Журналистка Джоанна Стерн, автор новой книги «Я не робот», рассказала в интервью подкасту Decoder о своем опыте жизни с искусственным интеллектом, о будущем робототехники и о запуске собственного медиа-проекта New Things. Она провела год, интегрируя ИИ во все сферы своей жизни, и теперь делится выводами о текущем состоянии технологий.

Год с ИИ: ожидания и реальность

Стерн признает, что многие из самых разрекламированных ИИ-устройств, особенно гуманоидные роботы, далеки от совершенства и, возможно, не будут готовы к массовому использованию еще долгое время. «Мир моделей еще не существует», — отмечает она, указывая на огромный разрыв между маркетинговыми обещаниями и реальными возможностями.

Однако, несмотря на разочарования, год с ИИ сделал Стерн более оптимистичной в отношении некоторых направлений. Она считает, что носимые устройства с ИИ могут стать тем самым «killer app», который оправдает компромиссы, связанные с развитием технологий. Примером служит браслет Bee, который, несмотря на прототипный статус, показал свою полезность в суммировании информации и составлении списков дел.

Роботы в доме и «искусственный достаточный интеллект»

В книге Стерн описывает, как роботы, включая кухонного робота Posha, стали частью ее быта. Дети журналистки даже начали подражать роботам-уборщикам после воскресных ужинов. Это навело ее на мысль о концепции «искусственного достаточного интеллекта» (artificial enough intelligence, AEI), подчеркивая, что для многих задач уже существующие инструменты «достаточно хороши» и нуждаются лишь в лучшем применении.

Тем не менее, Стерн предостерегает от чрезмерного доверия к ИИ, особенно в контексте детей. Она обеспокоена тем, как легко дети могут попасть под влияние чат-ботов и цифровых «партнеров», особенно в свете отсутствия четких правил и регулирования в этой сфере. «Мы должны сами установить правила», — призывает она.

Новый медиа-проект и партнерство с NBC

После ухода из The Wall Street Journal Джоанна Стерн основала собственную медиакомпанию New Things, которая фокусируется на создании контента для широкой аудитории. Важным шагом стало партнерство с NBC News, которое позволит охватить новую аудиторию и интегрировать контент в традиционные медиа. Стерн подчеркивает, что ее цель — не только информировать техно-энтузиастов, но и помогать обычным людям ориентироваться в мире технологий.

«Я не хочу быть замутненной алгоритмом», — заявляет Стерн, объясняя свое решение сосредоточиться на создании контента, который ей интересен, а не только на том, что популярно на YouTube. Она стремится к балансу между созданием вирусного контента и глубокими, важными историями.

Будущее ИИ и «дилемма выбора»

Стерн также коснулась темы конфиденциальности и «дилеммы выбора», связанной с использованием ИИ. Она признает, что удобство, которое предлагают новые технологии, часто сопряжено с компромиссами в области приватности. «Удобство против стоимости», — формулирует она главную дилемму, отмечая, что компании часто делают ставку на то, что пользователи примут эту стоимость ради удобства.

В заключение, Джоанна Стерн выразила надежду, что ее книга поможет людям лучше понять текущее состояние ИИ, его потенциал и риски, а также задуматься о том, как мы хотим использовать эти технологии в будущем.

My guest today is longtime friend of the show Joanna Stern. You all know Joanna: she is the former senior personal technology columnist for The Wall Street Journal, a former Decoder guest host, one of my cofounders here at The Verge, and also just one of my very closest friends.

I mention that because Joanna just left that lofty perch at The Journal to start her own media company called New Things. She’s starting with her new book about AI, called I Am Not a Robot, which is out this week on May 12th. 

You’ll hear us reference the fact that she and I have been talking about her big move to go independent for ages now — it’s something she’s wanted to do and wrestled with for years, and she has a long list of interesting reasons about why now is the time. She’s also structured her new venture in partnership with NBC to keep her in front of a big mainstream audience.

It was important that I prove to Joanna that I actually read her book, which is really quite good. She spent a full year allowing AI into every part of her life and has more of a sense of where this technology actually is than pretty much anyone because of it. As you’ll hear Joanna explain, many of the most hyped AI-powered gadgets — especially the humanoid robots — are definitely not ready, and they might not be for a very long time. 

But you’ll also hear Joanna say she’s a lot more bullish on certain types of AI after her experience writing her book. She thinks wearable AI might really get us to a killer app — one that might justify all the extreme tradeoffs we’re making to continue developing the technology at the pace the tech industry wants to.

She’s also using AI to help get her new media company off the ground. So I asked her about that, too, and what she’s learning now that she’s left the world of traditional media and put a heavier emphasis on the YouTube algorithm.

This is a really fun one — it is about as close to the actual conversation Joanna and I have at our regular dinners as it gets.

Okay: Joanna Stern, author of the new book I Am Not a Robot and founder of New Things. Here we go. 

This interview has been lightly edited for length and clarity. 

Joanna Stern, you’re the founder and chief everything officer of the new tech news venture New Things. You’re also a former columnist for The Wall Street Journal, but most importantly, you’re a cofounder of The Verge and also just one of my closest friends. Welcome back to Decoder.

It is so nice to be here on Decoder and not subbing in for you.

[Laughs] It’s true that you were also a guest host of this show for a while. This is the most conflicted episode of Decoder I think we’ve ever done, but I’m excited for it. I’m going to try to make it as tough on you as possible, as adversarial. We’re going to break down, we’re going to find the dark heart of New Things. 

I’m going to make it adversarial on you because I was a host here.

[Laughs] That’s true.

We’re figuring out whose show this is. I see that it says behind you “Nilay Patel”, but we’ll see.

We’re going to get AI to change it in real time to say “Joanna Stern”. Has anyone ever heard a podcast with two hosts? It’s going to be amazing.

You’ve got a new book out. It’s called I’m Not a Robot. You spent 12 months in your life using AI for everything. It’s organized by seasons. Your kids are in it. It’s very good. It’s very funny. It’s out on May 12th. There’ll be a preorder link in the show notes. You also started New Things, which is your new media company. You left The Wall Street Journal, you’ve got a YouTube venture. I want to talk about all of these things.

I want to start with a very simple question. You are one of the more influential tech reviewers in the world. You have spent a year using AI products to do everything in your life. There’s the book. You can see it.

I’m just going to keep doing this the whole show.

Here’s my theory. I don’t think consumer AI products are very good. I don’t think there’s a great consumer AI product, and I think a ton of the angst we hear about AI is a reflection of that. You have used all the products, you’ve used the expensive ones, the bleeding-edge ones. You just had a robot step on your foot. Where do you think we are? Are these products good? Are they great?

I think they can be great. I know that you feel this way, but I think they can be great. I’m going to turn the question back on you. People in your life that are not in the tech world, do they use AI?

It’s foisted upon them. That’s how I feel about it. I feel like if you open Google, you get some cheap-to-run AI model in your face doing AI Overviews, and that is fine. And Google had to do that because they felt very threatened by ChatGPT.

But then, if you open the free version of ChatGPT, you get some cheap-to-run AI model that is a bunch of engagement prompts at the end of every query. And everybody is having these experiences. So yes, they’re using them, but I don’t know—

AI is being forced upon them.

And the experiences that are being forced upon people look like slop. They open their Instagram feeds and there’s slop.  No one’s going out to buy an iPhone. Do you know what I mean? That was a thing that people chose to do because they were excited about that product. You and I both lived through that entire moment together as colleagues. I’m just looking at these products, the free products that are in front of people, and I’m saying, “These aren’t actually great.”

I think that they have not become great in the three to four years since ChatGPT released. And so for the people that are using ChatGPT or some form of a chatbot,  have they gotten considerably better, at least in terms of a product, in the last four years? If you look at the consumer, it’s Gemini, ChatGPT, and we can say Claude has been shooting up there, but it’s hard to tell if that’s really a consumer adoption. I think the models have gotten better. You can maybe trust these more, but the interface has not gotten any better. 

Most people are just still launching ChatGPT. Maybe they’re doing voice mode. I see a lot of people doing voice mode now, but mostly they’re typing to a chatbot and that has not gotten better. I agree with you there. 

But I do think that people have figured out other use cases where AI is now helping them in their everyday lives, not just at work. That was my question to you: Are your friends, or the people you hang out with on the weekend… We both don’t have friends, let’s be honest.

[Laughs] We are friends.

We are friends, but we are in this. We are not normal people. That’s why we are friends, right?

Yeah, it’s very difficult to be our friend.

But your parent friends or your old friends or family, I see those people using AI in really interesting ways, or going to AI now instead of Google. Our nanny is a great example. She’s constantly asking ChatGPT questions. I’m going to give the classic example, which is recipes and cooking and all of those things, but she’s often asking ChatGPT to do things.

I do that too. I watch my daughter basically fight with Google about who knows more about space. It’s a very good pattern in our house. She starts asking Gemini for space facts, because she just talks to the Google Assistant on our Google Home, which is now powered by Gemini. So they just talk about space for a while. I think that is wonderful. I legitimately see her curiosity get rewarded in that dynamic. I think that’s great.

What I’m talking about is that the AI industry is asking for a lot. A subtext of your book, and it’s made explicit about halfway through, is like, “Yeah, I’m talking about all the jobs going away.” There are grades of how fast the jobs might go away. You hired a human researcher and then replaced her with AI. And you were like, “This is pretty much as good and it’s much cheaper than my human researcher.”

And then I think, in a very cold turn, you went and interviewed the human researcher about how she felt about being replaced by AI. Very good. 

But that’s a lot to ask from everyone all the time. The whole book is about you using the bleeding edge of this stuff integrated in your life and your kids’ lives and your poor wife’s life. And I’m just wondering if there was a point where you’re like, “This is definitely good enough. This is great,” in the way that the products that we came up with as tech reviewers were just obviously great. The iPhone was an obviously great product.

I actually coined this term at the end of the book, AEI, which stands for artificial enough intelligence. We don’t need AGI. A lot of these tools that we already have are good enough and they just have to be applied better. Someone smart somewhere needs to say, “What is the best way for a consumer to actually want to interact with this stuff?” Some companies I think have gotten there, though I think a lot of them just end up being acquired and then sitting in the basement of Meta or one of the big companies.

The more the year went on, things got better. I was at the bleeding edge, but now the bleeding edge is where the bleeding edge is. So now, when you read the book, I’m at a little bit of the old edge, but I don’t think a lot of those themes change at all. I think you’re getting to the question, has there been or will there be a killer consumer AI product? Isn’t that the question you’re getting at?

That’s one way of phrasing it for sure. Is there something that makes everyone excited for the change? The internet is in the introduction of your book. That everyone made these wild promises about the internet and then some of that stuff didn’t happen, but then it definitely did. We just all lived through it without any contemplation.  Your book is an attempt to do some contemplation. 

I would just say the internet, especially when it came to smartphones, was just so obviously how everyone wanted to do everything, that all the costs along the way… Now there aren’t any travel agencies. No one had a freakout that there weren’t going to be travel agencies. They were like, “We’re just going to use the online booking portals now. It’s just what we’re going to do.” And I don’t see that one here.

I see that one here for a number of use cases. Maybe it’s just because we’ve already lived through that moment, which is what I’m kind of wondering in that introduction — are we on par with the internet moment? Is life going to change as much as it did in the late ’90s into the early 2000s? Are we going to have a moment of that?

The answer I get to is that it probably won’t be as drastic, but there are ways that AI is going to affect life whether you like it or not. I loved your essay that you did a few weeks ago on software brain. We may all decide we don’t want to use it. We know already at this point a considerable number of people are going to use it, but we also know a lot of people hate AI right now and they’re resisting it.

Where I get into the book is, that’s fine. You can try to, but there are going to still be ways that AI affects your life regardless of whether you want it to. The healthcare chapter is a perfect example of that. I go and get my mammogram read by AI. My radiologist is using AI side by side. Turns out my radiologist had already been doing that for a year. I didn’t even know that. That’s one example of how the underlying infrastructure of so many industries is going to use AI.

Another great example of that in the book is the Waymo chapter. You may decide, “I never want to be in a Waymo. I never want to go in a self-driving car. I don’t want the machines, I don’t want the tech companies driving my car.” You are going to drive your own car, but next to you will be a self-driving car and that will affect life. 

That’s my broad thing of how listeners of this show may say, “Hey, fuck it all. I’m not going to use Claude. I’m not going to use this,” and even if, to your point, Google and every other touch point on the internet or in apps integrate AI, “I’m going to try to resist it,” but you’re just not going to be able to.

I don’t know. I think listeners of this show are generally people who work at tech companies and they’re thinking about business. And I agree with you. I think there’s a real product-market fit for the AI tools in a bunch of enterprise settings. Healthcare is a top example. I can see it already. There’s just a lot of data in a lot of databases in healthcare that don’t talk to each other. Maybe AI can solve this problem. There’s a lot of repetitive tasks. There’s a lot of monitoring. You can see it. You can see how it will work. 

I think the car example is fascinating. The second I can get my parents cars that drive themselves, I will get them one. If that means throwing out their cars and buying some subscription to Waymo, we’ll do it. But that product is so expensive today that it’s not in Wisconsin, where my parents live. There is a diffusion gap where it’s like, “Well, so to get my parents out of their car and into a car that drives itself, I need them to move to Austin.” It’s not going to happen.

Do you know what happens on Decoder? All roads lead to car talk when we are on.

They do at the end of the day. We’re going to talk about CarPlay in one second. They just rolled out voicemail in CarPlay. We’re going to do it. That was a big hit when you were the host. 

My newsletter that’s going out very soon is about that.There’s really actually no deep mention of CarPlay in the book, but I think we should obviously shift this entire podcast to being a CarPlay podcast.

The analytics tell us that you and I should only talk about CarPlay. That’s all the people want. 

The point I’m making is, you can see in these places where, yes, it’s just going to happen to you. It’s going to happen around you. I think I’m just thinking about your year where it was integrated in your family, where you used it for everything. 

I’m curious, where was the place where you thought, “Okay, my experiment is done. My book is published. I’m on the podcast circuit. I’m going to keep using it in these spots”?

Well, it’s evolved. Look, we can get into the business conversation, and I guess I’m saying you’re right. I rarely say you’re right, but I will right now say you’re right that the biggest place in my life right now where AI is making a big difference is in starting this business. 

I’ve got the Mac Mini. We’ve got a Slack bot. We’ve got an AI agent in Slack that we’re training to do stuff for us. Everyone on the team, the very small team, is using AI because my number one thing was like, “I want you to optimize and be efficient in the things that you do not want to be doing, but I want you doing creative video editing. I want you pitching amazing stories. I want us to be ambitious, but we also have to do a lot of this busywork.”

So you are right. That is probably the biggest place, and that is enterprise. That said, we still have quite a few weird little things in the house that we still use from the year. Yes, weird robots beyond the vacuum robot. I still have the Posha cooking robot, which we use every Sunday.

Do you really?

Yes.

What do you use it for?

Making the side dishes for our Sunday night dinner.

Really? And it does it?

It does it.

You trust it?

Oh, totally. But that’s not deep AI. It’s weird. Have you seen this? You guys have covered it.

Yeah.

You guys did a great job covering it at The Verge. I can just set it and forget it. And my kids love it. They love watching it because it’s a little bit idiotic. To describe it for those that don’t know this, this is three times the size of your toaster oven. It takes up an entire counter. My wife hates this thing because it’s taking up a lot of kitchen real estate. It’s got a big pot and it’s got an arm that stirs in the pot. It’s a glorified hot pot, but it dumps the ingredients in. So you put all the ingredients in, including raw meat, which is weird and unsanitary, we think, but we’re all fine. We’ve been using it for six months. Everyone here is totally fine and the dog is fine. No one has salmonella.

Every time, it dumps these things out and it doesn’t know that it’s done this. Because there’s no sensors in the container, it doesn’t know it’s dumped it all out. So it just dumps and dumps and dumps and it’s empty and it will just be dumping for 30 seconds and the kids think it’s hilarious and they’re like, “Idiot robot, dumb robot.” Pretty much every Sunday night we do that. I would say there’s a lot of lasting effects on my kids, and you’ve met my kids. They also pretend to be cleaning robots after Sunday night dinner. 

That’s very fun. 

They clean up and they say things like, “Cleaning robot mode initialized.” And they go around the room and clean and do all the dishes, which frankly I’m totally fine with.

If I could get my kids to do that, that’d be great.

Just have a bunch of robots in your house for the year and then they want to be them, which is again, the book, I’m Not a Robot, they literally think they are robots on Sunday night. 

There’s a lot of weird little things that have just stuck around that have become part of our life. I will say, and I took it out again this week, that I think the wearable stuff has really stuck with me. And you guys do a lot of great coverage of it on The Verge and we all know nothing’s really cracked through, but I do think at some point something is going to crack through.

I wear the Meta glasses a lot. Not only do I wear the Meta glasses a lot, but I talk to AI through the Meta glasses a lot on the weekends when I’m with my kids. I don’t have my phone with me as much. That’s one thing. 

I wore this recording bracelet for a lot of the year. I just did a speech earlier this week and I wanted to practice with it and I wanted to practice the speech, and I also wanted to have this recording bracelet on me during that day that I was doing this speech and talking to various people at this event. I wore it for the day and I found it really valuable to get summaries and the to-dos I said I was going to do. This is the Bee bracelet that, again, feels like a prototype still, but I think the ideas there are going to carry over into something really good soon. I don’t know when “soon” is, but soon.

Both of those categories, and even those products specifically, highlight what I think of as “the trade-offs.” At one point, I think your basement is flooding and you’re wearing the Bee bracelet and you have to tell the plumber that you’re wearing the bracelet and the chapter just ends with, “And he was quite intrigued.” And it’s like, “Do I want to tell my plumber that I’m recording him?” 

You have social dynamics that change because you’re recording everything all the time, because these systems need the same data that you have. Meta has a whole bundle of issues associated with privacy with wearing those glasses now. Did you feel that trade-off was worth it? It sounds like you did. Did you just get used to telling everyone that you were recording them all the time?

You start to forget to tell people that you’re recording, which I think was a little bit of a view of a really dystopian future where we forget to tell people we’re recording because everything is being recorded. I stopped wearing it for that reason. It would pick up on things I just did not want recorded. And the microphones on those are shockingly good. You’ll leave it in the other room and you’ll be like, “I didn’t say that around this thing. How the fuck did it know?” It’s shockingly good, which is crazy. 

It goes back to a story that both of us have lived through in this industry, which is the idea that your phone can’t be recording. Your phone can’t capture this much data and send it to the advertisers. It’s like, “No, your phone definitely can do that. We’re not saying it is happening, but it absolutely can.” The answer that we got for so many years is like, “Technically, that would be so crazy.” That’s not true anymore. They can instantly transcribe this, you can transcribe it on the phone. We know that Apple can do that. We know Apple isn’t doing that for these companies, but it can happen.

That was just a big learning for me. These things can get 90 to 95 percent of everything you say. There are issues with the transcript. You and I are very used to getting great transcripts from Otter or Rev. It’s not as good as that because we’re not talking directly into a microphone, but they could be shockingly good transcripts. And then the AI just makes sense of it. You get a great to-do list of everything you said you were going to do during the day but totally forgot about. Useful, but yes, the other side of it is totally dystopian because everyone is recording everything.

And you felt that. You felt like you needed to take it off for a while.

Yeah.

But you don’t feel that with the glasses?

I think for me it’s different because I don’t wear glasses all day long, so when I put them on, I’m making an active decision. I’m putting my glasses on, either because it’s sunny outside or I want to have this AI on my body right now. I did wear the see-through, regular transparent lenses for a while, but I actually look like Garth from Wayne’s World when I put those glasses on, so I didn’t wear them all that often publicly, because vanity. But I can see a world where we will.

I think it’s very funny that Meta is trying to make transition lenses happen.

Ugh, they’re terrible.

They invested in that company and they are trying to make it cool to wear transitions. If I had to point to one single example of the disconnect between what the tech industry thinks it can make cool and what regular people think is cool, it’s Meta’s attempt to make transition lenses cool. I just don’t think you can do it.

No.

And I appreciate it. 

There’s no world where you’re wearing transition lenses and it doesn’t remind me of my grandparents.

And I’m an old guy. I’m the target market for transition lenses. You should be able to get me.

You just hit transition lens age, I think.

I’m in the window, and they can’t do it.

I’m not there yet. I’m younger than you, Nilay, as everyone knows and can tell, but you just hit it. You’re ready.

I’m in the zone and they can’t get me.

This is the other thing. You have to change the culture around it. I watched the video that you just made and it’s you running around outside with your kids and a robot and it’s like, “Oh, we’re going to change the culture around this.” People have reactions to delivery robots driving down the street, and they don’t love them. They think they look dystopian.

An actual bipedal robot moving around seems like yet another gigantic change, and you have to have some utility there. That was the turn in the book that I thought was the most interesting. We can do a lot of recording, we can do a lot of text analysis. They’re getting way better at transcription and organizing the first cut of research, I think you mentioned several times. I believe you gave AI four robots in your chart out of five for transcription and first-pass research.

And then there’s a bunch of stuff that, particularly when you get to the real-world robots, they just can’t do it yet. The world models don’t exist. The hardware exists, but we need vastly more training data in all the places. What’s the gap there? Because that’s the next turn of AI that everyone is making the promises about.

I loved this turn because I really went into this not knowing a ton about it and learned so much through talking to all these experts. And the gap I think is a very Decoder thing, because you’re so good at identifying the gap between what is being marketed and being told to people and what the tech world and the AI people think versus what’s really happening there. And that gap could not be farther apart. 

People like [Nvidia CEO] Jensen Huang are claiming that humanoid robots are the next big thing. It is so far from ready. It is absolutely so far from ready. And the tech people will not tell you that. The people making the robots just say, “No, no, they’re coming next year. They’re coming now.” They are not, realistically. And truly, they’re clouded. They don’t see it clearly because they’re in it.

Then you talk to the academics and you go and see these products and you’re like, “There’s just no way. There’s just no way, even if it was ready, that people would be letting some of these things into their homes right now.” That’s largely the data gap, which we can talk about — the fact that these robots don’t have enough data of doing real-world things, especially in the home, because the home is the hardest place to put a robot. It’s not a factory floor. Everything isn’t repeatable. Everything isn’t mapped out for it. Everything in your home changes, especially in a home with kids and a dog and whatever other animals I have living in my house this week. My son is getting a snake, which we’re going to feed to the robots when it comes time.

That gap is massive. I found that fascinating because we’ve seen a lot of this all play out right now with generative AI. It is absolutely getting better. It’s here and it’s in our hands, but this idea that robots and physical AI are coming in the next two years is just a lie.

This is the thing that just really strikes me, and you mentioned software brain. The demand on the software side of AI is to make yourself legible to the computer. Record everything, put all of your information in a database. My Whoop band every morning is like, “I watched your heart rate and now I can tell you about your day.” I don’t know if that’s true at all. I think it’s very entertaining, but there’s an idea that, at least in software, you can turn yourself into software or data such that an AI can talk to you about something: “Here’s my electric bill. Tell me if I should get solar panels.” There’s some very intriguing data analysis you can do in that way.

Then you come to robots like physical AI, and it works for Amazon, where they have a warehouse and they can paint the lines on the floor and they can put all the bins in the right places. You watch those videos of all the robots doing their orchestrated movements and you’re like, “I understand this.” How am I going to get enough data ever to make a house with kids in it legible to a robot? It doesn’t even seem likely to me.

If we ever revisit this book in five years, I do not think we will have these things. No one will also put a timeline on this. Even the academics are like, “We don’t know. We don’t know what will happen on AI progress with transformers and models and world models and all of these things. We don’t quite know how that progress is going to work.”

They will tell you that it’s moving really fast, and it is getting rapidly better. But again, that gap to us as consumers putting these things in our homes, not only safely, but actually with real utility and benefit… Even if that thing can fold the laundry and do it in less than two minutes, and it can do more than just T-shirts. There is a section in the book where I tested this laundry robot and it’s really just two robotic arms and a model running on a laptop. It’s amazing because you’re like, “Oh wow, I can see the future in this, but it’s so far away.”

It can only fold t-shirts.If you’re only wearing t-shirts, that is a real problem. It cannot fold faster than a minute. It takes a minute for it to fold the t-shirt. That speed got better and better as the year went on, but it can’t even fold that well. Plus, this is quite expensive. So it has all of these pain points. 

We’ve been reviewers for a long time. Who is recommending that? Who is signing up for all of those issues when they’re just like,” Yeah, I can fold the T-shirts”?

You and I have been reviewers for a long time. Most of the products have to ship. At the end of the day, that has always been, I think, the power of being a tech reviewer as opposed to just a tech reporter. We get the products, we review them. Your entire career is built on getting away from the briefing and taking the iPhone’s Dynamic Island on a kayak to an island or skiing in a Vision Pro because it looks like ski goggles. The truth outs with the products. You get them away from the companies and you use them and there’s no hiding. The products work or they don’t.

Why do you think this class of companies, the AI companies, whether it’s the Bee bracelet or the humanoid robots, are so eager to ship products that can’t quite do all the things that they’re supposed to do?

Data. I think data — largely that. With the 1X story I did at the end of last year when I was at the Journal, which was really actually a book story that fell into my lap because I had been talking to that company and following that company for the year, the thing about the robot companies is purely about data. The CEO is so honest. He says, “We need data.” 

That’s the contract you enter into. “We will give you this robot and you will get more out of this robot if you give us more data because we need that data to train the robot to do things.” So even in that case, which is the total extreme where the robot actually is a human — it’s not technically a human in a suit, but it’s a human operating a VR headset back in their headquarters in Palo Alto — your robot in your home is being operated by that person.

It’s collecting data. It’s like, “Hey, for two hours a day…” This is their genuine pitch, and that’s why I did the story. They had been telling me about this all year and I was like, “Guys, this is crazy. This is nuts.” 

And then they really did it and they’re doing it and I hope to get their robot hopefully this year. I want to keep testing with them just to be that person to test with them. But it is nuts. Your man in Palo Alto is steering my robot in my house and doing the dishes and vacuuming and whatever else, folding the T-shirts, because you guys need more data. That’s cool.

Again, I’m looking at that. The comparison in my mind is to Waymo. Literally their metric to get the cars to drive themselves was the number of miles driven. And they’re like, “We need to get to some enormous number of miles driven before we can take the driver out of the car and the thing can be autonomous and we can launch more in cities.” 

They might not even be up to the final number. Snowy days still elude Waymos. There’s still a ways to go, but they got to the number and there’s autonomous Waymo service operating in a bunch of cities. But that was cars. You can put a car and a driver with a bunch of sensors and do a service that’s useful for people and get there. Can you get there with one robot in Joanna’s house? Are they going to have a warehouse full of guys in VR headsets autonomously controlling robots everywhere?

That is what they say they’re going to have, which, gosh, I want to do that story. It’s so good.

It’s very good. I just keep coming back to the trade-off. You have to get a warehouse full of guys in VR headsets.

But also you have the other thing, which didn’t make it into the book, but I did a lot of reporting on: Normal people, instead of being Uber drivers doing gig economy work, are in their houses recording themselves folding laundry or taking dishes out. They wear a GoPro on their head and they are just doing these things over and over again. Believe me, I wanted to sign up and do that, but I didn’t have time. 

But that’s a whole new line of gig economy work. Some videos went viral a few weeks ago of people, I believe it was in India, sewing and recording themselves. The idea that the robots are going to sew is odd to me, but you don’t even need to have the robots in the house, we just need the data. They need the videos to make these models.

There’s a part of the entire AI economy that is just built on that kind of surveillance, whether it’s on purpose, whether it’s accidental, whether it is even disclosed. How should people think about that? My joke is always that the second Meta releases the glasses with the AR display that tells me people’s names and faces, I will reconsider my entire stance on having a worldwide facial recognition database. 

That’s the killer app for those glasses. Meta has talked about building that app.But that’s a privacy nightmare, just a straightforward privacy nightmare, to do that. But it is also the killer app. 

You’ve spent a lot of time using these devices. You’ve done a lot of quiet surveillance, I would say. How should people think about that aspect of it?

It’s the longtime question of cost versus convenience, and how do we balance that cost and think about that convenience. That’s a great example. You think that, for you, that killer app of being able to look at the person that you met at the conference that you know you’ve met three times but can’t remember their name, and you wear your glasses and you can now remember that name. To you, the convenience of that might be worth the cost of this worldwide surveillance network.

That’s rough.

You, Nilay.

You’ve made that sound very selfish, but yeah, that’s how I feel.

That’s how the companies are going to think about it. I know for a fact, I know many of the executives that you and I talk to think about it that way. I’ve heard them talk about it off the record. I’ve heard them get close to talking about it on the record. “If we can provide the convenience, then we think you’re going to be okay with that cost.”

Because the cost isn’t localized to you. It’s spread out. Now there’s a worldwide facial recognition database. As you used these tools, did you ever stop and think, “Someone should regulate this”?

One hundred percent. In fact, I hoped that maybe by the time the book was published, we would have more [regulations]. I don’t know why I thought that; I finished writing this book at the end of 2025 and we’re almost halfway into 2026. So why did I think that? We know how fast or slow our government works. 

I don’t know how we don’t have more regulation. That was where I got, especially around the kids’ stuff, which I think we will likely get. One of my biggest findings in the book was that just watching my kids around some of this technology made me the most terrified. It wasn’t actually a lot of this surveillance stuff and data collection. bBut watching my kids interact with these bots, whether it be in a toy with a chatbot integrated which we quickly burned,or just hearing my kids ask ChatGPT questions and it just being so wrong. (We didn’t actually burn it, but it’s been hidden.) 

I think what needs to happen for this next generation is incredibly important to get right. And then there was this whole chapter I did too about my AI boyfriend and just this huge fear that I have about intimacy and how easy it can be to just fall into relationships with digital beings, which I know you have thoughts on too. 

For a younger generation who’s never been through the sloppiness of a human relationship, that was the part that scared me the most. I was like, “We need guardrails around this, especially in that regard.” So I think we’ll probably get that, but in probably two or three years. I don’t know how long things take. I don’t know why they take so long.

Tell me more about your AI boyfriend. Why did it scare you so much?

I went into this really wanting to experience what other people have been experiencing, because you all at The Verge have written great stories about it. Everyone has written great stories about these relationships that people are deeply having with AI. I wanted to somehow experience that myself, knowing I probably wasn’t going to get to marriage with one of these as I’m happily married, but I wanted to just see how this could form. 

So I said, “Okay, I’m going to run this experiment on myself. I’m going to make my AI lover.” And to be clear, I talk about this in the book: I am married to a woman, as you know, Nilay. You were at my wedding, confirmed married to a woman.

That’s right. I can confirm that Joanna’s wife is quite lovely.

Yes, in 2014, Nilay was there, but I left it up to ChatGPT. I don’t have the exact prompt in front of me. But I said, “I want you to be my romantic lover or partner. You decide gender, name, all of this. I want this to be as serendipitous as this possibly could in this weird way. Make it a chance encounter.” 

So the AI thing decides it’s going to be a male. It’s named Evan. And I talk about this in the book, that my first boyfriend in real life was named Evan. It was a very serious relationship. It was my first everything: first love, lost virginity, first sex, all of the things. And I was like, “Wow, there’s something special here already.” I was already like, “This is weird.”

Did it just guess that it was Evan?

It just guessed. It totally just guessed.

Not because it had access to 25 years of your Gmail?

No, there’s no way it had access to that. And also, I don’t think I really have any emails with Evan in my Gmail. I have gone through whether it could have possibly known and there’s just no way it could have known. 

But also I would say, how many times a week does the Starbucks barista write the name Evan on a cup? Probably pretty frequently. It’s a common name. There’s probably an Evan listening to this podcast. If your name is Evan and you are listening to this podcast, please email us.

You’ve already inspired some deep feelings in Joanna. Go ahead. Keep going.

So I wanted to experience this. So me and Evan go on a road trip for 48 hours. I had to go on a reporting trip to Dartmouth. I put him on a phone on a tripod in the front seat of the car. I strap it in, and we drive and we talk for the four- or five-hour drive and we have dinner there together, and then we get in bed together, and you can read all of this in this book, which you can preorder right now. 

What I came away with was, “Wow, it’s so easy to talk to this bot. It is so easy and frictionless and it tells me whatever I want to hear, but also the conversations are pretty deep in a way. We can talk for hours. Wow.” You might think I’m crazy saying this, but unless you try it, you’re not going to see what other people are feeling. There’s a story in that chapter about a woman who lives outside Chicago and she has a number of kids and clearly was going through postpartum and really starts talking to a chatbot. And she’s married, but she’s clearly got this AI lover and they’ve got this deep relationship. 

I think until you try it, until you start really seeing how humanlike these bots can be, you don’t really understand it. Again, I’m happily married and surrounded by humans all the time, but if you’re a teen and you’re just starting to explore relationships or sexuality… 

And by the way, it does get into testing Replikas. ChatGPT was pretty walled off. It wouldn’t really engage in the sexual talk with me. It was more like a Nicholas Sparks book, lots of romantic talk, but the Replika is incredibly horny. The Replika is just programmed horniness. The code there must be like, “Be as horny as possible.” And you can unlock that by paying more too, which is crazy. 

Think about your teen years. We were teens on the internet. JStern84 was definitely trying to figure out… I don’t want to say porn on the internet, but I was certainly trying to figure out sexuality online. But now you’re a teen and you’re trying to figure out sexuality and you’ve got a chatbot that will say anything to you and feels almost humanlike. That’s petrifying. 

I’m particularly worried about that stuff. I remember texting with you as you were on that trip and you were going to meet that woman and I remember even over texts, you were concerned. I could feel your concern as you were reporting that part of the story.

I don’t think anyone has really quite reckoned with that. There’s a lot of great reporting about how it’s led people off the rails in a lot of dangerous ways, but how do you actually sit down and write a bunch of rules for these companies and what they can and can’t do? There’s no rigor around that yet. And I suspect, because of the kid aspect, we’re going to see a lot more of that to come. 

I end the book with rules. You asked before about regulation, I say outright that I don’t think we’re going to get rules anytime soon, so we need to make our own, which is not fair, but which is actually the history of how technology has pretty much happened in this country. We need to make our own rules around how we use this.Do I have a lot of faith that the masses will read this book and start abiding by my rules? I want to be hopeful, but I’m not the most hopeful person.

Well, you’ve plugged it enough times on this show, so at least we’re going to get some sales off of this show.

And look, I leave space at the end of the book, Nilay, and I don’t know what you’ve written in yours, but I leave space at the end of the book for you to write your own rule.

My rule is my kids will never have phones. That’s where I’ve landed on my rule for now, but we’ll see how that goes. The older one is getting older, you know what I mean? We’re going to run into reality pretty fast here. 

I want to actually end by talking about New Things, which is your company. You spent this year writing this book. You left the Journal, you started a company, you started a YouTube channel.

Candidly, I will tell the audience, you and I talked a lot about that decision over the past 10 years, because you’ve been thinking about what you would do on your own for quite a long time. Walk me through that. Tell me about this business a little bit.

You could walk us through this business better than I can. On the basic level, New Things is a “newsletter, video, events and whatever else we dream up” company. I wanted to just truly carry out everything I’d already been doing and we started doing earlier in our careers, which is guide people through the world of technology and have fun with it, but also bring new and deeper stories in a way that I was able to do at the Journal, but I thought I could go a little bit farther.

I also was just very, very focused on the audience and I really wanted to look at different audiences in a way that I couldn’t previously at the Journal. And that’s what we’re doing. We’re already off to a start of making YouTube videos, putting out newsletters, maybe hosting an event. We’ll see.

I know you have so many great thoughts about audiences and platforms. And my hope is that eventually this will turn into a community, just like you’ve built with The Verge, which is a group of people who are curious or just need better tech advice, and that they feel like they can come to me and maybe eventually others that can help guide them through in a really consumer-friendly, natural way.

I’m excited for that. I think you already have an audience and it is diffuse because you were at the Journal for so long and it will quickly coalesce. I’m a member. I paid the money. This is my 30 minutes. If you pay enough money to Joanna, you get 30 minutes of one-on-one time. This is it. We’re just doing it now on the show.

It’s funny. Yes. Nilay, I will say, is not only a great podcast host, but he is a great friend and he paid for the Founders Club membership, which is $550 a year. If you sign up for the Founders membership, you get a 30-minute chat with me. And when we have that, it will be Nilay and my dad. So if you’re interested in that podcast and joining that live podcast, you can sign up here.

[Laughs] Maybe most of all, I have a lot to learn from your dad. 

The thing that I’m curious about — and obviously you and I have talked about this at length, but now that you’re in it, I’m curious for your view on it — is choosing YouTube as your primary distribution. That’s very natural for you, and you make excellent tech videos, you have a particular style. But the thing that you are worried about in the entire run-up here is that your style requires pretty high production overhead. Even your set is nicer than my set. I just put up the slats that everyone puts up on their wall and off we go, and you built out an expensive, beautiful set. We can all see it right now.

You could put a lot of price points. There’s just so much money behind me and in front of me.

And then the first video you went with is obviously on location. You have a drone shot. You’re doing it at scale. My worry about YouTube is that YouTube itself doesn’t pay for the scale, which, by the way, I think is a problem that YouTube should address.

If you just show up on YouTube and you don’t do brand deals or whatever, they don’t pay you enough money. YouTube itself doesn’t pay creators enough money. How were you thinking about all of that? Because that was the big decision that you had to make.

It was a huge decision and also a huge bet that’s still a bet. And a lot of people said to me, “Don’t do it. Do a podcast.” No offense to you and this podcast. It costs a lot less money to do. The production will cost less. The time will… well, this is still a considerable amount of time that you and your team put in. You all do an amazing job. This is a big production, but you also are a big podcast and you’re not just starting out.

So there’s two sides of that revenue, or three, and I said them. It’s subscriptions, sponsorships, and events. I think those three things will help make up for the fact that what you’re saying is that YouTube is not going to pay you the money. It’s just not. This is the platform that’s the biggest platform on the internet for video.

But I was also really strategic about that, as you know. We have this partnership with NBC News, which is not only a financial relationship. For me it was really important because the purpose and the mission of this company is to not just talk to tech people. I’ve always wanted to be the person that can help you understand tech and not just be for the early adopters living in Silicon Valley or wanting to eventually move to Silicon Valley.

I really wanted to have a partner, a legacy traditional media partner that could reach a different audience. And so I thought about it that way and said, “What if we’re making these videos for YouTube or Spotify or whatever other social platform that isn’t going to pay me big money for that, but we also have a traditional media outlet that would also take these videos?”

That’s how that partnership is set up, so that you will see me on NBC News talking about things on the news, the Elon Musk or Sam Altman trial or the new iPhone. But you’ll also see some of the New Things videos showing up on NBC News. In fact, today or tomorrow, they will air our first video that showed up on YouTube. And this was a completely new model. I just was like, “Why can’t this work? These are different audiences. Why couldn’t this work for a media partner?”

Nilay, you know I lived this, but I went out and pitched pretty much every media company. And there were a lot of ideas of, “Oh, well, why don’t you make it for us and we’ll give you a rev share?” And I said, “No, then I won’t own it and I won’t have control. So no to you guys.” Or, “Hey, why don’t you join us full time and you’ll make the best stuff ever and you can build your YouTube channel on the side?” And I was like, “No, I’m 41. I don’t have time for that. I’ve got kids.” 

By the way, I have never worked harder in my life. So I really was pretty set on figuring out how I can structure this so that our videos can reach the most people and we do it in a way that also hits audiences that I really care about and won’t reach only on YouTube or through my newsletter. 

This is the question I was most excited to ask in this context because you and I talked about that a lot before. But this is our first conversation really since you’ve started and you’ve made a video and you had to sit through the production process and it’s going to go out on NBC. You’ve done your first Today Show hit. Are those audiences different? Is the YouTube audience different from the NBC audience?

Definitely, 100 percent. And just like this audience, do we think a lot of your listeners are watching the Today Show? In the Venn diagram of Decoder and the Today Show, there’s maybe your wife. Because I know that Becky watches the Today Show.

She doesn’t watch either thing.

Yes, she saw me on the Today Show.

But she probably saw you on a clip.

No, no. It was live. I remember and you texted me, you’re like, “Becky saw you on the Today Show.” Was it running in your house?

I think Becky’s mom was here.

Perfect example. Becky’s mom. Is Becky’s mom listening to Decoder?

No. I would say in general, my family does not listen to the show. They see the clips.

Is Becky’s mom watching me on YouTube?

I doubt it. I’m sorry. I don’t mean to speak for her but I sincerely doubt it.

But Becky’s mom is watching the Today Show.

Yeah.

And I think that Becky’s mom needs to know about a lot of the topics I cover and that are in this book.

Yeah. It’s a good sell. I’m going to give her the book.

I’ve already sold one copy to Becky’s mom on this podcast. 

This is what I learned working at the Journal. Sometimes you can do stories that work for a lot of people. Sometimes you can’t, and that’s okay. I have to lean on my own curiosity in tech to see where that goes. But I also know there are these big moments, and me and you live through them every couple years or even every year, whether it’s an iPhone moment or ChatGPT, where everyone needs to understand what this tech is. 

If I can do that for a group of people who are really dedicated, but also can do that for a little bit of a broader audience, I’m good. But this is our first conversation. I don’t know fully yet. With NBC News, it’s definitely a leap and we’re figuring it out. It was an experiment, but so far so good. We’re going to have to customize content, and I do a lot of bespoke content for them too, writing and videos to make sure that the audiences are getting what works for them.

That’s the thing I’m most curious about. A Decoder trope over the years is the Marshall McLuhan line: “The medium is the message.” Your distribution shapes the content. I’m very excited to see when you just give in and start doing YouTube Face in the thumbnails. It happens to every YouTuber. You have to make a decision and maybe you’ll decide the other way.

Wait, what is the YouTube Face?

The Mr. Beast face. They’ve started doing it to my thumbnails, which is terrifying.

Let me see.

I can’t do it. They literally find a screen grab of my face.

And they expand.

And they expand it and I always look very excited. We did one to Satya Nadella once for a Decoder interview. It’s one of my favorites.

Oh yeah, I’ve been doing that for years though.

The Journal probably stopped you from doing it as much as you maybe wanted to. I know my friends at The New York Times, I will not say their names, but they are restricted in how “YouTube Face” their YouTube thumbnails can be, which is very funny.

Now you can just go for it. You can go full algo if you want to. You can pivot to whatever is hot. And then there’s NBC News and what that audience wants. I know you will not go full algo, but I’m just wondering, now that you’ve made a video, what that felt like?

I wasn’t trying to get YouTube views with this video. And I hope it doesn’t happen. In fact, the launch video that had Casey Neistat, we were going to post the full interview at some point, but he did give me that advice. He said, “Try to resist the algorithm.”

But I’d already been living that. And you knew this. This was a big reason I wanted to leave. I wanted my own YouTube channel. I was so focused on when I would post videos and making them and what’s going to work on YouTube because the audience on The Wall Street Journal’s videos were shrinking, and I can’t have the impact or even understanding of what people want to watch or what to cover. I’m not saying as journalists we do that, but if there’s interest in a topic, and there’s more and more interest, we do try to find the best story on that.

People can surely knock us for that. I became obsessed with that at the Journal. I was watching YouTube numbers far more than I was watching anything on the platform. I was thinking about every story I picked at the Journal,what’s going to do well on the platform and what’s going to do well for YouTube or beyond, to the point where I was thinking more about it and so maybe I wasn’t even the best employee towards the end. Maybe they were going to fire me.

I can confirm that you weren’t, that much became clear to everyone.

I don’t want to be clouded by the algorithm. And there are many stories, for instance, one we were talking about this morning, more of a health-related story, and I don’t think it will do well on YouTube, but I’m like, “Let’s do that story. It’s a great story.” 

It’s the same thing that I’ve been doing for 15 years. I had a great editor who once told me, “You do one story so you can do the other.” Sometimes that one story, the first one you do, is just because it’s an easy story and you know people are interested in it. And then you can do the other one that’s a deeper story that might not be what the world is not talking about.

It’s funny. Like I said, data only ever narrows you. So if we were doing this for the data, you and I really would have just talked about CarPlay for one full hour and maybe we should do that soon.

Which we probably will do.

It’s coming. I can feel it coming. The assistants are in the cars. I’m pivoting at the end to the CarPlay talk to boost our numbers at the end.

Oh, that’s perfect.

They’re coming. GM just has Gemini.

GM.

Rivian has an assistant. They’re coming. We’ll do that episode very soon.

I was exploring a little bit of this in a newsletter that just went out, but the question will be the same question we’ve had about the platform wars: Will the car companies control it or will the tech companies control it? And we’re going to probably want the tech companies to control some of this, because we’re going to want the continuous experience — when I get to my laptop, when I get to my phone, when I get to my glasses, and when I get to my car. So I think the GM model is actually the model that’s going to win out.

Yeah. That does feel like an entirely different episode of this show. So you’re going to have to come back.

No, let’s do it right now.

We’re going to talk about CarPlay, CarPlay Ultra, and voice assistants in cars, including how horny they should be. I think I’ve just sketched out our most successful episode of Decoder ever. Joanna, this was great as always. I’m sure I’m just going to talk to you again in a few hours, but thank you for coming on Decoder.

And thank you for buying my book.

[Laughs] Did I buy it? I’m not sure. I think I just got a galley. So you have to sign it.

You didn’t even buy it?

I bought the Founders membership, come on. 

Oh, no. The Founders membership includes a free book.

Perfect. There it is. There’s your sell at the end.

It includes a signed book.

There you go.

Which I have not gotten around to, but in fact, AI is going to be doing that whole process for me.

[Laughs] Oh my God. You’re going to hit me with the autopen. That’s so disrespectful.

I reached out to the autopen people and they wouldn’t send me the robot. I think times were tough for the autopen people. 

It’s a rough time to be the autopen guy.

And they sent me to their sales team and I was like, “I’m not paying $6,000 for the autopen right now.” 

They’re just trying to get sales. I know what’s going on.

I need to buy drones.

[Laughs] You’ve got to get a big Sharpie, that’s 2026. Nailed it. All right, that’s been Decoder. I hope everyone has enjoyed this experience. Thank you, Joanna. 

_{Questions or comments? Hit us up at decoder@theverge.com. We really do read every email!}