← Young and Profiting with Hala Taha
Young and Profiting with Hala Taha
Peter Norvig: Transforming AI Into the Ultimate Human Advantage | Artificial Intelligence | AI Vault
00:58:14
Loading summary
Transcript60 lines
- [00:00]
A
You know, I'm not worried about these Terminator scenarios of an AI waking up and saying, I think I'll kill all humans today. I guess I'm more worried about a human waking up and saying, I want to do something bad today. In human centered AI, the goal is to build systems that do the right thing for everyone. And part of it is saying you want to consider everybody involved. And I think when you do that, you don't end up with good results.
- [00:24]
B
We're diving into the world of human centered AI with none other than Peter Norvig. And he's not only authored major AI textbooks and established software tools, but also implemented numerous successful AI systems, including the Google search engine.
- [00:38]
A
I don't want technology that makes me disappear. I want technology that respects me and let me choose how much the machine is going to be doing and how much I'm going to keep control.
- [00:48]
B
A lot of jobs might get replaced by AI. So do you feel like AI is going to generate a lot more entrepreneurs and solopreneurs in the future?
- [00:58]
A
Absolutely.
- [00:59]
B
And I think, hey yeah, fam. We're still continuing with the AI Vault series, and by now I hope you realize that artificial intelligence is no longer some futuristic concept. It's here and it's reshaping everything. For some, AI sparks excitement and limitless possibility. For others, it raises tough questions about ethics, control, and what it means for the future of work. That dilemma is exactly why today's conversation matters. We're diving into the world of human centered AI with none other than Peter Norvig, a true pioneer who's been at the forefront of AI for decades. He's not only authored major AI textbooks and established software tools, but also implemented numerous successful AI systems, including the Google search engine. Peter believes that AI shouldn't be about replacing humans, but about amplifying what we can do, making us more capable, more creative, and more efficient. So get ready, yap fam, because this episode will challenge the way that you think about AI. And by the way, if you're new to the channel and new to Young and Profiting Podcasts, first off, welcome. You're going to love it here. And secondly, make sure you follow and subscribe to the show so you never miss an episode like this. Without further delay, here's my conversation with Peter Norvig. Peter, welcome to Young and Profiting Podcast.
- [02:12]
A
Great to be here. Thanks for having me.
- [02:15]
B
I'm really looking forward to this conversation. I love talking about AI and I can't wait to pick your brain on that topic. But first, I want to talk a little bit about your career journey. So I learned that you worked at some awesome companies like NASA. You actually worked at Google, but it turns out you started in academia. So I'm curious to understand, why did you decide to transition from academia to the corporate world?
- [02:40]
A
Yeah, so I've been in a lot of places. I'm a AI hipster. I was doing it before it was cool. Started out, you know, got interested in it as a subject in the 1980s, and at that time, really, the only way to pursue it was through academics. So got my PhD. And it was sort of the assumption back then that you get a PhD, you're going to go be a professor. There was much less back and forth between academics and industry than there is today. So that's the path I took. But then I started to realize, you know, we didn't quite have the word big data back then, but I saw that that's the way things were going. And I saw as a young assistant professor, I couldn't get the resources I needed. You know, you could write a grant proposal, get a little bit of money, get a couple computers and a couple of grad students, but I really couldn't get the resources to do the kind of big projects I wanted to do. And industry was the only way to do that. So I set out on that path.
- [03:44]
B
Yeah, I love that. It's so funny that you say, like, you were doing AI before people knew it was a thing. For me, it was, like, surprising because I feel like we hear about AI so much, but it turns out that AI has been a thing for decades. Can you talk to us about kind of when you first discovered AI and how long ago that was?
- [04:03]
A
Yeah, so it's definitely been here sort of right from the start, you know. So Alan Turing, one of the founders of the field, writing about it in 1956, sort of foreseeing the chatbots that we have today. But of course, we didn't know how to build them back then, but it was definitely part of the vision of where we might go. So I guess I got interested. I was lucky that I had a high school that at that time had a computer class and also had a class in linguistics. And I took those two classes and took talk to the teachers in the classes and say, hey, it seems like there's some overlap between those two. Can we get computers to understand English? And they said, yeah, that's a great subject, but we can't really teach you that. That's kind of beyond what we know how to do. So you're on your own pursuing that goal. And that's more or less what I've been doing since, with some side trips along the way.
- [05:06]
B
So I always say that skills are never lost, they're really just transferred. So I'm curious to understand what skills do you feel like were an advantage for you in the corporate world that you took from academia?
- [05:21]
A
Yeah, I certainly agree with that idea of transfer. I guess the idea of being able to tackle a complex problem, being able to move into an area that hadn't been done before. And so academia is all about kind of invention of the new. And for industry, it's a mix of you want to make successful products, but sometimes in order to do that, you've got to invent something new, and that's harder to do because you don't know what the demand for it is going to be. There's nothing to compare to. And yet you have to design a path to say, we're going to go ahead and build this and we're going to put it out. Customers are going to have to get used to it because it's not going to be familiar to them.
- [06:15]
B
Yeah. And speaking of building something new, you were responsible for Google Search, and that was a while back when Google really was just not starting off. But there was only 200 employees when you joined them in 2001. So what was it like working for Google back then?
- [06:31]
A
Yeah, that's right. So it was an awesome time. The company was 2001, so it's three years old, 200 people all in one building. I came in and I got the honor of getting to lead the search team for a while, for about five years. So it was a time when, you know, it's not like I invented it. Google search was already there, but they were three years old, and it was really the time when they were trying to ramp up the advertising business. So a lot of the key people who had built the search team had moved over to help build the advertising platform. And so there was an opening and I had just come on board and so I got the opportunity to be a leader of the search team and bring that forward over the next five years. So that was super exciting to be sort of right in the middle of a transformative time in our industry.
- [07:33]
B
Yeah. And I think a lot of my listeners, they don't realize that the Internet was actually much different before Google. Like, Google really changed the way that we use the Internet. Can you help people understand what it was like before Google Search?
- [07:47]
A
Yeah. So I guess there was a couple of Things. First of all, there was directories and lists of sites. And so I remember from the various early days, you know, 1993 or so, and there was a site that was Internet site of the day. And so it was just. You go there and it says, hey, look, here's a new website that you might not have heard of before. And it was like, wow, today 10 new websites joined the web and they picked out a good one and you could sort of keep up through that way. But then a year or two later, that no longer worked because there were thousands of new sites every day, not just a couple. And so Yahoo was one of the first to try to deal with that. And they took this. It's not going to be just one person saying, here's my favorite site today. It's going to be a company organizing the sites into kind of a directory structure. And that worked okay when the web was a little bit bigger, but as it continued to grow, that no longer worked. And then we really needed search rather than manually curated lists of directories and so on. But in the early days, the search systems just weren't that good. I guess we had some experience as a field of doing. It used to be called information retrieval rather than search. And it was sort of. It worked. The techniques we had at the time work for things like libraries. But the problem there was, in a library, everything that was published is like a real book or a real journal article that's already been vetted. And so the quality is all at a pretty high level on the web. That just wasn't true. And so we needed new systems that not only said what's relevant to your query, but also what's the quality of this content? And other companies really hadn't done that. And Google said, we're going to take this really seriously and we're going to work as hard as we can to solve that problem. I think others didn't really see that as an opportunity. There's a story of, in the very early days, people were saying, here's Google, it's rising. Yahoo was far bigger and far better known. Maybe Yahoo should buy Google. And that never happened, in part because the Google founders thought they had something more important. Whereas Yahoo said, oh yeah, search, that's kind of important. We've got a homepage and it's got all this stuff on it, and you got to have search on the homepage, but you also need like daily comics and horoscope. So why would search be more important than horoscope? That's sort of how they felt about it, and Google felt, no, we think search is really, really important and we're going to do an excellent job of it. And so that was something new that, that other people hadn't thought about totally.
- [10:52]
B
And people who are my age and all these listeners who are tuning in. Google is a verb for us. Google is how we use the Internet. But something is changing now with AI. Now, a lot of us, instead of going to Google, we're going to chat gbt, and instead of, you know, putting in a search query and then digging around for information ourselves, we're just asking a question and getting ChatGPT to spit out the information. So how do you think AI is going to change search and the way that we use the Internet?
- [11:25]
A
Yeah, I think there's always been changes and that's always been true. So. So Google's had a dominant position, but there's always lots of places that people go to, you know, so if you wanted breaking news, you went to Twitter. If you wanted a short explanation of something, you might go to TikTok or YouTube to see a video. So it's going to be lots of ways to access this, and we'll see how that changes as AI gets better. Right now, sometimes it works and sometimes it doesn't. So it's a little bit of a frustrating experience, but there certainly seems to be a path to say we can have something that's a much better guide to what's out there. Both in terms of answering a question immediately is one aspect. Rather than saying, I'm going to be pointed to a site that has an answer, I can get the answer right away. And then also kind of guiding you through and maybe summarizing or giving you a whole learning path. So right now you sort of have to make up that path yourself. But I think AI can do a good job of saying, where are you now? What do you know? What do you want to know? And we're going to lead you through that.
- [12:41]
B
Yeah. And AI also is just using the information that was inputted into the system. Right. So it might not have all the information available that you could potentially find on the Internet, is that right?
- [12:52]
A
Yeah, that's certainly true. Right. Depends on what it's trained on. And. And we're at a point right now where the training of these big AI models is very expensive, and so it's harder to keep them up to date. Right. With the Internet search, if something new happens, some new news is there, it's pretty fast of getting that indexed and making it available. But with the large AI models, it's just too expensive to update them instantaneously. And so you miss out on the newest stuff. But that will change over time, and we'll come up with new ways of getting things out faster and faster. When I first started at Google, this transition, when it started, we said, well, we're kind of like a library where you can go to look things up. So it's okay that the library catalog only gets updated once a month. And now that would seem crazy to say. You're only getting information that's a month old. But in the earliest days of Google, that was the case. And then we went to daily, and then hourly, and then even hourly wasn't fast enough. And you had to get faster and faster.
- [14:03]
B
Yeah, it's so interesting how fast technology changes young and profiters. You know, I talk a lot about getting ahead in business, and that means putting yourself out there. But let's be real. Being visible comes with a cost. The moment you exist online in any capacity, even if it's just posting, buying things, joining a group, or having a phone, pieces of your personal information end up in places never meant for them to be. That's the reality of the online world. And it's even worse for entrepreneurs. If you've got a website or a company, for sure your information is out there. And here's the problem. Corporations, called data brokers, collect and sell everything about you. Your address, your phone, even your relative's info. Anybody can buy it. And this exposed data is what fuels the risk of harassment, stalking, or scams. And that's why I use delete me. Deleteme ensures that my data is not online. And not only my data, they make sure that my mom's data, my sister's data, my brother's data, that they're not online either because my career puts them at risk. Delete me's privacy experts remove your data from hundreds of data broker websites. They monitor those sites and they repeat the removal from me all year long. It helps me protect myself from the risks of online attacks and stay focused on building my business. And they can help you too. Yap, bam. Because privacy is power. Get 20% off. Delete me consumer plans. When you go to JoinDeleteMe.com profiting and use promo code Profiting at checkout. That's P R O F I T I N G code. Profiting at checkout. Again, that's joindeleteme.com profiting with code. Profiting at checkout. Yeah, fam. Of course, before I built young and profiting, I had a million doubts. Every time I thought about starting, it just felt so daunting. I didn't know if I was cut out for entrepreneurship, but one day I stopped putting it off and I turned my dream into a reality. Step by step I built my dreams and now I'm running a nearly eight figure company. I bet a lot of you guys out there are thinking about starting a business, but you need a little push. Take this as your sign. It's time to stop thinking about what if and start doing. And one of the easiest ways to do that is to use Shopify. Shopify powers 10% of all US E commerce, from big brands like Gymshark to small business owners Getting started, you don't need a big team because Shopify handles everything from web design and inventory to customer service and shipping. Their marketing tools help you find and keep customers and their point of sale connects online and in person sales. Shopify even helps you sell globally in over 150 countries. With 99.99% uptime and the best converting checkout on the planet, you'll never miss a sale. Turn those what ifs into and keep giving those big dreams their best shot with Shopify. Sign up for your $1 per month trial and start selling today at shopify.com profiting go to shopify.com profiting again. That's shopify.com profiting. YAP Gang here's the truth in business, cash flow isn't just numbers on a spreadsheet. It's the fuel behind every smart move that you make. Cash flow powers your business. At Yap Media, we've trusted Intuit products for years now and I can tell you that intuit QuickBooks is essential when it comes to our cash flow and ensuring that we have strong cash flow. They have powerful money tools that are built right into the platform that help you get paid faster and pay your bills smarter. With QuickBooks Money Solutions, you can say goodbye to manual bill entry and reduce manual work by half. No more juggling passwords and nine different spreadsheets. QuickBooks does it all for you. QuickBook money tools bringing your money and your books together in one platform with real time insights into your cash flow. I'm logging onto QuickBooks every single day, checking out the health of my business. So with QuickBooks you know what's coming in, what's going out, and when you should act to transform your cash flow and your business. Check out QuickBooks Money Tools today. Learn more@quickbooks.com money Again, that's quickbooks.com money terms apply. Money Movement services are provided by Intuit Payments Incorporated, licensed as a money transmitter by the New York State Department of Financial Services. I know that you wrote a book about AI with some Stuart Russell. In 1995, you wrote a textbook, the first edition of Artificial Intelligence. How has AI changed since you wrote that textbook?
- [18:31]
A
Yes, so we did the first edition in 95, and we're up to the fourth edition, which we did a year or two ago. And there definitely are changes. And first of all, I think we did the book because we saw changes Even back in 1995, where in the earlier days, in the 80s and the start of the 90s, the sort of dominant form of AI was called expert system. And what that meant was you build a system by going out and interviewing an expert, say an expert doctor, and ask them, in this situation with this patient, what would you do? And then you try to build a system that would duplicate what the doctor said. And it was all built by hand, you know, programmers sitting down, trying to understand what the doctor said and trying to encode that into rules that they would write into the system. And it worked to some extent, but it was very brittle and it just often failed to handle problems that were just slightly outside of what it had anticipated. So in the 1990s, there was a big switch away from this expert system hand coded approach towards machine learning approaches, where we said, rather than telling the system how to do it, just show it lots of examples and let it learn by itself. And so we felt like the existing books had missed that change. We wanted to write a book about it, so we did that. But of course, things continue to change. And so I guess, what can I say about what's changed over the four editions? I guess one was at the start we felt like, well, AI, this is part of computer science, and computer science is about algorithms. So we're going to show you a bunch of cool algorithms. And we did that. And then in the second edition, I think we felt more like, okay, you still got to know all the cool algorithms, but if you had a choice, you're probably better off getting better data rather than getting better algorithms. So we're going to focus a lot more on what the data is. And that continued to be more true in the third edition. And now I feel like, okay, now we've got plenty of data, we've got plenty of algorithms, you still have to know about them. But really the key to future progress is neither of those. The key is deciding what is it that you want, what is it that you're trying to build. So we have a great system that says if you give me a bunch of data, I've got an algorithm that can optimize some objective that you're shooting for, but you got to tell me what the objective is, what is it that you're trying to do? And for some tasks, that's easy. If I'm playing chess, it's better to win than to lose. But in other tasks, that's the whole problem. And so we look at things like we have these systems that help judges make decisions for parole, who gets out on parole and who doesn't. And you want to parole somebody if they're going to behave well and you want to not parole them if you think they're going to recommit a crime. But of course these systems aren't going to be perfect, they're going to make mistakes. And so the question you have to answer is, what's the trade off between those mistakes? How many innocent people should we jail to prevent one guilty person from getting away? And so there's this trade off. You're going to make false positives and false negatives and what's one one worth against another? And you know, we've, we've, even before there was AI or any kind of automation, we've had these kinds of discussions in our societies. Going back to judge Blackmon in England more than a century ago, who said it's better that 10 guilty men go free than that one innocent man be jailed. Now I don't think he meant it that literally, like, you know, 10's the boundary and 9's okay, and 11 would be bad. But with today's AI systems, you have to specify that, right? So you have to build the system and there's got to be an exact number in there of saying what is the trade off point? And we're not very good at understanding how to do that. Right. So we built a software industry and we have 50 years of experience in building debugging tools and so on. So we're pretty good at making reliable software. There are still every week you'll see some kind of bug or something, but we're getting pretty good at that. But we don't have a history of tools for saying how do we specify the right objective, what are the trade offs, how important is it to avoid this mistake versus that mistake? And so we're kind of going by the seat of our pants and trying to figure that out. And so I think that's where a lot of the focus is now is how do you Decide what you really want.
- [23:33]
B
I want to dig into this a bit because I think it ties in with this idea or the fact that AI is not yet in all instances at human level intelligence. Right. And that's not always a goal. I read of your work where you said human level intelligence is really not always the goal when it comes to AI. So I want to read you a quote from Dr. Fei, Fei Li, who came on the podcast, episode 285. She's the co director of the Human Centered AI Institute, which you're also a fellow. And it was an awesome conversation. And she said the most advanced computer AI algorithm will still play a good chess move when the room is on fire. So she's trying to explain that, like, AI doesn't have like human level common sense, you know, it's still going to play a chess move even when the room is on fire. So let's start here. How do you feel AI stacks up right now against the human brain as a tool?
- [24:30]
A
Yeah. So that's great. And Fei Fei is awesome. And I've heard many of her talks where she makes great points like that. Let's see. So I guess I would try to avoid trying to make metrics that are one dimensional. Right. How does AI compare to humans? For a couple reasons. One is, I don't want to say the purpose of AI is to replace humans. Right. We already know how to make human intelligences. My wife and I did it twice the old fashioned way. That was awesome. It worked out great. So instead of saying, can we make an AI that replaces a human? We should say, what kind of tools can we make so that humans and machines together will be more powerful. Right. What's the right tool? And so we don't want a tool that replaces a human. We want a tool that kind of fills in the missing pieces. And we've always had that. There's always been a mix of subhuman and superhuman performance. So my calculator is much better at me at dividing 10 digit integers. So I rely on it rather than trying to work it out myself. And I think we'll see more of that of saying, what are the right tools for people to use now in terms of this generality versus general AI versus narrow AI, I think that's really important. And so there's multiple dimensions we want to measure. So we want to focus on both generality and performance. So how good are these machines and how general are they? So yes, we have fantastic chess playing programs that are better than the best human chess players. And recently it's also true in Go and we see sort of every week it's true at something else. But we haven't done quite as well at making them good at being general. So we have these large language models, the ChatGPT and Gemini and so on, and they're good at being general, but they're not completely competent yet at doing that. So they'll surprise you in both ways. They'll give you an amazingly good answer one time and then the next time they'll give you an amazingly bad answer. So they're not reliable yet at being general. And then we have incredible tools that are narrow. And so we're kind of looking at this frontier of how can we make things both more perform better and more general. And so I think, you know, we'll get to the point where we'll say, here's an AI and it can make a chess move and it can also operate in the world. But right now we separate those two things out and we say we're going to have the chess program that only plays chess and then we're going to have the large language models. And it won't be as good at chess, but it will be good at some aspects of figuring out what to do in unusual situations.
- [27:51]
B
Could you give us some concrete examples of AI that we might want superior human level intelligence versus AI that we wouldn't want to have human level intelligence with?
- [28:07]
A
So I guess it's always better for it to be better, but sometimes we need that and sometimes we don't. Sometimes we want to make our own decisions. And I guess part of that is I see too much of people saying AI is going to be one dimensional and automation is going to be one dimensional and the more the better. And I think that's the mistake that I'm worried about. And there's a great diagram from the Society of Automotive Engineers of level of self driving cars. And they define that as five levels of self driving. And they did a great job of that. And that's really useful. And now you can say, where is Waymo or Tesla? Are they at level 2 or level 3 or what level are they at? And that was useful, but the diagram they used to accompany those levels was worrying to me because they've got this diagram and at level one they have this icon of a person behind the car holding onto the steering wheel. Then when you get up to level five, that person has disappeared and they've just become a dot like outline, right? And so it's like, I don't want technology that makes me disappear. I want technology that respects me and I don't want this trade off to be one dimensional of if I get more automation then I disappear more. I'd rather have it be two dimensional and let me choose so sometimes I might want to say I've got a self driving car and I trust it. I just want to go to sleep, it should take over completely but sometimes I might want to say it can do all the hard parts but I still want to be in control. I want to be able to say oh let's turn down that street or go faster or go slower or let's make an unscheduled stop. So I don't want to say just because I have automation that I've given up control. I want me to come first and let me make the choice of how much the machine is going to be doing and how much I'm going to keep control.
- [30:17]
B
Hey YAP gang. As a CEO I'm always looking for ways to streamline our creative output. At yap, I need one place to handle everything. Not just our website, but the entire look and feel of our brand. When I first heard about Framer, I thought, oh, just another website builder. But I was totally wrong. And I love being wrong when the alternative is this good. Framer already built the fastest way to publish beautiful production ready websites and it's now redefining how we design for the web with the recent launch of Design Pages, a free canvas based design tool. Framer is more than a site builder, it's a true all in one design platform. From social assets to campaign visuals to vectors and icons, all the way to a live site. Framer is where ideas go live, start to finish. Ready to design, iterate and publish all in one tool. Start creating for free@framer.com design and use code profiting for a free month of framer pro. That's framer.com design and use promo code profiting again, that's framer.com design promo code Profiting Rules and restrictions apply. Young and Profiters I can't believe we're already wrapping up the end of the year. It's about to be 2026. This past year has been so eventful. I've traveled so much for work and I squeezed in some time for fun too. But lately I've been soaking up life here in Austin. Great food, great vibes, amazing people. I've really tried to settle in. That said, this year is not over yet and I still have a few more trips on my calendar. One last trip to New Jersey to Visit my family and maybe a beach getaway to Aruba if I can sneak it in. The best part is, even when I'm away, my place doesn't have to sit empty. Somebody else can enjoy it too. With Airbnb's co host, network hosting is easier than ever. You can partner with a local co host who handles all the day to day management, from guest communication to design and styling and on site support. So this day runs smoothly even when you're away. Think of them as your local co host superhero who knows the ins and outs of creating amazing guest experiences. Turn your space into an opportunity without adding more to your plate. Find yourself a co host@airbnb.com host foreigners can we take a moment and talk about something essential for anybody starting a business? I'm talking about your business identity. When you're building an empire, you've got to think about what the world sees and what stays private inside your business. Things like legal documents, security, and especially privacy. I get it. Starting a business can feel really overwhelming with all the paperwork. But imagine building your complete business identity in just 10 clicks and 10 minutes. That's where Northwest Registered Agent enters the equation with reliable, straightforward support. They've been helping founders for nearly 30 years and have over 1500 corporate guides. These are real experts who know their stuff and can help you. When I was first figuring things out with the app media, I would have loved something like Northwest Registered agent. For just $39 plus state fees, they set you up with everything you need to start a business. An llc, domain name, business email, local phone number, business address, and even a registered agent. Plus, they help you protect your identity by using their address instead of yours on state documents. That commitment to privacy is why I feel confident endorsing them. They treat your data with respect, don't wait, protect your privacy, build your brand, and get your complete business identity in just 10 clicks and 10 minutes. Visit northwestregisteredagent.com paid yap that's Paid Yap and start building something amazing. Get more with Northwest registered agent@northwestregisteredagent.com paid yap so, like Dr. Lee, you are an advocate for human centered AI. Can you help us understand what that is?
- [33:59]
A
Yeah, so a couple of things. So first of all, you know, I'm, I'm essentially a software engineer or a programmer at heart. And so I look at what are the definitions of these various things. And so software engineering is building systems that do the right thing, but artificial intelligence is also building systems that do the right thing. So what's the difference. And I think the difference is that the enemy in software engineering is complexity. We have these programs with the millions of lines, we have to get them right. And the enemy in AI is uncertainty. We don't know what the right answer is. And then in human centered AI, the goal is to build systems that do the right thing for everyone and do that fairly. And so that kind of changes how you build these systems. And part of it is saying you want to consider everybody involved, so you want to consider the users of your system, but you also want to consider the stakeholders and the effect on society as a whole. So we go back to us talking about this aid for judges in deciding who gets parole. If you took a normal software engineering approach, you'd say, well, who's the user? Okay, it's this judge. So I want to make this program be great for them. I want a pretty display with graphs and charts and so on, and numbers and figures and diagrams so that they can understand everything about the case and make a good decision. And yes, you want that in human centered AI, but human centered AI says, well, you also got to consider the other stakeholders. So what's the effect on the defendant and their family? What's the effect on past victims and potential future victims and their family? What's the effect on society as a whole of mass incarceration or discrimination of various kinds? And so you're not just serving one user, you're serving all these different constituencies. I mentioned this idea of varying autonomy and control. So not having to give up control if you have more automation. I think there's the aspect that it's multidisciplinary and multicultural. I think too often you see companies say, okay, I want to build a system, so the engineers will build it and get it working. And then afterwards we'll kind of tack on this extra stuff to make it look better or make it more fair or less biased and so on. And I think when you do that, you don't end up with good results. You've got to really bring in all these people right from the start, both in terms of being aware of what it means to build a system like this, and then also that, you know, as we were saying before, a lot of these problems is deciding what is it that we want, what is it that we're trying to optimize. And different people have different opinions on that. And so if you get a homogeneous group of engineers, they might all think the same thing. And they say, great, we're agreed, we must have the right answer. But then you go a little bit broader to other people from other parts of society and they might say, no, you forgot about this other aspect. You're trying to optimize this one thing, but that doesn't work for us. So you've got to bring those people in right from the start to understand who all your potential users are and what's fair for all of them.
- [37:40]
B
So one of the things that worries me is that we live in a capitalistic world. So while it's nice to think that people are going to have like a human centered approach with AI, I do feel like at the end of the day, companies are going to do whatever is going to impact their bottom line the best, like most positively. Right. So what are the ways that you think that there'll be some guardrails against not using AI in a human centered way?
- [38:08]
A
Yeah, so that's certainly an issue with capitalism, not specifically for AI at all. Right. So that's kind of across the board. And so what do we have to combat that? So part of it is regulations of various kinds so governments can set in and get rules. Part of that is pressure from the customers saying, here's the kind of company we want, here's the kind of products we want. And part of that would be competition of saying, you know, if you build a system that doesn't respect something that users want, somebody else will build one that's better. And I think we're in this kind of wild west period now where we don't quite know what the bounds are going to be. And so, you know, there's so many of these sets of AI principles now. So all the big companies have have their own sets. I helped put together the Google one. Various countries have had legislation or sets of principles. The White House put out their set of AI principles a couple months ago. The professional societies like the association of Computing Machinery has theirs. I actually joined an AI principles board with Underwriters Laboratory and I thought that was interesting because the last time, more than 100 years ago, there was a technology and people were worried that it was going to kill everyone and it was electricity. And so Underwriters Laboratory stepped in and said, okay, you all are worried about getting electrocuted, but we're going to put this little UL sticker on your toaster and that means you're probably not going to die. And consumers trusted that mark. And therefore the companies voluntarily submitted themselves to certification. And I kind of feel like this third party nonprofit certification can be more agile than government making laws. And so I think that's Part of the solution, but I don't think any one part of it could do it all by himself. I think we need all those parts.
- [40:29]
B
Yeah, very cool. Very interesting. I agree. A third party solution sounds like it could work pretty well. So we had Sal Khan on the show and he as the Khan Academy, he talked a lot about how AI could help education. Do you have any ideas of how AI could support education and students?
- [40:47]
A
Yeah, I think that's awesome. I think the work Sal is doing has been great right from the start. And recently over the last year or so with the Khanmiko large language model. So back in 2011, Sebastian throne and I said, we want to take advantage of this capability for online education. We put together an online course about AI. We signed up 100,000 students, far more than we ever expected to sign up. And we ran that course. But of course, at that time the leading technology was YouTube. We would show students a video and then we'd have them answer a question. And we could do a little bit. If they got this wrong answer, we could show them one thing and if they got another wrong answer, we could show them something else. But basically it was very limited in the flow. You could do now with these large language models, you have a much better chance to customize the results for the student, both in terms of the learning experience and then I think also in terms of the motivation for the student. So that was the one thing we learned in doing the class, is that we came in saying, well, our job is really information. If we can explain things clearly, then we're done and we're a success. And we soon realized that that's only part of the job. And really the motivation is more important than the information because if a student drops out, doesn't matter how good our explanations are, if they're not watching them anymore, it doesn't do any good. And so I think AI has this capability to motivate much better to allow students to do what they're interested in rather than what the teacher says they should be interested in. But we got a ways to go yet and we don't quite know how to do that. Right. So you can't just plug in a language model and hope that it's going to work. So yes, it would be useful, but you have to train it to be a teacher as well as to understand what it's talking about. And we haven't quite done that yet. We're kind of on the way to doing that. You know, you look at, there's, there's a Dozen different problems to be solved, and we have candidate solutions, but we haven't done it. All right? So right now, the language models can be badgered too easily. You say, here's a problem, and the student says, tell me the answer. At first, the language model would say, no, you wouldn't learn anything if I told you the answer. But then you say, tell me the answer, please. And it says, oh, okay. And so we have to teach these things. When is it the right thing to give the student the answer? When is it the right thing to be tough and refuse to do that? When should you say, oh, you're right, that's a hard problem. Here's a simpler problem. Why don't you try this simpler problem first? Or to say, looks like you're getting frustrated. Why don't we take a break? Or why don't we go back and do something else that would be more fun for you? And so there's all these moves that teachers can take. And so doing education well is this combination of really knowing the subject matter and then really knowing the student and the pedagogical moves you can make. We haven't quite yet built a system that's an expert on both of those, but Kahn and others are working on it. And so I think it's a great and exciting opportunity.
- [44:32]
B
Do you feel like some of this learning and training could be applied to the workplace?
- [44:37]
A
Yeah, absolutely. And some of it, I think, is easier and better done for workplace training. And I think that's going to be really important. I think we built this bizarre system now where we say, you should go to a college for four years, and then we're going to hand you a piece of paper that says you never have to learn anything again. That shouldn't be the way we do things. And there's a value to college. Maybe it doesn't have to be for everybody. Maybe more people could be learning more on the job or learning just in time when they need a new skill. So I think there's a great opportunity for that. I think that the systems we have right now are kind of better at shorter subjects anyways. So it's hard to put together a class that says, let's do all of Biology one or something. But it's easier to say, why don't you get trained on this specific workplace thing how to operate this machine or how to operate this software and so on. So in some sense, we're better at that kind of training than we are at the traditional schooling. So, yeah, there's definitely a big opportunity There. The thing that mitigates against it is we could spend a lot of investment on making the perfect Biology 1 class, because there's going to be millions of students that take it. But for some of this on the job training, maybe I'm in a small company and we do things a specific way, and there might be only five people that need to be trained on it. So right now, it's not really cost effective to say, can I build a system that will do that training? But that's one of the goals to say, can we make it easier for somebody who's not an expert programmer, not an AI expert, to say, here's some topic I want to teach and I should be able to go ahead and teach that. And I think that's something that's oddly missing from our sort of standard playbook. Right? So you look at, you know, we have these office suites, and what do they give you? They give you word processing and spreadsheets and PowerPoint presentations, and sure, that's great. Those are three things that I want. But I think a lot of people want this. I want to be able to train somebody on a specific topic more than they want spreadsheets. But we don't have that yet. But, you know, maybe someday we will. Maybe that'll be a standard tool that would be available to everyone.
- [47:20]
B
So this conversation made me realize that there really is no better time to be an entrepreneur, because as we were talking about, a lot of jobs might get replaced by AI. And when you're an entrepreneur, when you own the business, you're sort of in control of all those decisions, and you're the one who might end up benefiting from the cost savings of replacing a human with AI. So. So do you feel like AI is going to generate a lot more entrepreneurs than solopreneurs in the future?
- [47:50]
A
Absolutely. And I think it's a combination. So I think AI is a big part of it. I think the Internet and access to data was part of it. The cloud computing was a big part of it. Right. So it used to be, you know, if you were a software engineer, the hardest part was raising money because you had to buy a lot of computers just to get started. Now all you need is a laptop and a Starbucks card and you can sit there and start going and then rent out the cloud computing resources as you need them and pay as you go. And so I think AI will have a similar type of effect. You can now start doing things much more quickly. You can prototype something and go to a release product much faster and It'll also make it more widely available. Right. So, you know, there's a lot of, you know, so I live in Silicon Valley, so I see all these notices going around of saying, looking for a technical co founder, right? So there's lots of people that say, well, I have an idea, but I'm not enough of a programmer to do it, so I need somebody else to help me do it. I think in the future a lot of those people will be able to do it themselves. Right. So I had a great example of a friend who's a biologist and he said, you know, I'm not a programmer. I can pull some data out of a spreadsheet and make a chart, but I can't do much more than that. But I study bird migrations and I always wanted to have like this interactive map of where the birds are going and play with that. And he said, and I knew a real programmer could do it, but it was way beyond me. But then I heard about this copilot and I start playing around with it and I built the app by myself. And so I think we'll see a lot more of that of people that are non technical or semi technical who previously thought, here's something that's way beyond what I could ever do. I need to find somebody else to do it now I can do it myself.
- [49:59]
B
Yeah, I totally agree. And we're seeing it first with the arts. For example, now you can use Dolly and be a graphic designer, you can use ChatGPT and be a writer. So so many of the marketing things are already being outsourced by AI. It's only a matter of time where some of these more difficult things, like creating an app like you were saying, is going to be able to be done with AI.
- [50:22]
A
Absolutely.
- [50:24]
B
Cool. So what are the ways that you advise that entrepreneurs use AI in the workplace right now?
- [50:33]
A
I guess so. You know, you could help build prototype systems like that. You can do research, you can ask, you know, give me a summary of this topic. What are the important things? What do I need to know? As you said, creating artwork and so on. If that's not a skill you have, they can definitely help you do that. Looking for things that you don't know is useful. And so I think just being aware of what the possibilities are and having that as one of the things that you can call upon, it's not going to solve everything for you, but it just makes everything go a little bit faster.
- [51:16]
B
Yeah. Do you think that AI is going to help accelerate income inequality?
- [51:25]
A
I think it's Kind of mixed. So any kind of software, any kind of goods with zero marginal cost tends to concentrate wealth in the hands of a few. And so that's definitely something to be worried about. With AI. We also have this aspect that the very largest models are big and expensive. They require big capital investments. And if you'd asked me two years ago, I would have said, oh, you know, all the AI is going to migrate to the big cloud providers because they're going to be the only ones that can build these large state of the art models. But I think we're already going past that. Right. So we're now seeing these much smaller open source models that are almost as good and that, you know, don't impose a barrier of huge upfront costs. So I think there's an opportunity, you know, yes, the big companies are going to get bigger because of this, but I think there's also this opportunity for the small opportunistic entrepreneur to say, here's an opening and I can move much faster than I could before and, and I can build something and get it done and then have that available. So that's part of it then. The other part is, well, what about people who aren't entrepreneurs? And we've seen some encouraging research that says AI right now does alleviate inequality. And so there have been studies looking at, well, you bring AI assistance into like a call center and it helps the less skilled people more than the more skilled people, which kind of makes sense, right? Sort of. The people who are more skilled, they already know all the answers. And the people that were less skilled, it brings them up almost to the same level. And so I think that's encouraging because that means there's going to be a lot of people who are able to upskill what they do and they'll get higher paying jobs. Right. They're not going to found their own company, but they're going to do better because they're going to have better skills.
- [53:50]
B
Yeah, makes a lot of sense. Okay, so as we close out this interview, let's talk about the future a bit. What scares you the most about AI right now?
- [54:01]
A
Yeah, so, you know, I'm not worried about these Terminator scenarios of an AI waking up and saying, I think I'll kill all humans today. So what am I worried about? Well, I guess I'm more worried about a human waking up and saying, I want to do something bad today. And so what could that be? Well, misinformation. We've seen a lot of that.
- [54:29]
B
And.
- [54:30]
A
I think it's mixed of how big an effect AI will have on that. It's already pretty easy to go out and, and hire somebody to create fake news and promulgate it. And the hard part really is getting it to be popular, not to create it in the first place. So in some sense, maybe AI doesn't make that much difference. It's still just as hard to get it out. And maybe AI can fight against that misinformation. So I think the jury is still out on that. But if you did get to the point where an AI could create, knew enough about an individual user to say, I'm going to create the fake news that's going to be effective specifically for you, that would be really worrying. And we're not there yet, but that's something to worry about. I worry about the future of warfare. So you're seeing these things today. You just saw a tiny little personal size drone shot down a Russian helicopter. So we've had half a century or so of mostly a stalemate of saying the big countries have the power to impose themselves on the others, but none of them are really going to unilaterally do it on a large way. And we have smaller regional conflicts now. We may be transitioning into a world where we say the power is not just in the big countries, it's in lots of smaller groups. And that becomes a more volatile situation. And so there could be more of these smaller regional conflicts and more worries for civilians that get caught up in it. So I'm worried about that as well. And then, you know, like you said, the income inequality I think is a big issue.
- [56:30]
B
Well, let's end on a positive note, I guess. What excites you the most about AI?
- [56:36]
A
So a big part of it is this opportunity for education. That's where I spent some of my time and I'm really interested in that now. So I think that can make things better for everyone, just making everyone more powerful, more able to do their job, able to get a better job. So that's exciting. I think applications in healthcare are a great opportunity and I got involved a little bit in trying to have better digital health records and that really didn't go so far, mostly because of bureaucracy and so on. But I think we have the opportunity now to do a much better job, to invent new treatments, new drugs. You've seen things like AlphaFold figures out, here's how every protein works. And it used to be you could get a PhD for figuring out how one protein worked. And AlphaFold said, I did them all. So I think this will lead to Drug discovery lead to healthier lives, longevity and so on. So that's a really exciting application.
- [57:47]
B
Yeah, it's so interesting to me that AI can do so much good and then there's also such a risk of it doing so much bad. But I feel like any good technology kind of brings that risk along with it.
- [57:59]
A
Yeah, I think that's always true. Right. If it's a powerful technology, it can do good or bad specific, you know, especially if there are good and bad people trying to harness that way. And some of it is intentional bad uses and some of it is unintentional. Right. So internal combustion engines did amazing things in terms of distributing food worldwide and making that be available, making transportation be available. But there are also these unintended side effects of pollution and global warming and maybe some bad effects on the structure of cities and so on. And, you know, we would be a lot better off if, you know, when cars were first starting to roll out in 1900, if somebody said, let's think about these long term effects. So. So I guess I'm optimistic that there are people now thinking about these effects for AI as we're just starting to roll it out, so maybe we'll have a better outcome.
- [59:01]
B
Yeah, I hope so. Well, Peter, thank you so much for joining the show. I end my show with two questions that I ask all of my guests. What is one actionable thing our young and profits can do today to become more profitable tomorrow?
- [59:16]
A
I guess keep your eye on what it is that people want, you know, So I said the problem in AI is figuring out what we want. I'd worked some with people at Y Combinator and I still have this T shirt that says on the back, make something people want. And very simple advice to entrepreneurs, but sometimes missed. And so I think that's true generally and I think AI can help us do that.
- [59:48]
B
Yeah, it's so true. The number one reason why entrepreneurs and startups fail is because there's no market demand. So make something that people want and what is your secret to profiting in life? And this can go beyond today's episode topic.
- [60:05]
A
I guess. You know, keep around the people you like and be kind to everybody.
- [60:18]
B
Love that. Where can everybody learn more about you and everything that you do?
- [60:24]
A
You can look for me@norvig.com or on LinkedIn or thanks to Google, I'm easy to find.
- [60:35]
B
Awesome. I'll stick all your links in the show notes. Peter, thank you so much for joining us.
- [60:39]
A
Great to join you, Hala.
- [60:41]
B
It was so great to connect with Peter and Dig into a perspective on AI that feels grounded and deeply human. Peter spent decades at the center of technology, and what stands out to me is how committed he is to creating tools and that elevate people rather than overwhelm them. His lens on human centered AI is a powerful framework for anybody building, leading, or innovating in this new era. Here's a couple takeaways from this conversation. First, clarity matters more than complexity. Peter reminded us that great technology doesn't start with bigger models or fancier algorithms. It starts with defining the right goal. Entrepreneurs who know exactly what outcome they're aiming for will use AI more effectively than those who are just chasing the trends. Next, human judgment remains irreplaceable even as AI becomes more capable. Its value depends on the choices we make. Peter emphasized that people will set the objective, interpret the results, and decide what good looks like from AI. For founders, that means leaning into your taste, your creativity, and your intuition. AI can accelerate your work, but it can't choose your mission and it can't replace your human intelligence. Finally, learning must become a lifelong habit. Peter offered us a refreshing view of education as something that is continuous, adaptive, and personalized. Entrepreneurs that have that mindset in this new era, those who stay curious, update their beliefs quickly and experiment often will thrive. When you stay adaptable, you stay ahead. All right, Yap fam, if this conversation got your wheels turning, I want to hear from you. So take a second and share your thoughts on Peter's human center approach to AI. Let's keep this dialogue going and build a community that uses AI with intention. And if you want to follow me on social media, you can find me at Yap with Hala on Instagram or LinkedIn. Just search for my name. It's Hala Taha. All right, Yap fam. This is your host, Hala Taha, AKA the podcast Princess, signing off.