that they could buy couldn't operate at their scale. We've had a number of blowups because of that. So when I joined Amazon, engineers at Amazon were brilliant at scaling but from a practical, lots of scars kind of approach. And Jeff hoped that by bringing a former academic in, you get some more robustness, a better fundamental approach to scale and reliability. And we did lots of large projects around removing all single points of failure or how to best measure. What does it mean to measure? If 50% latency on your web page means nothing. Well, it means that 50% of your customers are getting a worse experience. You need to know how much worse. So, I think if I think about CTOs, there's four or five different types and there's no real well-developed, well-described one. First of all, there's the data center manager who reports up to the CIO. You have the CTO that is the second person in the startup, often the co-founder, the first coder. And then you have the big thinker, the role that I got when I came into Amazon, sort of like look at the whole picture, what are the kind of things we need to do. But then when we started doing AWS, your world changes. You go from an internal focused CTO to an external focused CTO. Scott Dietzen, who was at BEA I think at that time, he called it really an external technologist. The ability to talk to your customers, look at how they are using my products, and what are the problems that I see with 10, 20 of my customers that they may not see as a single problem themselves, but where you can find if I can build a tool for that, I can really help my customers. So your role changes from being purely internal to being external and bringing back things. Then over time, I've become more and more interested in those organizations, both profit and nonprofit, that try to solve hard problems. With hard problems, I mean hard human problems. The United Nations expects that by 2050 we have two billion more people. How are we going to feed them? How are you going to make sure they have an economic future? How are you going to make sure they have health care? Those kind of problems are the ones that I mostly focus on today. On one hand, by the way that we build technology at Amazon, but also by looking for those often young businesses and how can we support them. Take for example the Ocean Cleanup project. It's a massive problem. The grand ocean garbage patch is full of fish nets and plastics. There's about 30 rivers that sort of contribute mostly to that. So these guys have built plastic with GPS in them, threw them in the river to see where they end up, or they have these AI cameras on boats to sort of first where do these boats go, but secondly also what are the things that they're seeing. Working with those kind of customers is extremely satisfying because we are solving actual real human problems. We're not building spam filters.

There's a long history by large technology companies, if you think about IBM, Oracle, or McKinsey, the consulting companies, of predicting the future. And always when I read them, I always felt that they were a bit self-serving. And the things that I saw in the real world were different. I see significant human problems, some of them are our own nature, but quite a few are caused by technology or the way that we use technology. There's a leadership principle at Amazon: with success and scale comes broad responsibility. That means you don't only have a responsibility towards your shareholders to make money. You also need to serve your customers. At AWS, of course, that is builders. How can we help builders best? Then I write these predictions about things I see around me. I have the fortune to travel the world. I was just in spring in Sub-Saharan Africa: Nigeria, Rwanda, Kenya. In Kenya, we were in Nairobi. Of course you stay in the center of Nairobi, beautiful buildings. You drive 15 minutes and you see how the real Kenyans live. Most of them are day laborers. They go to the bank, they get $2, they try to buy something, they try to sell it, they bring the $2 back and hope to have 40 cents left to buy food. They probably have that, but then they don't have money left to cook it. Why? Because these big butane canisters cost $10 and they don't have $10. There's a young company called Koko Networks that built a sort of ATM where you can go with a canister, plop it in, and get 15 cents of gas. Then you can take that home and cook your food. This is not a world-shocking problem, but it is solving really hard problems that people have. Those kind of companies and those kind of problems are often the things that I like to serve through the predictions. This year, I do think there are things every year. I could have done 10 predictions, but five works fairly well. Three is a magic number, five is better. Was it two years ago? A topic that has become more and more openly discussable is menopause for women. It's a major problem for them, but even mothers don't talk to their daughters about it. However, that is slowly changing. What kind of things do we see in the startup community that are trying to build technology products or medicine to solve those kind of problems? So I also try to talk a little bit about things that people normally find a bit harder to talk about.

Well, it's one of these problems. I did this documentary series called 'Now Go Build' TV series probably two or three years ago. I was in Japan for that, and that's where I first really dove deep in this particular topic. Japan is a society where kids take care of the elderly. The grandparents live with the grandkids and the kids, and that was normal. But the younger kids want to start making careers, and there is a clear shift in Japanese society happening where the elderly are no longer taken care of by the younger people. As such, that's a shocker not for the young kids, it's a shocker for the old people. They're really on their own. I think there are a lot of technology solutions possible to help these people. Loneliness is something we'll get to, but anything we can do as technologists to keep people longer out of hospitals or care homes, the better it is. If that means a company in Japan that I met called ZWorks, they try to do all these innovations around elderly people to help them. Make a sensitivity pad in your mattress, and maybe in the middle of the night you go to the toilet and you don't come back within 15 minutes, the alarm goes off. This doesn't seem like world-shocking revolutions, but it does mean that people can stay independent longer. They don't feel that old. More and more people becoming 100 years old is actually not that strange anymore. Many of these people may have lost their partner, or live far away from family. Especially in the US, after high school you probably move away from where you grew up. You go to college somewhere in upstate New York, you go to work in California, and people are not close anymore. That truly is a loneliness epidemic. So what can we do? I love the work that is happening around companion robots. Why? Because I always thought that people would treat a robot like a piece of metal mechanics. Turns out it's not at all the case. They treat them like pets. Kate Darling at MIT has done all the research around this. I can tell you absolutely amazing stories about how people get attached to their devices. The story I think is also in the predictions: 80% of the people that have a Roomba have given it a name. One of the stories that Roomba tells: someone sent an old Roomba back because it needed to be repaired, and they looked at it and told the customer, 'We'll just give you a new one.' And they said, 'No, no, no, we want Mur back.' They want the old one because they're attached to it. Or Kate tells this story where she divides two groups of people and each gets a baby dinosaur. They really fall in love with it, give it a name, play with it. Then she gives them a hammer and tells them, 'Kill the dinosaur.' Nobody in the group will take the hammer and kill the dinosaur. They have no problem going to the other group and killing their dinosaur. Why do I tell these stories? Because while these things are technology, we treat them as living beings. Especially, before I was in computer science I was in healthcare. The application of these kind of tools in hospitals is incredible. If you're a young kid, a doctor is someone that is tall and dangerous or whatever you want to think. But for example, there's the Huggable robot, a small green thing with a nice snout that talks to the kid. Kids with a Huggable have no problem talking to a doctor because they have their mate with them. Those kind of things are crucial for us as technologists to think about. We see a problem, not necessarily a problem that is going to make us a ton of money, but a problem that is really important for people's psyche, for people's feelings, for loneliness. Who wants to be lonely? Nobody. If we can't do this through pure social interaction, what's the next best thing? Amazon has Astro, but there are more robots like Astro that will go search for you through the house and ask you if you've taken your medicine. In families or people that have these kind of robots, their medicine taking goes up 80% because they're being continuously reminded. Early on when we launched Alexa, I got a story from a man who was suffering from early dementia. He knew what was coming. The only thing he had to do was put a post-it note on the device saying 'It's called Alexa.' Because one of the things he noticed was that during the day his caregivers were getting more and more irritated with him because he would ask the same thing over and over again: 'What day is it today?' Because he had completely forgotten. The thing with that voice assistant was that you could ask it the same thing over and over again and it would never get angry. It would always be happy and always give the same answer. So there are technologies that we can develop to help people have a good life. If you can do that, spam filters are fine.

Well, I don't think it's an end to loneliness. It is an assistant. That's an antidote. People are still lonely. I have seen elderly ladies with a fake cat next to them stroking the cat, and the cat meows, and the old elderly lady is extremely happy about that. So why not? The fact that we have had a phase in our existence where we think all relationships need to be human to human, well maybe not. Maybe there is more to that, especially because we are getting older. Look at all the baby boomers. Everybody over 65 or over 70. There's a very large group of people that is growing older and older. They come with a new set of problems. Can we solve those as technologists? Can we make a contribution to that? That would be great because building technology is cool, but helping people is much better than that.

Well, I do think the most important thing is that I don't necessarily want to put a timeline on it. I think in the past two years ago or three years ago, I think I said 5 to 10. I think five is now more realistic. But I've seen a number of problems arising that I think we need to be aware of and take action on before that. It's not about... We have the AWS Center for Quantum Computing together with Caltech, an institute about quantum. I see different institutions making massive progress in terms of error correction networks, things like that that actually make me believe that it's no longer a vision. It's an execution now. It becomes an iterative process in improving the technology that we have created step by step until it becomes workable. There is a ton of other problems to solve: how do you program these beasts? How do you do that with something that is not necessarily 100% reliable? There's a ton of things that still have to be solved before we have production quantum machinery. But one of the things these things will do is solve existing problems much faster. There are things that would now take a thousand years in compute that will then take a year or six months or two months. The one thing that they will be able to do really, really fast is decrypt. Anything that is elliptic or the usual RSA encrypted kind of things that are perfectly fine now because it will take way too long to decrypt, will then be decrypted with a snap of your fingers. The reason for the prediction is to put in people's minds that we need to start taking action now if we want to be protected five years from now from the ability to decrypt anything that we have now. All the hyperscalers—Google, Microsoft, Amazon—we have all been building post-quantum cryptography. We've been inserting that into our front end. Unfortunately, we're not the only players in this world. There are lots of people with their own data centers, with their colos, that will have a hard time protecting themselves. So really getting it into people's minds that they have to start taking action now if you want to be safe a number of years from now when these things become reality. The other thing is that you can't just lean back right now because a lot of the data stealing is basically data harvesting. Nobody is interested in encrypted data being retrieved; they'll wait till the machinery is there to be able to decrypt. Your medical data now will be decryptable in a few years from now. I don't necessarily mean script kiddies, it's really state actors or commercial actors who are interested in that kind of data even if it's five years old and might have all the relevant data for them to take action on. That's the scary part.

There are of course different types of encryption. You have lattice-based encryption, so there are types of encryption that you will be safe from this type of decryption, but you need to start installing that now. OpenSSL was 500, 600,000 lines of code. You can imagine that because it had all sorts of different types of encryption units. At some moment, we realized at Amazon that running all of our content with OpenSSL was a major vulnerability. So we reimplemented TLS with a limited set of encryption technologies and just put that in front of S3 for example. First of all, much cheaper, but we made it open source and we made sure that it is fast and inspectable and more importantly written in a way that we can use automatic reasoning to see whether or not we're protecting ourselves. The most important part: you hear more and more about automatic reasoning. People like Byron Cook have been working at Amazon for the past 15 years on these kind of problems, proving to ourselves that we are protecting our customers. That's not a press release that is going out, but there are technologies that you can use to prove to yourself now. The open source project originally called s2n was purely intended to get post-quantum encryption in the hands of everyone. Now I told you, you got scared about the data harvesting. Let me scare you a little bit more. Probably every one of your home devices hasn't been updated for a number of years. They were Linux whatever version, no automatic updates. Your automatic garage door might have a little bit of encryption in it between the key and you. Think about hotel rooms: your key is an encryption encoder. There are so many small parts. Coming back to these house devices, I recently looked at how many devices were on the network in my house, and it was something like 56. There's your Sonos, your garage, your light opener, the thing that keeps track of this. Are all these things up to date? Are all these things protected? Are all these things up to new standards? We have a number of years to get this done, but we do need to start now.

First of all, there are a number of areas. There is just using newer encryption technologies, but you can imagine that some piece of data that you encrypted 15 years ago, you're not going to go through all of your data and re-encrypt it again. But are there gateway possibilities that protect you from that? Are there other ways? Let's use our inventive brains and find solutions to this, because I believe we can, but only if we make it a focus. The focus is not that quantum is five years away. The most important part is that we need to keep ourselves and our businesses safe. If that happened to be quantum or if it happened to be an alien spaceship that lands and suddenly turns out to be able to decrypt all our data, that doesn't really matter. We need to start protecting ourselves.

Do the work. Do the work. The same goes for everything now. Let me take a whole different example. When at Amazon S3 we went from eventual consistency to strong consistency, that was a major change that impacted every piece of technology. How can you make sure that you touched every edge of this particular problem? Because it is either 100% or not. Automatic reasoning plays a role in that: annotations plus other technologies that you have to prove to yourself that what you've built actually does exactly what you want it to do. Definitely automatic reasoning will continue to play a more and more role in giving us a secure feeling that we are on the right path. The same is for example we recently launched Q Developer, the spec-driven IDE with an AI system. We use automatic reasoning there to limit the number of hallucinations that can happen. Not that automatic reasoning knows terribly much more, but it knows that a squared plus b squared equals c squared is probably a triangle, and as such it knows all the triangles in the world. If then this LLM says this is a triangle, automatic reasoning will say no that is not a triangle. That's just a simple example, but we start building more and more technologies to prove to ourselves that we are doing the right thing.