Could AI help us, not replace us?

0:00This message comes from Charles Schwab. When it comes to managing your wealth, Schwab gives you more choices, like full-service wealth management and advice when you need it. You can also invest on your own and trade on Think or Swim. Visit schwab.com to learn more. This is the TED Radio Hour. Each week, groundbreaking TED Talks. Our job now is to dream big. Delivered at TED conferences. To bring about the future we want to see. Around the world. To understand who we are. From those talks, we bring you speakers and ideas that will surprise you. You just don't know what you're going to find. Challenge you.

0:37We truly have to ask ourselves, like, why is it noteworthy? And even change you. I literally feel like I'm a different person. Yes. Do you feel that way? Ideas worth spreading. From TED and NPR. I'm Manoush Zomorodi. On the show today, building AI that puts humans first. Fifteen years ago, millions of people around the world might have had their first interaction with AI when they talked to... My name? It's Siri. Yes, Siri. Apple's voice-controlled virtual assistant was added to new iPhones in 2011. That was a year after Apple CEO Steve Jobs had set his sights on the technology and the people who'd been building it.

1:28He's kind of surprised us. He literally called us on our iPhones at work. And he's like, hi, Steve. Yeah, sure you're Steve. Right. Tom Gruber was Siri's chief technology officer and head of design. Steve Jobs invited Tom and his other two co-founders to his house. I mean, imagine being at Steve Jobs' house in Palo Alto. There's Ansel Adams on the wall. There's a beautiful Carver amp. Like, his taste, this guy. And it's all private and quiet. And for three hours, we talked about what it would be like to build products together. And the reality distortion field completely worked at that point. Like, we were so seduced.

2:07They were made an offer they couldn't refuse and sold their company to Apple, who we should mention is a financial supporter of NPR. Siri was folded into Apple products and debuted on October 4th, 2011. Celebrities were ready to demonstrate how seamlessly this little voice could fit into our everyday lives. Remind me to clean up tomorrow. Okay, I'll remind you. Find me a store that sells organic mushrooms from Arizona. This organic market looks pretty close. Set timer for 14 minutes. OK, 14 minutes and counting. It was one of Jobs' last projects ever. The day after Siri launched is the day Steve Jobs died.

2:50And apparently we're told that he did get to see the demo.

2:56So that was 15 years ago. So at the time, were you worried about the ethical considerations since people did attribute human thinking to Siri? I mean, did you start to think, hmm, this is at the forefront of getting this kind of technology into regular people's hands. We really need to think about what's OK and what's not OK. Oh, yeah, absolutely. If people took Siri too seriously, it means they're out there on the part of the curve where people are inclined to see agency in inanimate objects more than other people. But in general, I saw that, yeah, of course, AI was moving ahead while I was at Apple for the deep learning networks came to power.

3:36And we were then saying, look, we've got to lay down the ethical foundation because this stuff is coming fast. And we need to make sure that we don't use AI to exploit people, that we use it to augment and work with people. In the fall of 2016, big tech companies, including Amazon, Facebook, Google, IBM, and Microsoft, competitors with each other, they came together to form a nonprofit called the Partnership on Artificial Intelligence to Benefit People and Society. Apple soon joined as well. Tom Gruber was their representative. And in 2017, he shared his thoughts on what their goals should be from the TED stage.

4:16I'm here to offer you a new way to think about my field, artificial intelligence. I think the purpose of AI is to empower humans with machine intelligence. As machines get smarter, we get smarter. I call this humanistic AI, artificial intelligence designed to meet human needs by collaborating and augmenting people. That was around the time that you gave your TED Talk in which you introduced the idea of humanistic AI. Can you define what that was in your mind? The idea was really like, where do we stand in the world of the future of AI? And so there was really two paths I saw. One path was, you can sort of say machine intelligence or machine-centric, delighting and celebrating how smart machines were getting relative to humans.

5:10And there were businesses being formed to use that to automate human work. And then the other way of looking at it, which I call humanistic AI, which wasn't just my idea, it was a lot of people had this idea, was that the purpose of AI should be to actually help people do things they're trying to do by either augmenting their intelligence or collaborating with them as an intelligence. And so it turns out that down the road now, that was a watershed because now we see companies that are raising money by the billions with the explicit goal of automating white-collar work. And then we have other companies that are raising billions of dollars and saying that our job is to help solve some of humanity's big problems and make people smarter.

5:57So that's why I wanted to give it a label. And that's why I call it humanistic AI. Yeah, I mean, I have to admit, I was in the audience. I didn't get it at the time. I was like, this guy lives in the future. But the future came pretty fast. 2022, 2023, LLMs like ChachiPT and Claude and Perplexity, regular people started using them. And do you get that sense that people get it now? Oh, I think so now. It is 2026. And in the U.S., AI anxiety is at a fever pitch from fears about the destruction of white-collar work. What is happening? Why is this happening so fast? I value my brain. I value my ability to think.

6:38I don't want to outsource it. To protests in towns where massive data centers are being built. To debates over the government using AI as part of its military operations. Anthropic refusing to allow its powerful AI models to be used for autonomous lethal weapons. For those who have been calling for ethical guidelines all along, this moment represents an opportunity. It's time, they say, for these conversations about how AI is built to move beyond Silicon Valley to the public. Because AI is no longer a futuristic possibility. It's here now. And decisions need to be made. Recently, a declaration of human rights for the AI age was signed by a cross-section of leaders from across the political spectrum, including Ralph Nader and Steve Bannon.

7:30Ensuring AI allows humans to flourish could just be the one idea that everyone, no matter their politics, can agree on. Making sure tech is built to serve humanity has been Tom Gruber's goal for decades. It all started when he was studying psychology and computer science in college in the 1970s. I was a little frustrated that the psychology of the day didn't have very good experimental methods to truly understand the mind. There wasn't really cognitive science yet. Then a computer showed up at my university in the late 70s, so I was able to start programming. And just by reading some papers and so on, I discovered artificial intelligence.

8:11And I said, wow, this is the way to start studying intelligence, by making it. A few years later, Tom began a graduate degree focusing on cognitive science and AI. working to build machines that could mimic the mind at least a little bit. The models that we could build then were nothing like human intelligence. They were just crude approximations. So then we decided, okay, well, how would you augment human intelligence? And one of the ways is to shore up where there's some kind of impairment in human capability. And so the first project I did was to build what I call the communications assistant for people who can't speak, who have like cerebral palsy in this case, but also ALS and folks like that, that have neurological conditions that makes it hard to speak like we're speaking here.

8:58And so we built an AI program that had an LLM, a little language model, maybe a TLM, tiny language model, that actually predicted the next word and the next phrase based on single motor action, like a switch on their muscle. They could actually communicate in sentences. So was that voice then or no, not voice? Yeah, it was voice even then. They all sounded like a strangled Swedish person back then. Something called a Botox voice synthesizer, very primitive and no prosody, but it was voice. I mean, this is pretty common, right, Tom, that like a lot of extremely hard technical problems, They're often built for people who need them the absolute most.

9:45But then once the tech gets going, often developers find a way to get it into the hands of all of us. Is that what happened with Siri? Siri wasn't so much by disability, but it was definitely a kind of handicap. Imagine you're in a car and you're trying to text somebody or you're trying to get the directions. And you are cognitively loaded. Meaning you're concentrating big time. Yeah, you should be concentrating on the road. and if you're distracted, it's dangerous. You're not supposed to touch it. So it's as if you can't use your hands and you're not supposed to look at it. And so how would you use your mobile phone?

10:19Well, you have to have a voice interface. That's a dialogue. And so that's the kind of thing that Siri was built for. Where are we now with AI in our lives? It's not just a voice in the room. We're sort of at a tipping point in terms of understanding how they might change the way we live. Oh, absolutely. We are at a tipping point. I mean, everybody now has access to a real amazing, intelligent partner to help them do things in their lives. It's almost free. It's unbelievable. And it's the same conversational user interface in the front end. So it's really kind of held that style. But the back end has gotten extremely smart.

10:57Like, you know, I mean, I shouldn't get excited. I've seen AF for 40 years, right? And all of a sudden, it's just like a million times different than it has been in the past. as Yuval Harari says, you know, anything that's made of words, AI will own. I mean, who else has read everything in the world and can talk about it? So anyway, I am simultaneously freaked out and excited and scared and unbelievably optimistic. However, we have to act now on the basis that we have superhuman ability to talk a good game, which means persuade people, do things against the will, convincing them and so on, but also helping them in ways that they couldn't do otherwise.

11:38We're not seeing any regulation around AI here in the United States. Where does this go back to your idea of humanistic AI? What sort of guardrails do you think we need to be putting around it right now? Yeah, the humanistic AI framework would say that the objective function of the AI, that is the thing that the AI is optimizing for, should be human benefit, not, say, profit or something else. And human benefit is hard to measure. So there's no easy prescription. So we have to decide if we're going to put guardrails on a thing, we have to solve the engineering and scientific problems of detecting that something the AI is doing is harmful to humans, we have to build theories of human harm and human benefit into our objective functions.

12:30So that's what I think we should be working on. I think it's not just a matter of regulation, it's a matter of the scientific agenda for AI research. We have a choice in how we use this powerful technology. We can choose to use AI to automate and compete with us, or we can use AI to augment and collaborate with us to overcome our cognitive limitations and to help us do what we want to do, only better. In a minute, how ethics could be built into AI going forward. Today on the show, AI that puts humans first. I'm Manoush Zomorodi, and you're listening to the TED Radio Hour from NPR. We'll be right back.

13:26This message comes from WTTW. Join Jeffrey Baer as he tells the story of how railroads shaped Chicago. Riding the Rails with Jeffrey Baer premieres April 13th at 7 p.m. on WTTW. This message comes from Charles Schwab. When it comes to managing your wealth, Schwab gives you more choices. like full-service wealth management and advice when you need it. You can also invest on your own and trade on Think or Swim. Visit schwab.com to learn more. This message comes from Capella University. That spark you feel? That's your drive for more. Capella University's FlexPath Learning Format lets you earn your degree at your pace without putting life on pause.

14:08Learn more at capella.edu. It's the TED Radio Hour from NPR. I'm Manoush Zomorodi. We're spending the hour talking to Tom Gruber, one of the inventors of Siri, about how to build AI that puts humans first. Because it can feel like right now AI is on a collision course with humanity. Recently, for example, the U.S. Department of War struck a deal with OpenAI, while its competitor, Anthropic, fought back over military use of its models. And people responded. Some deleted OpenAI's chat GPT in protest or downloaded Anthropics Claw to show support. Tom believes that if regulation falls short, it will be on consumers to push the market towards AI that feels safer.

14:59Meanwhile, he says there are technologists who are trying right now to shape AI behavior that could prevent future doomsday scenarios. For instance, imagine a scenario where the AI escapes and takes over. Maybe it takes your money, maybe starts running cyberbots and attacks people and so on. And so there are things that are prerequisites to those scenarios, like can it lie to you effectively, masquerade itself as something else, prevent itself from being turned off. These behaviors you can put guardrails around, and people are studying that. Yeah, because I feel like every so often there's some report that comes out that's like, oh, Claude manipulated the developers in some way.

15:40They got to fix that. Oh, yeah. We're just seeing the beginning of this. So we have to come to grips with this. This is one area of just safety that there's no easy answer. But I would hope that we have competition among the big AI model companies in, like, which ones are going to be safest. So, for example, back in the old days, there was a brand allegiance to Volvo because it really went to the great extent to be safe, like the best airbags and the best heat seats and all that sort of thing. So I think we should be able to have AI compete on how safe it is so that people who know could buy the one that is safest.

16:17Do you see that happening anytime soon? Globally, yes. We see a lot of distorted thinking at the national level here and ideologically driven policy and so on. I think hopefully we'll get past that. But the key thing is that we still have a free market system and we still have the freedom to choose among a set of AI products. I mean, you probably know a lot of these mostly guys running these companies. What do they say? Do you bring this up with them? I don't really hear much about safety from Sam Altman. No, you don't hear from him or Elon, but you hear from Demis, from Satya, from Microsoft, definitely from Dario Amato from Anthropic.

16:59So I think the studios that make the foundation models, Demis and Dario, Anthropic and DeepMind, Google, have always been safety conscious and have serious, well-funded teams working on that part of the problem. And that's the sort of healthy, normal thing you would expect from a company that's worth trillions. I mean, when I first started really digging into tech and reporting on all of this 15 years ago, I have to say I was a lot more optimistic and excited about the possibility of building these sort of altruistic systems. But like we've been down this road, Tom, with social media, with the surveillance economy, attention economy.

17:45It seems like any chance for a tech company to take advantage of its users, it will. So what makes you think this could be different? One thing that's different is that the foundation infrastructure of AI, the models like Claude or Gemini and these models, they're super expensive to build, both in money, time, especially talent, which is very rare for now. Anyway, they're hard to build these things. And there's only a few of them, maybe 10. The good news is that they're fairly omnipurpose at that point. And once you have the models, then the rest of the world builds applications on top of them.

18:23So the rest of the world can compete on humanistic applications. I mean, look, you can go out there into the dark web and get the most nasty evil software in the world, or you can go out there and find lovely games or educational things or whatever. You can find everything in the application world. I think we're going to see that with AI too. Is that kind of how Apple created the app ecosystem and you have to adhere to certain privacy standards? Yeah, that's right. Well, that's a very good analogy there. It would be great if there was something like an app store for AI that would do the work of at least minimally establishing whether an application is in the benefit of humans or not.

19:03X is using AI to do evil things right now. It's already happening. So we can't stop. I can't stop them. You can't stop them. Only a government can stop them and they're not stopping them this year. So, okay. But you don't have to use X. You don't have to use Crock. There's a talk that I know you're giving that says that instead of talking about giving up our privacy and self-determinism, we should start thinking about instead of big brother being what tech has sort of become, surveilling all, we need to think about big mother. Yeah, exactly. I tried to figure out what would be the right symbol, and I found, I think, an African elephant is kind of a cool big mother.

19:44It's smart, The matriarchs are the ones that run the herds, are the wisest and smartest and so on. So imagine a female African elephant with her calf and think about her value alignment. She will do anything to protect the calf. And obviously humans too, human mothers. They nurture their children. They teach them right from wrong and truth from falsity. And then they show them skills to survive in the environment they're in. That's what AIs can and should do. Machines can now have access to everything in your life that's digital. And with that, you can build amazingly good recommendations about how to use your attention.

20:24Maybe give you insights on things you could discover or learn from. At the same time, that same data is extremely powerful if it's used in a surveillance economy against your interests to addict you, to make you buy things. This is why it's a real ethical choice. How you optimize the AI is actually a massively important societal choice today. It's not a technical thing that we leave to the engineers or to the boards of directors of companies. It has to be done at the society level. And so I think Big Mother is a way to say, go ahead, have a lot of data. Like, mothers know everything about their kids, but mothers are aligned with their kids' interests.

21:03That was Siri inventor and AI pioneer Tom Gruber. You can watch his TED Talk at TED.com. As Tom said, there are people around the world who are trying to build AI with a big mother mindset. Sure, knowing more about our minds than mere humans can, but using those superpowers to make us better without trying to take advantage of us. Priya Lakhani says she is one of those people. I think AI could be the single most positive technology to impact everybody's lives in an educational context. But it has to be developed responsibly. 15 years ago, Priya was an entrepreneur building schools in India. But back home in the UK, she found that schools were struggling too.

21:5420 % of children cannot read or write well enough in the United Kingdom. him. There were some statistics that were just as poor to do with mathematics. And it was just a real shock to my system. I thought, there's a fundamental problem here. And if we don't fix this in the UK, then we're definitely not going to be able to help the sorts of places where I'm trying to make a difference. And so just out of pure curiosity, I went to schools. What she saw were two big problems. You may have maybe richer resources, more teachers in different ratios. But once you close the door of the classroom, you will often find there is a teacher stood at the front delivering a sort of one-size-fits-all delivery of education.

22:34And then the second problem that was prevalent across all schools was every teacher workload was a massive problem. There were teachers who were stressed, they were anxious, they had far too much to do to the point where they were considering leaving their roles. We were tens of thousands of teachers short in the United Kingdom. You know, how can we have a professional who is trained for this position enjoying their role in inspiring children, imparting knowledge of a subject that they're really passionate about and being able to do so confidently and in a way where they're not exhausted. She realized the outdated tech in classrooms could be the solutions if it could curate a tailored education experience for each child and give teachers instant data about where exactly each student was struggling.

23:21So she looked around at what was on the market. There was one very large company in the US that was talking about this already. It was touted as this massive AI system for education, sort of similar sounding, personalizing learning, reducing workload for teachers. But they were using a type of machine learning that is very common for retail systems. So recommender systems that track user behavior and then recommend what you like, right? So if a student is on the system and they're learning biology, you get a lot more biology. Kind of like TikTok. Right. The problem with that is that that is not going to work in education.

23:58That's because in education, sadly, sometimes you need to give someone what they don't like. And it's often you'll find, particularly with younger people, they don't like the subject that they're struggling in. Right. You can't ignore that. What you need to do is have a system that's a little bit more complicated, give them some things they like to keep them engaged. But you do have to hand them the foundational prerequisite knowledge that they're missing in order to increase their knowledge and skill set in areas where they're struggling. And so it was going to be a more complex system from the outset, and it didn't exist.

24:30And that is where the journey to create Century really started. Century Tech is the educational tech company that Priya founded in 2013. She says the platform's goal is to support overworked teachers, not take their place. There is, of course, a big debate in education right now about the role of AI in teaching. Some believe funding should go straight into schools, not into technology that's trying to make money off of learning. Others think the only way to make education accessible to every child will be with the help of AI. That's the camp that Priya is in. But she says that doesn't mean replacing the hard, even grueling work that happens in the classroom between teachers and their students.

25:14Here's Priya Lakhani on the TED stage. We need to combine artificial intelligence with neuroscientific theory and the learning sciences to learn how every single brain in this room learns. Because if we can fix learning, we can improve outcomes. We can personalize education for every single one of us and provide intelligent insights to teachers to reduce the workload. So 12 years ago, I built a team. They built the technology. It exists. students use it in over 140 countries, I thought it'd be really important to share with you some student feedback that I have on our platform, because it tells us what children's expectations are when they use an AI education partner.

25:55So I get feedback like, I think Century will help me achieve things that I thought were impossible. It's a golden child, right? My life's purpose has been fulfilled. And then these sweet, lovely, innocent children send me messages like this. I don't like this website. It makes me able to do my homework. Wait, I'm being bribed. I will give you£100 ,000. I'm not joking. You just need to give me no work. Give me a button to do the work for me. If I was to go into a classroom as a student to use Century Tech, what would my experience be like? The teacher would walk into the classroom and say, right, everyone, do you want to log into the machine and they could either set them learning material usually they set them a set of questions the teacher has their dashboard open and they can see how the students are performing so information would light up about who is struggling the teacher would then walk around the classroom and be able to make those interventions real time okay so this student why did you answer that question in that way and the teacher's then using their expertise so that's a blended learning environment and the teacher would then utilize that information to then be able to not just stand in front of the class assuming everyone's taking in the knowledge and asking people to put their hands up and spot checking with their knowledge.

27:13You can walk around and have that very sort of targeted intervention with students. Okay, so that's in the classroom. But what about at home? So half of them will use it as teacher sets homework. To Manoush, you've been given this assignment by your teacher. It's on Pythagoras' theorem. You may be a student where before that it has automatically given you some maths work on roots and powers because it knows that you don't understand that and there's no point you doing the Pythagoras work if you don't understand roots and powers in the same way right if you master that work it can then do what's called a smart recommendation and stretch you as well so some people will use it for assignments in that way and then the others will use it in a flipped learning way so teachers will be planning on teaching a lesson they'll say right we would like you to learn the lesson on Century the week before.

28:02Students do that in their own time. Teachers then receive the information of how the students have done. And then that lesson the following week is a far more targeted lesson as to what did they understand, what didn't understand. So 50 % of the platform is about the teacher, because we fundamentally believe that empowering teachers is one of the best ways to improve education. They should not be doing data analysis off spreadsheets in the evenings, which is what many of them sadly have to do, they should be receiving insights from our platform instantly so that they can then go into the class the following week, focus on that particular concept, but not have to just teach it one size fits all.