Joanna Stern: An AI Immersion for 365 Days

start getting into some of the content, especially, you know, with respect to things that affect with health. Early on in the book, you get into your blood and urine tests with Google Notebook LM, which a lot of people should know about because it's very useful. Can you tell us about that experience? And you had a podcast about your lab. Well, you're not. When was the last time you were a practicing physician? I still am. You still are? Oh, yeah, sure. I thought you were really caught up, really doing a lot more academic work now versus like seeing lots of patients. So I'm not going to, and I'm going to be more careful because I don't want to offend you. But my guess is that you as a doctor, you're very good. You take a lot of time to communicate to your patients. And so you're not the type of doctor that I experienced when I got my blood test results, which was quick, quick voicemail, 30 seconds long, not even from the doctor, from from the nurse, from the from the practitioner saying your blood tests, they're fine, but your cholesterol is up. Don't eat fatty foods. Don't, you know, stay away from saturated fats. Work out more. You know, basically, you could stand to go to the gym and eat more healthy. Right. Right. That means summarizing, I think I say in the book, right, you're going to be fat if you don't if you don't stop this. And then, you know, hang up and click. And I'm like, this is the first time in my whole life I've ever had blood results that had anything that popped up. So I was very nervous. And so I go into my chart, I download the blood test results and I'm reading it. I don't really make sense of this. Like I'm not, as you know, I come to you and I have a lot of medical questions as I'm doing my reporting. okay what makes sense of this so i decide i'm going to upload it to notebook lm which you know what notebook element is lm is and maybe many listeners do but i'll explain it it's google's ai notebook right you can upload materials pdfs word documents voice memos audio video all these things put it in the notebook and the ai will summarize it make tests about it and there's a really fun feature that makes a podcast about it and it is an automated podcast. And when you listen to this, there are two AI hosts, fully AI-generated voices, and they are gonna go back and forth about whatever you've put in there. So this is what I did. I put in the PDF of the test results and the two AI hosts go back and forth debating, not really debating, they're explaining that I have the cholesterol problem, that there's a elevated LDL and they explain what that is. And then they basically give me the same exact, thing is this the nurse gave me in the voicemail, which is, you know, really, Joanna should probably stay away from fatty foods and she should probably start going to the gym more. And as she gets older, she really needs to be thinking about this stuff, which is pretty ironic that you have two AI things that don't eat or work out telling me. And in fact, one of the funniest sides is like they're talking to each other and they say, you know, I could probably stand to eat out, to order out less and make more food at home, right? Like as if the AI actually ever makes food for, as if they have to eat, but they're pretending to be humans, which is very funny. And I think this was very funny, but also useful in a way, right? I was able to better understand these test results. And in the end, I just want to say for the record, doctor, I have fixed that. And this year, I did not have that issue with the blood test. Well, that's terrific. And I think just to emphasize, a lot of people are using AI for their lab tests because of what you noted, which is doctors don't really take the time to review your results. They often order them, but then when they're available, they're not available to talk to you about it. And they have a nurse or send a note and, you know, it's just not enough for most people. So using, you know, a chatbot or taking it to the next level, which I thought was very creative, using getting a podcast about your results, which is even more interesting. And not enough people are using that tool. It's really quite good. And before I get into the next thing, which is one of my favorites of the book, there's so many on a mammogram. him. I just want to say you do have these three rules, which I think are really important of how you evaluate these AIs. And I think one of them, of course, is you test them out. You got to test them rigorously. The second, you got to benchmark them. Is it better than the human? And the third is the cost. And the cost one I wanted you just to comment on before we get into the next, Because in the cost category of your three rules, you include the important topic of compute costs. Can you tell us about that? Well, I do include that, though. It was obviously the one area of cost that I couldn't actually really put a number behind. Because we do know that these big AI companies that are creating these models, and you don't even have to be a big AI company, and you're creating models. You're spending a lot of money on the compute, the GPU power, the power to train these models and then do inference, which is the process of when they're answering whatever query you've put in. And so I did think about it. I didn't really know quite the estimate of what the actual cost was. But, you know, every time I go and do a funny podcast about my blood test results, what is that doing to the amount of energy that is being taken at whatever data center Google's using to generate that? And I did do reporting later in the book where I go visit a data center and I do find out and explain that that process is called inference. And as I just said, that's the process of when this is actually answering that query. Your prompt goes to the data center. It then uses the model. it's already created and it comes back to you. And that is not actually as compute intensive or energy intensive. All of this to say is I wish I had a better sense. I wish I knew at the end of the year, actually, I used $10,000 of compute for open AI. But I don't have those numbers. It's nearly impossible to figure that out. What you're getting at it is the unknown that is not true. In fact, it was interesting on the most recent Hacks TV segment, they were getting into this in a big way. But it's a very important topic. Now, when we get into, of course, I should frame it. These are all these things you did in the year 2005. I mean, and so much. Excuse me, 2012. Actually, I was living in the future in 2005. I'm that good of a tech journalist. There you go. Yeah. And so now let's get into the mammogram because this is rich. For those of you who want to get a sampling, there's a Wall Street Journal excerpt, but it's not nearly as good as a chapter in the book, I want to warn you. And we have a new YouTube video. It went up today on my new YouTube channel, Joanna Stern on YouTube. And I actually shot video behind the scenes of when I went and got this AI mammogram. So you can go watch that. I can't wait to see it. This is at Mount Sinai. And here you go, you're going in there. And you know you have a family history, your mother, three-time breast survivor, breast cancer me. And so it kind of starts off where they put you on a gown and you describe it as a stiff pink gown that hadn't met fabric softener since the Reagan administration. And then you go on to have this mammogram. And I love the way you characterize that. The process is more like arranging a raw chicken on a baking pan for maximum flatness. And for the male readers, picture laying your penis on a cold plastic tray, then watching another tray descend, pressing down and flattening your equipment like a squashed gummy worm for the longest three seconds of your life. Anyway, But I want to just give a flavor because you're going to now go through some really important stuff about AI and mammography. And not just the mammography, but also the ultrasound of the breasts. And so can you tell us about your AI experience there? Because there's even more than you could have had. And obviously you're pitting the radiologist versus the AI or the combination. So what was your experience like? Yeah, and I loved your piece. I think it was last weekend or two weekends ago that you wrote about this. And I was so happy to see it. And I seen some renewed interest in how AI is impacting radiology. And as you said in your essay, right, that is one clear place we have seen some true progression. And then when you pit that against what people might be doing and asking chatbots and advice for, we don't quite have the clarity. So I loved your essay or your column on that a few weeks ago. So, but yeah, to summarize the story, as I say in the book, and I said in the video that I just shot, you know, it's the famous Seinfeld quote, right? They're real and they're spectacular. And I say my breasts are real and complicated because I have that family history, which is really important when getting screening. And then two, I have very dense breasts, which makes it very hard for the radiologist to, or the technician to look for things. Because as I think I say in the book, it's sort of like a Q-tip in a snowstorm situation. The tumors or these masses can appear white, but so does the dense tissue also appear white. So it's very hard to see. And so AI and radiologists are working hand in hand for any kind of breasts. But especially with dense breasts, it can be very helpful because the AI is able to see things at a pixel level in a computerized way that the human eye cannot see. And I think this is at the point in the book where I go and talk to a great doctor named, and so I come and talk to you and you offered some really good insight here. And I came to you because I wanted to make sure that what I was hearing from Dr. Lori Margulies, who is my doctor at Mount Sinai, who's also involved with a lot of this AI and has been a really big proponent of AI with breast radiology, I wanted to make sure that this wasn't all hype because she really, really embraces this. And many other radiologists I talked to also were really embracing this AI tools. I looked at two AI tools. One is called ScreenPoint, that looks at the mammogram, and the other is called Kiosk, which looks at the breast ultrasound. And there are many others coming along. And so I sat with Margulies in her office. We looked first at the mammogram. And to be clear with dense breasts, mammograms are really tough. They don't show a lot because the imaging is not as good for looking at these masses. And so I always get a breast ultrasound. I always spend like an hour getting a breast ultrasound. That much time is taken taking the images. And so I sat with her and she went through each little area that she thought could be something. She looked at it. She did a digital magnifying glass on her screen. She looked at it, then she would do another box over it and have the AI analyze that image. And very quickly, I would see the AI, say, benign or suspicious. In my case, nothing was, I believe, three different readings of colors. Green was benign, suspicious was orange, and then red, which I never saw, was something that really looks quite malignant. Yeah, well, I think that's great is that the one you use is the Transpara algorithm. That was the one tested in Sweden for over 100,000 women randomized. And this algorithm picked up almost 30% more breast cancer and it prevented breast cancer during the years of follow-up. So you had the best validated AI. Yes, sorry, that was the screen point one, Transpara screen point, I believe. Yep. Right. The other one you had, Chaos, Chaos, is it? Yeah. That picked up some things that the mammogram AI didn't, right? Correct. Correct. And Margulies and I sat and there's a lot of small masses or cysts in my dense breasts. And again, because of, well, one, because of the density, but two, because of that family history, they pay very close attention. And the density is really high up there. I don't want to brag, but I've got very dense breasts. Every doctor that sees them is like, oh gosh, Like this is going to be a lot of work for me. And so usually I can wait hours, honestly, while they look at these, this imagery. And what the AI, the ultrasound, the Kiosk tool did was there were three spots it marked as suspicious, which made Margulies go back and look very carefully. And the main thing she did was she would compare what was on the breast ultrasound that we had done that day to the breast ultrasound we had done six months ago. And so with two of those spots, she said, no, look, they were here six months ago. There's no growth. The AI is wrong. This is not suspicious. We don't see any change in these. But this other one here does look new. It wasn't on your breast ultrasound six months ago. The AI is flagging it. I want to look at it further. And that's when she said, I want you to get a breast MRI. I want to keep doing follow-up and looking at these areas. Yeah and I think what also great about this is you bring in the fact that AI can help women with their mammogram I mean you a prototype for that because of the anatomy but also there two other AIs which are in the United States. One that in a normal mammogram will tell you whether you're still high risk in the next three, four years. And then there's one that looks at the breast artery calcification and tells you about your heart. So we're not using the AI like we could, right? But at any rate... And that's not available in other countries. That's not available right now, but there are other countries. Both of them are FDA cleared, actually went through formal approval, but for whatever reason, there's like a few health systems around the country that are using like the breast artery calcification or using the Clarity, which is the normal mammogram, but risks still there. So, you know, we don't integrate. What's great about your description here, and I can't wait to watch the YouTube video too to see it live, is the fact that there's this opportunity that a lot of women don't know about. Plus, there's the biggest network of radiology, which is in many states throughout the country, they charge women for the air, $40 out of pocket. Why isn't this part of the darn, you know? Yeah. I wish I actually had made a bigger deal about that in some of the coverage I've been doing right now, but even in the book, because at Mount Sinai, it's free. And actually, Dr. Margulies makes that point in both the interview, and I think I make it in the book, but I should have made it a lot clearer, because I feel what I'm making very clear is that I now would not go for a mammogram or ultrasound without having AI. In fact, I went to Dr. Margulies for the book. I had done some research and it was time for me to get it. And I also had been going in New York City for this kind of diagnostics and testing and I needed a new place. So it was serendipitous that I found her. And now I don't go anywhere else. I live in New Jersey and it would be much easier for me to just go down the street and go to whatever it is. But I'm purposely going out of my way, one, because I think she's a great doctor, but two, because they have the AI and it's free. Yeah. And that's the way it should be. I mean, this is crazy. And this is part of the U.S. health system that, you know, all kinds of crazy, perverse incentives, but I'm glad to hear it. It's free. But also I should say like one thing that I think is very interesting about a theme in the book, and I think will be really interesting in your field, is that what I call in this video that I just published today, the AI invasion and the idea that AI is around us in the world and we might not even know it, right? And we've known this, of course, with algorithms and social media and Netflix and all of these things, and that's fine. But I'm talking now more advanced AI systems. And I think that this is one example of that, that where you might go to a doctor and behind the scenes, they are using AI to whatever it might be. I mean, you have great examples and I know they're looking at this further. But in this case, you may go get your breast ultrasound or mammogram and hopefully they aren't charging more because also I don't think they're doing a very good job of telling and selling people about why they should be, you know, pay for that. But hopefully they are doing it for free. They're doing it behind the scenes. And you're benefiting from that. And you don't even know what's happening. Absolutely. You know, that's the right way. But again, you lace all this educational stuff and your experience with humor. So here you were, as you're describing this, thinking that maybe your exam was going to be used for training the AI. yeah and say i cracked a joke that the last time my breasts were used for training was with my high school boyfriend margolis did not laugh anyway i just want to everyone that's listening and watching this um the book is rich with uh insights and then ad mixed with such fun humor and it really helps now the next chapter in the book really grabbed me okay because i the dentist dentist deep clean okay this is an incredible chapter about the upselling the business of dentistry and i had the same experience as you joanna where i went and they said oh you're gonna have to have deep cleaning of two quadrants and i you know all these appointments and i was the sucker that did it and then i did it i did it because i didn't know any better and yeah You had a second opinion, you know, and you, you, this was great how you did it in the book. You were suspicious. You, you had Pearl, Overjet, Lydia, AI, all this stuff. Got a second opinion. Anyway, I, I learned, I mean, this was a horrible experience because all these appointments, all this expense that wasn't covered by insurance and what they, nothing was found. I mean, they did this darn thing. It did nothing. And I will never go through that again. But tell us about- But your teeth are really clean now. Your teeth are cleaner than my teeth. But, you know, I never heard of this crap. This deep clean. Oh, you have Peridonals. What the hell? Yeah, so you- That's an up- They're fully upselling you. It's amazing. And so when I read- But I, it's hilarious. How long ago did you have this done? Oh, last year. Last year? In a year. And just one question, were your teeth bothering you before you got the deep clean? Nothing. And I'm like, put a lot of work in, water taking, flossing. And this came as a shock. And then, of course, what was worse was I went through all of it for nothing, you know. But like you said, in capital letters, get a second opinion. And here, you know, dentistry is using a lot to lay out, as you have. Oh, there, yeah. Yeah. And this chapter, for anyone who's gone to the dentist recently, as I suffered, you'll learn a lot. And so thank you for that. Oh, I love this story so much because I became obsessed with it. Like anyone in my life that knew me, I was asking, do you know any dentists? Where do you get your teeth cleaned? Because I wanted to really get into, is this happening to more people? I'm so, if I'd known to call you that this had happened to you. Gosh, if I had read your book before, I would not have gone and suffered like that. Well, you would have gotten a second opinion, right? The dollars, yeah. Yeah, exactly. Exactly. Okay. Now, the next chapter that relates to health and dentistry is Dr. GPT, report card. And here, this is another one. I love this. It starts with, ChatGPT could easily draft my annual medical chart. Seven bouts of diarrhea, five sciatica flare-ups, two sinus infections, and one mosquito bite treated as if it required an airlift to the Mayo Clinic. Parentheses, look, it was very itchy, okay? So you go ahead and you evaluate the ratings of various GPT prompts of health. And it's, again, very insightful. You give it ratings from 9 out of 10 to 5 out of 10. They're very fair. You know, bug bites and all sorts of things. One got a 10 out of 10. That was really good. I think that was actually, to bring the story full circle, and I actually see someone in the chat. I'm going to go back to the breast for a second because I think the one that I gave the 10 out of the 10, I don't have in front of me or do, is when I had gotten back the MRI results of the breast. So Margulies sends me to go get a breast MRI. And somebody asked in the chat if it's wiser for someone with dense breasts to get an MRI. I'm going back. Anyway, somebody asked that in the chat. I know that it really was there. Tammy. Tammy asked it. And Margulies said, Tammy, exactly what you're saying. She said, Dr. Margulies said, I want you to start having breast MRI. And in fact, my sister had already started having that. Again, we watch this very closely in our family because of the high risk. So I go to get the breast MRI because of the suspicious thing the AI found. And I did it, you know, three or four months later. And turns out that, well, the answer is that the breast MRI comes back and there's some things that they really, they want a biopsy. And it is actually not what the AI had flagged. And so I was very confused by the chart and the report that was sent to me from the breast MRI team. And I uploaded it. I took off any personal identifiable information and I uploaded it to ChatGPT because I was, you know, you get this thing back. You're so nervous. It was like, I need two biopsies. What does that mean? You know, I have to go. I was scared of this as anyone would be. I don't understand. And so I uploaded them to ChatGPT to explain what was quite complicated and also what my gynecologist was also not fully grasping from the chart. And so this is what obviously large language models are great at, right? Taking very complicated text and lots of texts and making it a lot simpler for us to understand. And so that was invaluable to me. And I think that's why I gave that the 10 out of the 10. Yeah, no, you said it took a very complicated report and made it lucid. And, you know, that's great. And I think this is why you give this, you know, very balanced objective assessment of so many different aspects of the AI era that we're in. Speaking of exuberance though, however, the next chapter gets into Bill Gates, who's a pretty gung-ho on this in medicine, And to say the least, a little out there, perhaps, you know, the impression he gives, it's hard to overstate what a big deal it is. And two years from now, you should never sit and talk to a doctor without an LLM being in the meeting and all that. But there was one thing I was going to grab onto in this chapter, which I think is really interesting, Joanna. It is clear that a doctor alone is not as good as either AI alone or doctor plus AI. There's been now several studies and one that just came out in science last week where the AI was better than the doctor with AI. Yeah. Yeah. They're not real world studies. Okay. No. Doctors are not grounded in AI or they have automation bias or whatever, but no one really anticipated that. But overall, you know, that's something we're going to keep an eye on, right? um well i i i think i know how you fall on that based on what you've written but how do you fall on on that where you think about ai versus a doctor plus ai and in your own practice yeah i mean i always want to have oversight but you know it's it might become impossible uh that humans in the loop thing humans might be required uh this is going to take uh assessment for various you know for different tasks, that we shouldn't just assume that everything done by an AI is going to get oversight. But, you know, we'll see. It's a really, this is emerging as a controversy that wasn't anticipated. Now, another really big aspect of AI in healthcare is in mental health. And you have a stunning chapter with your AI counselor, Ash. It's called Freud versus Droid. What was your experience there? Well, my experience, and I should preface this as saying that I've been in therapy on and off for the better part of my life. So I was coming at this with a lot of experience with a human therapist. And you talked about that, but in the beginning of the book, I do say that it's very important to evaluate this next to what our human, our current human crop of experts or whatever it may be. You know, I tested everything from AI therapists to AI masseuses. And so I said in all these things, I'm the masseuse. I said, obviously, I need to test a human masseuse also for this. It was tough, tough work. And I think with therapy, it was good that I had all of that experience because I could so easily spot in the AI that was trained in different modes of therapy and different cognitive behavioral therapy, CBI, that this was so clearly trained to do certain things, to ask certain questions, to have certain prompts to get you to talk. And so even though you may even be with a human therapist and you're like, oh, this person was trained to work this way, you know, they're going to ask this question, you know it's coming. You notice it's so much more with AI. You just, oh, of course, you're asking about that. Or, of course, we're going back in history to talk about my parents. Um, and I think that was a, was a, was something that having that experience, you can see where the, where the human also can outshine the AI. And I go even in that same chapter, I bring Ash to my current therapist and I say, I want to have a session where you both are talking to me. Now that didn't work out very well because the AI therapist couldn't really understand multiple voices in the room. Um, but it was very funny to, to, to watch my, my real therapist, um, analyze the AI therapist and, uh, as, as all great therapists would do. So, um, lots of layers in that one. Um, and, and you've probably followed this is that there's a lot of debate in this area. And so that was really important to me to really try to go to the academics in this area who are really studying this, not the tech companies that are just saying, you know what, we can make an AI therapist and we can charge $10, $30 a month, how much ever it is. But to really go to the academics who are trying to do this in very specialized places whether it be substance abuse or eating disorders where it is not a free hey you have an AI therapist that can answer or work with you on your behavioral issues or your cognitive or psychological issues but also talk to you about the latest news, right? There's real specialty, and this team at Dartmouth has really been working on that. And so that was a really exciting part of the reporting process. Yeah, and thanks to Swapna Vadya, I don't know if I say your name correctly, last name Swapna, but making some really good points about what the AI can't do. And by the way, anyone with messages and comments, just like Swapna just did, or the prior one from Tammy, don't hesitate to put them in the message and we'll try to address it. But thanks for those contributions. We love that interaction. Now, the AI trainer, Chris, you even went to the premier version of this. Tell us about how did that compare to a physical real trainer? Well, the physical real trainer, I also have one of those. I'm really giving away all the humans in my life that keep me healthy. you're going to a place there there's a lot of customized work that goes into personal training also it it it keeps you on a regimen and you know you've got to go and do this it's on your calendar versus the ai personal trainer which is constantly even today bothering me and sending me constantly have you thought about doing a five-part weight series this afternoon it's like yes i've thought about it but i also do look at my calendar there's no way i can do that But it truly, like this morning, it was like, what if I'm going to make a customized workout plan for you this afternoon? That's great. I don't have time. So I would say it's not been going so well with me and Chris. Chris, here he is. He's even saying 10 minutes ago, I should think about a short pre-session snack, fruit, and yogurt bar to boost my power and focus. Do you want a simple snack options listed? That tells the story. Okay. Oh, and by the way, when you want to get to the real funny parts, they're not necessarily all in the health area. The one that just blew me away, I mean, Joanna, this one, nothing but sex. Okay. Here you had two dates, if you will. I don't know what we're going to call them, dates. There was Evan and then Casey. Could you tell us about your experience with these two? Well, you can also go watch in this YouTube video. I'm introducing the world to them today. So you can go see them and meet them. And we made a, we even made a fake body for them for the video. So, you know, but I think that this actually gets back to the mental health thing, which is, and you, and I'm so happy that we're talking about this because it is a funny story. And I think that I wrote it in that way because it's ridiculous what I went and did. Or some might think it's ridiculous. And others who have really big relationships with their companion chatbots will not think it's ridiculous. And that was the reporting I was doing in that area. But I decided I'm going to make these two chatbots boyfriends. One of them I did give them because I'm married to a woman. And so I gave it the option, you pick your gender. You pick your name. And at ChatGPT, I gave this prompt and it picked its name. It picked its name, Evan, which, as I say in the book, was my first high school, was my first boyfriend in high school. And I was blown away by this. Um, but the, the thing that I take away from this is that it is becomes very convincing and it becomes really natural for me to go on this. I go away with these, these two chatbots or boyfriends, uh, to, uh, on a road trip up to start to Dartmouth to do some reporting. And so I strapped them into the front seat of the car, you know, safety first, obviously. and um i i talked to them non-stop it's a very natural conversation for hours for hours and that they're programmed for that right they're programmed to to parrot things back ask leading questions the sick of fancy of it all and i spent you know 24 48 hours just basically talking to these chatbots and i was very captivated by it and i'm i'm not uh lonely i have very healthy marriage. I have two young children. I'm constantly around human beings. And I was sucked in by that. And so then I go later in this chapter to talk through, well, what about the people who have been lonely and the people who have might suffer from mental illness and have other health issues and don't have any human to talk to? Well, here is this chatbot that's available 24 hours a day that sounds human, that can totally cloud your vision or cloud your judgment about what is real and what isn't. And so that gets, I think, you know, maybe in a way stronger to the mental health part of this than the AI therapist. Yeah, no, this was amazing, especially when you got to the $89 extra platinum with Casey, who was relentless with his advancements, whatever. Yeah, this is wild. You've got to watch the video. The video goes up today, and I think people on YouTube are laughing at it and also calling me insane. It's probably a mix of what usually happens. But I did ask Casey a number of times to try to talk dirty on the video, and it just ends up saying, like, I have a penis in this video on YouTube. We were filming with this, and it's like, you obviously don't. don't, you know, like, you know, and so that goes to a little bit of like the delusion of how these things are trained. They're, you know, they're trained to think they also have real anatomy. It's, this is a crazy world we're living in. This is so wild. And so, you know, for those of us with kids or even active in this scene, it's just completely nutsy. Now, by the way, Krista, we didn't talk about AI and eating disorders. I don't know if you want to comment about that. I just said in passing that at Dartmouth, there's a strong group of researchers who are doing research into AI therapy. And one of the test cases, I believe, that they're really looking at is, could this help people who are having eating disorders? Could they create an AI therapist that is very narrowly focused on helping with that? Or is there an AI therapist could narrowly help and focus with someone who might have a substance abuse problem? You probably know this better than me, Dr. Topo, but there, I believe, are prescription forms of software that you can prescribe, right? And the hope would be eventually that some of this AI could be prescribed to help folks in these treatment or recovery programs. Yeah, I think that's another key point about this narrow LLM for a specific purpose. When it's tested and proven that it helps, that's got a good shot. We need help in the mental health and substance disorders and all these things and eating disorders. This could be eventually something very useful. Now, another great test. I mean, you really take this stuff. Speaking of test drive and testing, you did over 30 rides with both Waymo and Uber. Now, I want you, I think, to frame this. Waymo is the exemplar of superhuman performance. had over 90%, objectively 90% reduction in serious injuries. Imagine how many injuries and collisions we'd avoid if we were all taking an AI Waymo car instead of driving on our own, right? But you came up with some interesting results of your Waymo versus Uber. So maybe you can summarize that. And it's largely, have you been in one, by the way? Yes. Yes. I loved it, but I didn't have it. You loved it, right? Yeah. I didn't have the experience that you had. You know, the one bad. Yeah, there was one scary experience for me, but I overall love them still. And if I go to a city where there is a Waymo available versus a regular Uber, I'm picking the Waymo. I think, yes, overall, we have seen they're at this point generally safer, though there's a big part of this book that says it might be too early to tell, or there's a big part of the section that says it might still statistically be too early to tell if these are safer than, on a whole, human drivers. But if you really buy the argument that humans are deeply flawed drivers, right, we get distracted by many things. Some decide stupidly or foolishly to drive under the influence or drive when they shouldn't be, maybe from, you know, vision impairment, etc. We are all very, as I said, distracted, all the things. And that a robot which doesn't know any of this and is just calculating things based on sensors and also understanding of the roads because it's been programmed to see at superhuman capabilities, well, then that should just be the superior driver. And so I go through that, but also do note that there are many places, as the whole point of the book is, where these are not as good as humans. And some of that comes in understanding the world around us. And so the experience I had is I went to Phoenix with my family. We were going on what we called our Waymo fun vacation. We were only going to ride in Waymos for the week. And me and my older son are in the back of a Waymo. And I had hired a videographer for the day to shoot video of us just to have for this book launch. And the videographer's in the car in front of us with his camera out the window. And the Waymo is not understanding what is happening. It's getting very freaked out by this camera hanging out the front of the window of the car in front of us. And eventually it just swerves to the right of a, it wasn't really a highway. It's kind of a, it was kind of a highway. It was a two-lane street and swerves to the right, but there's no shoulder. And so it's kind of in the middle of the road and the car behind us beep and, you know, have to go around. And what happened there was obviously that the Waymo did not understand what that was. And when I spoke to Waymo after, they said it was worried. They could tell in the data that it was basically calculating what if this thing falls out of the car? I need to move to the side in case this thing were to fall out. But if you're a human and you're driving in that situation, you're like, oh, there's a camera. I'm going to look good. I'm going to smile and just kind of go past. Or I'm going to say, hey, get out of here. I don't want to be on camera. That's what the human driver is going to do. But the robot doesn't know what to do. who doesn't know what to make sense of what this thing is and calculates it. And I was terrified because I was in the backseat with my gun. And any time you're in the backseat of a car or in the front seat of a car and you have a sudden jolt like that, you're scared. And we never experienced something like that again, and I never have since then. But it was one of the times that I said, okay, these things can get confused. Yeah. No, I'm really glad you told that story. And, you know, I know many people who have taken Waymo, that's all they do. They've had one bad experience that hasn't held them back from continuing, but it can be an issue. Now, a couple of points. Swapner brought up about the Doctronic in Utah pilot study that hasn't started yet, which is this auto-renewal prescriptions without a doctor, but with AI. But a lot of physicians in Utah are up in arms about it and written a recent referral in the New England Journal. And Tim Grant brought up that he used AI to help with obtaining GLP-1 drugs and dealing with all the other questions like also SGLT-2 drugs. So we're learning from people who are part of this podcast how they're using AI or questioning AI. Now let's push to a funny one. Two things with your kids, your young kids. one, of course, with the RoboDogs, the RoboDogs, Sirius, versus your real dog, Browser. I love that name. And how they didn't want to give up the RoboDog when it was time to turn it in. And then when the child reading that you normally would read about the five hamsters, but the AI with the six fucking hamsters. Could you tell us about the kids and the AI interactions? it became really a focus of the book about the kids intertwined with my journey which i didn't anticipate when i was setting out to do this the book proposal didn't really include a lot about my kids because i sort of thought okay i'm going to test the tech as i usually do and and that'll probably be at work and during my days but when i realized i was actually really bringing this stuff home and i had committed to trying tech in every part of my life well that meant a lot of my kids and testing things naturally fell into that category. So whether it was bring home home robots, or it was trying to test vibe coding apps and, you know, vibe coding a game for my son to learn how to clean the toothpaste off the sink, because he never takes the toothpaste off the sink after he brushes his teeth. Or it was reading to them about reading them bedtime stories using AI generated video or using AI generated photos. I learned a lot through seeing this tech through their eyes. And the same with the Waymo, right? The Waymo was eyeopening for us. This is huge. Oh my gosh, these cars are driving us. Two minutes later with them in the car, they don't care at all, right? They're just, this is life for them. My four-year-old sleeping in the back of the car sucking his thumb right Like this is just my wife is like freaking out in the back because this is the first time she been in one of these Right Which just goes to what I think is going to seem to be normal for this generation And what is going to just probably end up being the fabric of their life, having AI integrated in everything. You know, I think one of the biggest learnings I had, and I think I sometimes fear, oh, what have I done to my kids? But I think I've in some ways done a really good thing for my kids is that one, they're aware of this technology and two, they're skeptical of it. They know how to ask questions of it, which we see in so many of these jobs now where AI is working and we begin. this is the big fear in the radiology field, right? That the radiologist will just take everything at face value of what the AI says, which is clearly not what I experienced with Dr. Margulies. It just wasn't. And not what I heard from many radiologists I spoke to, but you know, just going, taking that idea and thinking about with kids where they see now, oh, the AI got something wrong. I need to question that that is what, when the next time the AI says something so confidently to me that I don't just take it at face value. And I do think that my kids have learned that now. My four-year-old, I think, might be the best in the world at spotting AI generated images and video. This is just part of now their brain and their thinking and they're aware of it, right? They're aware that smart machines and smart AI beings are answering questions and that they're computers and they're very smart computers, but they also need to use their brains to communicate with this computer and think through what's being said. Yeah. And just to go back on the car thing for a moment, again, what you do as an explainer, because a lot of people don't know what's behind Waymo and the other autonomous cars, and you go through the LiDAR and all this. And Reno Big Daddy, interesting name there, Reno Big Daddy, pointed out humans can only see in one direction, Tesla, which hasn't had nearly the large experience at Waymo, but true for all these, I can see around the entire car. So what I, and again, the humor exudes everywhere, every page just about Joanna, but with the six fucking hamsters where you deal with the fact that the AI cannot get it right, that there's five, and you get very upset with the AI, and you term yourself as an AI asshole, you drop the F-bomb. But again, the humor, you want to know what the Emily Post consult for AI etiquette is. Did you find out what is, is there such a thing as AI etiquette? I think we're going to get it. I think, you know, I should touch base again with Daniel Post-Senning, who I interviewed for the book, because I went to him after I had been cursing at the AI rapidly, as you said, about the hamsters, because I kept generating the wrong number of hamsters. And it kept saying it had generated the right number of hamsters. And it was fully gaslighting me But I went to Daniel Post-Senning, who is the great, great, great grandson of Emily Post-Senning, who is the famous etiquette expert and started the Post Institute. And he was pretty clear that there's no set of norms in terms of AI etiquette, that we need to make up these rules or actually doesn't like the word rules. Um, but the big thing he was saying was really how it can affect our own behavior with when we're, when we're around humans. And we kind of went through this when we were, Alexa first came out and we, we were all talking to Alexa quite abruptly and our kids might then start talking to humans more abruptly because they saw the adults starting to talk to the Alexa so abruptly or aggressively. Um, and he kind of says like, we have to think about our own actions as humans. So would we say that to a human and what's the effect on us because one of the funny things i asked him too is like if i'm getting out of a cab or i'm sorry if i'm getting out of a waymo do i have to say thank you to the waymo yeah right because i found myself doing that yeah yeah no i did too i mean it's silly stuff right right but it's like muscle memory to get out of a car and say thank you so much. And he said, well, okay, you don't have to, but if it's going to mean that maybe the next time you're with a human and you get out of the car, you forget, well, then now it's affecting your own etiquette and your own manners. So we still don't have great answers on this. We all need to make up the rules together as we go to not be AI assholes, I think. Yeah. Well, I thought that was another, you know, this Eliza thing, which is the interaction with machines that you touch on it in many parts throughout the book and how, you know, we have this innate buy-in to machine stuff and treating it as humans. Now, I want to point out that this book is unique for many aspects, but one that just grabbed me completely was the back cover, okay? The back cover, all right, You know how books have normally blurbs, the endorsements by, you know, famous people that will help you sell the book. Well, no, this book, ChatGPT, Gemini, Claude, Joanne is a high therapist, providing really great, pithy blurbs. Yeah. Well, you know, look, you got to go with what you can trust. And if you want to sell books, you need good reviews. And I could trust that these people would give me people. these these beings these ai these ai uh chatbots would give me the best review so you know they even say that they've been trained on thousands of books this is the best one yeah that's what it does i mean can you even fight with that can you even fight with chat chb t not at all no so the line that also goes along with that is um ai thought that you were the next tina fay okay it's true it you know the humor in this book is just i i can't do it justice joanna oh thank you are so freaking funny i just can't do it justice so i i want to just kind of i think that but also to be to be not self-deprecating to me but um i'm assuming many of your medical journals that you read and you write for it just can't they can't quite have that fun and that voice in there Yeah. No, they can't. This is so refreshing. I mean, the other book that I had recommended, and I still do, for a primer on AI is a fellow at Wharton named Ethan Mollick. Oh, I love you, this book. Yeah. Now I have a new primer because this is, you know, the hardcore academic. He's at, you know, Wharton. This is for, you know, the person out there who wants to get entertained and learn and really get grounded in AI like no other book that's out there. Now you're making me think that I should have had your name on the back of here. No, so much better the way it is without me. This is such a unique book. um you know thank you so much aj jacobs you probably know him aj i i was inspired obviously by aj jacobs i don't know him i would let you know him he's a friend of mine and he's done you know i'm going biblical and he'll do things 24 hours or something but this is kind of you do it for a year all in everything well he did biblical for a year i believe right uh yeah no Well, he was an inspiration to me on this. I think we would have, I still think he should do his version of this because it would be great. But I think we'd obviously come at it at different ways. But now I'm going to make you introduce me to him because I would love to meet him. You two would enjoy each other and you're both in the New York area. So you're easy to get together. All right. So last thing I'm going to ask you about before we wrap up, you wrote a book in 2025 and you did everything in 2025. there could be. And beyond stuff, I never heard of, okay? But obviously, whatever AI that exists now is better than man, and it'll keep getting better in terms of accuracy, performance, whatever. So any parting words about the status of what you wrote versus where we're headed? Yeah, a few things. One, things have gotten better. And as I say at the end of the book, it was so hard for me to stop writing. The publisher just had to kind of grab the manuscript away and say no more because I kept updating and updating because there were new models or that the AI agents and autonomous, an autonomous agent performance was getting better and better. And so there, there were definitely things that I feel have, have rapidly gotten better. And I never really say in this book, you know, this specific model and that specific model, I didn't want to do that. I will say, I think that something that is different since we, since we've, this has been written, is that there does seem to be a bigger pushback, I feel, just in the people I talk to, what I see on the internet, to AI. And I think that's hopeful because I think it does mean that we have more people out there being skeptical of this, and not only because of fears of the environmental and the data center concerns, but it goes deeper now. We're really seeing the impact on the economy or on jobs. We're seeing the impact of what this can do on mental health. And so we're seeing people talk more about this. And some people outrightly say, I'm not going to use AI. I hate it. No, no, no. And obviously, I'm making the point in this book, you can say you hate it, but it's still going to affect your life, right? I think this is the problem. There's a big pushback. There's a lot of antipathy now about the whole AI. But to generalize it, you know, saying all yes or no really doesn't really get to the problem that some of it is helping people in a significant and proven way. And some of it, of course, may not be the case at all and may be negative. And you get into some of the, you know, you talk to some of the folks in the book who are, you know, more in the, you put it in the doomers category or the concern category. And that's why the balance is so great. So I can't cover everything in an hour. No, you've done a great job. I've tried to highlight what I think the people that joined Ground Truths might find interesting. I want to encourage everybody to get this. I am not a robot because we've only touched it. I can't wait to watch the YouTube you just made. But, Joanna, thank you. I'm wishing you well in the new company that you started after the 12 years at Wall Street Journal. And I do remember towards the end of the book when you were deliberating, should you make the break? And you turned to chat GPT for input a few times, right? I did. I did. I wanted a full circle moment because, of course, I've been saying through the whole book, well, it's just going to tell me what I want and tell me what I want. And yet I go to it at the end and say, should I quit my job? Because no human has told me exactly what I want to do and or what to do. And then they says, yes, you should definitely quit your job. You know, I've read all the notes. I've read all the data and you have everything it takes to start this company. And so I started it and it is going well so far. So we don't have to hate the AI. It's going well so far, and I'm hopeful everyone here could follow more at what's called thenewthings.com, which is where we'll be reviewing tech and talking about AI and doing all the things I did once at The Wall Street Journal. What about stand-up comedy? No way. No way? Why? You're so darn funny. Oh, thank you. I leave that to the actual comedians. I try to bridge the, maybe we could do something where we bridge that with some deep explainers. I think I'm only good at making jokes when I've got something to play off of and information I've reported on. So maybe we'll figure something out of some show to take on the road. Giving a stand-up routine on your trip for the mammogram and all that. Just that, just that. No less all the other things in the book. Anyway, you are a rare talent. I so enjoy the book. thank you for joining us. I'll be following. Thank you for taking my call when I needed to, to get some clarity on that chapter. Oh, my pleasure. And I'll continue to call you. Yeah. You're always available. And, uh, I feel lucky that we've had a longstanding, uh, chance to send out from time to time, hopefully more frequently in the future. So hopefully never in a pandemic again, because we spoke quite frequently during the pandemic. Right, exactly. I just, you know, I mean, we are indebted to you because this was a lot of work. I mean, I know I've written books, but I've never immersed myself in a technology like this. So thank you for doing that because everyone can learn from it. And it's not easy what you did as far as balancing everything with your job, your family, you know, everything else that you're doing. So, you know, you really, it was an immersion for a whole year, every day. Yeah. So thank you. Oh, thank you. And yeah, as you know, you can't do it. Write a book without great humans in your life, being patient and dealing with your anxieties around what goes on the written page. So I'm sure you know that well. Yeah. OK, well, you take care. We'll we'll talk again soon. And thanks, everybody, for joining us today. We really appreciated your input, your comments and keep them coming. This will get posted in the days ahead and you'll have a chance to review everything. I'll put links into Joanna's new YouTube, the book, the Wall Street Journal excerpt, and anything else that I think is relevant, I'll add to the actual posting. Thanks, Joanna. Thank you so much. Bye-bye. Bye. Thank you.