The Daily - The Online Search Wars

Episode Date: February 15, 2023

Microsoft recently released a new version of Bing, its search engine that has long been kind of a punchline in the tech world.The company billed this Bing — which is powered by artificial intelligen...ce software from OpenAI, the maker of the popular chatbot ChatGPT — as a reinvention of how billions of people search the internet.How does that claim hold up?Guest: Kevin Roose, a technology columnist for The New York Times and host of the Times podcast “Hard Fork.”Background reading: When Microsoft released the new Bing, it was billed as a landmark event and the company’s “iPhone moment.”On the latest episode of “Hard Fork,” OpenAI’s chief executive, Sam Altman, and Microsoft’s chief technology officer, Kevin Scott, talk about an A.I.-powered Bing.For more information on today’s episode, visit nytimes.com/thedaily. Transcripts of each episode will be made available by the next workday.

Transcript
Discussion (0)
Starting point is 00:00:00 From The New York Times, I'm Michael Barbaro. This is The Daily. A few days ago, when Microsoft released a new version of its search engine, it claimed that it would reinvent how billions of people search the Internet. billions of people search the internet. Today, Kevin Roos, a tech columnist and co-host of the Times podcast Hard Fork, on whether that bold claim might actually be true. It's Wednesday, February 15th. Kevin, your judgment, as far as I am concerned, is pretty unerring when it comes to technology.
Starting point is 00:01:00 But a couple of days ago, when you claimed that there was a big breakthrough in how we use artificial intelligence online, I was very skeptical. And I was skeptical because you had just been on the show two months ago saying, lo and behold, there has been a big breakthrough in how we use artificial intelligence online. So that's a lot of breakthroughs, maybe one too many. So I'm going to ask you to make your case that something big has happened here. And please know that if it's not compelling, we don't have an episode, and we need an episode. Well, I should say I appreciate your confidence, but I am sort of like an advanced AI chatbot. I can be occasionally helpful and correct, but also erratic. So take this all with a grain of salt. But I do believe
Starting point is 00:01:46 we have an episode today because something very big and strange and I think important is happening in the world of internet search engines. Okay, well, tell that story. Make your argument. So, a couple months ago when we last spoke on the show, it was because this new AI tool had come out called ChatGPT, which was this very user-friendly interface for this large language model developed by OpenAI. And millions of people used it. It became an overnight sensation. But it did kind of feel for a little while like it was going to be a novelty. Right. I mean, in that episode, we treated it a little bit like a plaything. For those who missed our conversation, Kevin, you showed me how chat GPT worked, how it can write stories, right? I
Starting point is 00:02:38 mean, we wrote a story together, first in Shakespearean English, then in mafioso prose. This technology is incredibly creative and generative, and it just kind of knocks your socks off. But, you know, it felt like something that was kind of a party trick. Right. And a lot of people saw it that way. They played around with it and saw some cool things it could do, but maybe it didn't work its way into their daily routine. Right. But then, just a week or two ago, I got an email from Microsoft. And Microsoft, not known for sort of thrilling announcements, I was a little skeptical. But they were inviting me and a bunch of other reporters up to Seattle, to their campus in Redmond, Washington,
Starting point is 00:03:20 to hear about a new announcement they were making related to artificial intelligence. So I went to the Microsoft campus, I sat down, I had a little moment of fear that it was going to turn out to be like some boring update to their cloud services architecture that I would regret my entire trip for. But it was more exciting than that. They revealed a new AI-powered version of Bing. Bing, of course, being the also-ran backbencher of search engines. So here I need to say that my skepticism has reemerged. Right. And I don't blame you because Bing, for pretty much all of its existence,
Starting point is 00:04:03 has been kind of a punchline in the tech world. And it felt like being invited to the launch of a sequel for a movie that flopped the first time. Like, Bing, really, that's your big reveal. That's the thing that you were so excited about. But it was. Well, explain that. I mean, what exactly had they done with this relaunch? And why did it seem to really matter? So Microsoft, which has had a partnership with OpenAI, the creator of ChatGPT, announces that they are building OpenAI's software directly into Bing. They are basically taking this search engine and sort of Frankensteining it together with this AI large language model built by OpenAI
Starting point is 00:04:47 to make the new and relaunched Bing, which might sound like kind of a boring product update, but it is actually a radical change. You know, search is the basic atomic unit of internet activity, right? When you go online to look something up or figure something out, the first thing you do is probably you Google it. That is like a very basic part of how billions of people use the internet. And so a change to a search engine, even one as seemingly minor as Bing, that makes it work in a fundamentally different way is, I would argue, a pretty big deal. And what's so fundamentally different about an AI-driven search engine, like being from the traditional way you and I Google all day long? So I think it could be helpful to just remind ourselves of what search has looked like for the past 20 plus years.
Starting point is 00:05:43 Educate us. search has looked like for the past 20 plus years. Educate us. So right now, when you search for something on Google, what you see is some ads, maybe some image results, a featured snippet with an attempt to answer whatever question you have. But the main event is that you get links, blue links to websites. And those are the websites that Google believes are the most likely to contain the answer to whatever question you have. And so what Microsoft is trying to do using this open AI technology is to make a fundamentally different kind of search engine, one that doesn't just direct you to links, but actually tries to use this AI to answer questions directly, to write you a custom answer for whatever you are looking for.
Starting point is 00:06:29 Okay. We've clearly reached the moment in our conversation where I need to see the frigging thing to make sense of what you're talking about. Well, you can't see the thing directly because it's still invite only. You have to be sort of approved. There's a wait list.
Starting point is 00:06:46 Such a snob. I know. I know. Well, take it up with Microsoft. But I can do the next best thing, which is to share my screen with you so that you can see what this new Bing actually looks like. So here's the new Bing. Looks like a familiar search engine homepage.
Starting point is 00:07:06 Yes. So you go to Bing.com, and you can either use the traditional standard search box that comes up right in the center of the screen, or up at the top, there's a little button that says Chat. And if you click on that, it opens this new interface, what it calls an AI-powered answer engine. It says, Welcome to the new Bing.
Starting point is 00:07:26 So let's try it out. Today is Valentine's Day. And so I am going to make my wife her favorite dish, French onion soup today. But I need some side dishes to go with it. So I'm going to go over to the chat tab here. And now you can ask Bing anything. And I'm going to say I'm making a romantic Valentine's Day dinner for my wife. The main course is French onion soup. And I need some side dishes to go with it. She doesn't eat pork or shellfish. What should I make? It's a pretty long question.
Starting point is 00:08:15 Okay, you've put this up. It almost looks like a text exchange that you're having with Bing here, and it's responding that it's searching for an answer. It has some preamble. It says, French onion soup is a delicious and hearty main course, but it can also be paired with some side dishes. So it's recommending, it says,
Starting point is 00:08:34 one easy and tasty salad option is an arugula salad with cherry tomatoes, pine nuts, parmesan cheese, and avocado. You can dress it with a simple lemon vinaigrette or olive oil and vinegar. It's giving me one answer rather than a list of websites. And there are small sort of annotations on the answer where if I wanted to, I could click through and go to these websites and look at some of the text on them. But this is just pulling out the recipe right here.
Starting point is 00:08:58 And if that sounds good to me, which it actually does. So now I'm going to say, write a shopping list for the arugula salad. Right. You're just writing this right underneath this little essay you've been given about the salad. Sorted by aisle of the grocery store with amounts needed to make enough for two people. Okay, so it gave me the recipe now adjusted to make enough just for two people. So here's my shopping list now. It says produce aisle, arugula leaves,
Starting point is 00:09:37 cherry tomatoes, avocado, cheese and butter aisle, Parmesan cheese, baking aisle, pine nuts, condiment aisle, olive oil and balsamic vinegar, and spice aisle, salt and pepper. Right. And I do want to say the last thing in this list is a note to you, Kevin, and your wife. Quote, I hope you and your wife enjoy the salad and have a wonderful Valentine's Day. And then there's a heart amuse. Yeah, it's got those nice little touches too. Yeah, it's got those nice little touches too. So this is one thing that Bing can do, but it can also do a lot of other things.
Starting point is 00:10:12 Here's another example. I've been thinking about getting an e-bike. I'm not sure I'm going to get one yet, but I'm kind of curious about them. And I'm looking for one that can fit into the back of my car. So I'm going to ask Bing, what are some top rated e-bikes and can they fit into the trunk of a Toyota Highlander? So it's searching for top rated e-bikes. Now it's searching for Toyota Highlander trunk dimensions. And it's sort of combining these two searches. Then it comes up with a list of top rated e-bikes. Including prices and dimensions and little bits of reviews, stuff it's curating. Totally. And it says to fit an e-bike into the trunk of a Toyota Highlander,
Starting point is 00:11:04 you would need to fold down both rows of seats and possibly remove the front wheel and or handlebar of the e-bike. And then it says, based on the information I found, I think the M2S bike's stealth electric bike Sport R750 and the Trek Verve Plus 2 are the most likely to fit into the trunk as they have the smallest dimensions among the e-bikes I listed. However, I cannot guarantee that they will fit as I do not have the exact measurements of the e-bikes. What's fascinating about looking at your screen here, Kevin, it's not just assembling and curating, it's actually saying, in my opinion, I think this might be the one for you, right? Which is a whole other level of, I don't know if the word is autonomy or agency, judgment, right, is being applied here in a way that in search, we're absolutely not used to. Right. And I should say, like, I don't
Starting point is 00:11:51 want to pour water on how cool this all is. But, you know, I would say this is more of a starting point than an ending point for my e-bike search. Because now, you know, if I do want to go through with this, I have to actually go verify that the information that this chatbot is giving me is correct. I have to maybe go to the store and take some measurements and actually look at my trunk. So if I actually do want to buy one of these e-bikes, I'm going to have to do a lot more research to just fact check what this AI is telling me. But this is a decent starting place. Right. But this is a decent starting place.
Starting point is 00:12:24 Right. So I can clearly now see that this is changing the very idea of what search is for all kinds of reasons, right? The judgments that are being applied, the visual presentation, and above all, this utter lack of links, right? I mean, this is the un-Googling of search. Right. It is a very different kind of result than we're used to. And in some ways, it's a lot more like what we want search to look like.
Starting point is 00:12:53 Right. And of course, as convenient as that all sounds, it also sounds extremely disruptive. Yeah, it's an even bigger deal than it might seem like because this traditional search model, it's a massive economy that lots and lots of people rely on. And this new AI-powered search could change what that economy looks like forever. We'll be right back. Google-based economy. So walk us through the ways that an AI-powered search engine from Microsoft could put that Google economy at risk. Yeah, so one group that is likely to be affected by this would be advertisers, right? People who use Google and other search engines to try to sell things. So
Starting point is 00:14:00 if I am a manufacturer of microwaves, I can pay Google so that when people search for best microwave, my microwave pops up in an ad slot before their search results. That is a basic transaction. The reason that Google is one of the biggest companies in the world is because it sets the terms for online advertising. is because it sets the terms for online advertising. But if that whole model is broken by this AI-generated response style, then it's not entirely clear
Starting point is 00:14:32 how the company that makes the microwave is supposed to get its microwaves in front of potential customers. And we don't yet know how these AI-generated search engines are going to work with ads. So it may be much harder for that company to sell microwaves. That's one example. Another example is the world that we live in, the world of online media and publishers, which, as you know, rely on clicks
Starting point is 00:14:59 for their revenue. But that model gets disrupted with Bing. So I'll show you an example. What should we ask Bing about that's in the news right now? Like the Chinese spy balloon? Sure. I'll just say like, what's the latest news about the Chinese spy balloon? It says, hello, this is Bing. it says hello this is bing i like to introduce itself like yeah i know i'm on bing thank you for reminding me according to the latest news the u.s military has recovered critical electronics from the suspected chinese spy balloon downed by a u.s fighter jet off south carolina's coast on february 4, including key sensors presumably used for intelligence gathering. So it goes on, it has a whole paragraph, and then it has these little kind of annotations, these little numbers above the sentences with a link to some of the websites that it is pulling
Starting point is 00:15:58 and digesting information from, including Sky News, the New York Post, USA Today, CBS, CNN. None of which you need to read or send money to if you are relying on this summary of the news, which is just fundamentally not about clicking on a link because Bing is summarizing the news for you. Right. And you could click through and look at a full article. But if this is giving you enough information, if you feel like you're caught up on the Chinese spy balloon now, then maybe you don't click on any articles at all. Maybe you don't visit any news website.
Starting point is 00:16:35 Right. Well, cut out the middleman. Unfortunately, in this case, the middleman is our employer. Right. And Bing and Microsoft, they'll say people are still going to click through to these websites and publishers are going to do better under this regime in some way that I don't totally understand. But I think if you are a publisher who makes money from showing ads to people who visit your website, this is probably a concerning development. Right. The entire Google-based economy requires the simple act of clicking on a link. That's how companies get paid. That's how when you fall in love with a couch but don't buy it, that couch starts to follow you around the internet and show up on every page you go to, harassing you until you buy it. And if you don't click on that original link, that whole cycle breaks down and people aren't paid. And that massive economy essentially is in jeopardy. Exactly. So I have to imagine, Kevin, that Google is profoundly aware of the implications of all this and has a plan to respond. So what is its plan? So it hasn't produced a ton of publicly available products yet. But after ChatGPT launched, Google did declare something called a code red internally, which was basically them pulling the fire alarm. And then... Hello, everyone. Bonjour.
Starting point is 00:17:58 We're coming to you from Google Paris to talk about the next frontier for our information products and how AI is powering the future. On the day after Microsoft's Bing relaunch, they did hold an event where they announced a product. We're creating new search experience that work more like our minds, that reflect how we as people naturally make sense of the world. Which is their chatbot, their competitor to ChatGPT. Which we fondly call Bard. And how did that
Starting point is 00:18:30 work? Well, the demo didn't go so well. Let's say you're in the market for a new car. One that's a good fit for your family. Google showed off a number of things that Bard could do. Once you buy a new car, you'll have to plan a road trip. Bard can help you plan your road trip so you can take your new car out for a spin. Google showed off a number of things that BARD could do. It's still not public.
Starting point is 00:18:55 They're still only letting what they call trusted external testers use it. But some of the screenshots that they shared as a part of the rollout of this new BARD conversational AI chatbot, included a mistake. So one of the sample questions that Google asked BARD was, what new discoveries from the James Webb Space Telescope can I tell my nine-year-old about? And BARD responded with three bullet points, one of which said that the James Webb Space Telescope took the very first pictures of a planet outside our solar system,
Starting point is 00:19:24 which sounds plausible, but a bunch of astronomers quickly noticed that this was mistaken and that the first image of an exoplanet was actually taken in 2004 by a different telescope, not the James Webb Space Telescope. Okay. So after people start noticing this mistake, it has a really dramatic impact. Google's parent company, Alphabet, lost about $100 billion in market value that day
Starting point is 00:19:53 because of the decline in the stock. And some of that was related to this mistake by the barred chatbot. Got it. So Google's foray into competition with Bing is off to a rather rocky start. It was a very expensive mistake. everyone worries that what is happening with ai and search is going to undermine google's basic business model and the economy around it but isn't that also going to be true for microsoft i mean if you're bing until a few weeks ago your revenue model also required people to search for something find links click on them and companies got paid,
Starting point is 00:20:45 including Microsoft. So doesn't Bing as an AI search engine undermine that giant revenue stream for Microsoft? And doesn't that make Microsoft and Bing less profitable over time as well? Sure. But you have to remember, Microsoft is a tiny, tiny part of the search industry. And Microsoft has a lot of other ways of making money. So they are not nearly as dependent on search advertising revenue as Google is. And so Microsoft sees this as basically all upside for them. Ella said last week that for every 1% of market share that Bing takes from Google, they can make $2 billion more in advertising revenue. So for them, this chance to kind of eat away at Google's dominance in search is basically all upside. They don't have a ton to
Starting point is 00:21:39 lose. They very much see this as an opportunity to hit Google right in the core of their business model. And if they can do that, if they can make Bing the household go-to search engine, then they are going to be able to set the terms for how to monetize it and eventually maybe earn back some of that advertising revenue that they're losing by switching to this new model. Right. For example, it seems highly plausible that Bing's chatbot could be programmed to promote certain products above others and basically become very profitable as an advertising model.
Starting point is 00:22:14 Right. There are tons of ways you could monetize this new AI chatbot search thing. You could have it sort of subtly slip in references to certain products, or you could just have a little sort of sidebar on the AI search results that says, by the way, here are some sites where you can go buy this stuff that we've now told you about. Google and become supreme when it comes to search, which is that chat GPT, the technology that now powers this new Bing, it makes a lot of mistakes, right? I mean, Kevin, you talked to us about that in our first conversation about the technology. It makes math mistakes. It makes factual mistakes.
Starting point is 00:22:58 And its leaders have actually acknowledged this to you in interviews that I know you've done. acknowledge this to you in interviews that I know you've done. So isn't that possibly a very big vulnerability for Microsoft? Because the one thing we want our searches to be above all is accurate. Yeah, that is a real issue. And I would say the main issue that is underlying this entire trend in using AI for search. These large language models are very good at generating sort of clean, plausible sounding sentences by predicting the next word in the sentence over and over again. But that is a very different thing from looking up factual information. And actually, a few days after the Bing rela relaunch people started pointing out that it was making all kinds of mistakes too it was just making stuff up it got some numbers wrong and you can
Starting point is 00:23:51 see this i mean when i was testing the new bing last week i was looking for something to do with my family over the weekend and so i said you know what are some kid-friendly activities that I could do in my hometown this weekend? And it generated a list of options. And of the three things that it listed me, all of them had already happened. One of them happened the previous weekend. One of them happened two weekends earlier. One of them was a Hanukkah party that happened in mid-December. So none of them were actually things that I could do
Starting point is 00:24:26 that coming weekend. Which basically makes that an entirely useless search. Entirely useless search, but one that sounded very plausible. And if I hadn't gone to the underlying websites and actually looked at the dates, I would have had no idea. And I might have shown up expecting a Hanukkah party or a Lunar New Year party that had already happened and looked like a real idiot. So that's a benign example, relatively. But you can imagine this going way worse. People ask search engines for very important questions. Questions about how to save and spend their money. Questions about medications and medical advice. People really do put a lot of trust in their search engines. And if these search engines
Starting point is 00:25:05 are just generating plausible sounding but wrong answers, that could be a very big problem. Well, Kevin, given how important this is, does Microsoft have a plan to make AI being more factual? Yeah, they do. They say that this model will improve over time as they see, you know, the kinds of mistakes that it makes, and they change the underlying technology, and they make it more reliable. And, you know, there's one camp of people who says, yeah, that sounds right, because things on the internet, you know, some of them tend to get more reliable over time. Like, I've been thinking about when Wikipedia first came out, it was seen as very unreliable. Like your teachers and your
Starting point is 00:25:46 parents, they would say like, don't trust anything you read. Don't trust Wikipedia. Exactly. That was drilled into us as kids. And then over time, as Wikipedia got better and more popular and developed these kind of internal standards, it got more trustworthy. And we also sort of developed a sense as consumers of what kinds of things on Wikipedia were more trustworthy than others. Right. There's another camp of people who say, actually, no, that's not what's happening here. This is not just another kind of self-improving product because these AI models, these large language models are fundamentally untrustworthy. They don't have a way of distinguishing between true and false information. All they are designed to do is to predict the next words in a sequence.
Starting point is 00:26:36 And that kind of technology can never become trustworthy. So at the end of the day, Kevin, what you're really saying here is that this basic ritual in our lives is undergoing a revolution, how we search. And I now admit this is an episode. This does seem like a big deal. And my really big question as we conclude the conversation is, do I now need to download Bing? Because I'm a Google user in a Google-using household who works for a Google product-using company. So that would be a big change. Well, I'm not going to tell you how to live your life. Use whatever search engine you feel most comfortable with. I have started using Bing mostly to test it because I cover this stuff,
Starting point is 00:27:22 and I find it very interesting and sometimes very helpful. But I would say that no matter which search engine you use, the larger point here is that this is all happening so fast. I've been covering tech for a long time and I have never seen a new technology work its way into the heart of the tech industry so quickly. I mean, it was just a couple weeks ago, really, that we were talking about ChatGPT, this new and sort of weird and interesting chatbot. And now the same technology is being put into products that billions of people use every day. It happened so, so quickly. So I have never seen a moment like this where it felt like something so fundamental to the way the internet works, the way that society works really, is being upended so rapidly. Well, Kevin, as always, thank you very much.
Starting point is 00:28:20 Thank you, Michael. And happy Valentine's Day. And happy Valentine's Day to you. about his decision to work with Microsoft on the newest version of Bing, and about the future of search. To listen, you'll have to do some searching of your own, for Hard Fork, wherever you listen to podcasts. We'll be right back. Here's what else you need to know today. Police in Michigan have identified the gunman
Starting point is 00:29:09 who allegedly killed at least three students at Michigan State University as a 43-year-old man who had been arrested as recently as 2019 for violating state gun laws. The suspect's father told reporters that at his behest, his son had promised to get rid of his gun, but that ultimately he never did. I am filled with rage that we have to have another press conference to talk about our children being killed in their schools. During a news conference on Tuesday, the congresswoman for the district where the shooting occurred, Alyssa Slotkin, said she was outraged by the frequency of mass shootings in her community, including one at a high school just last year.
Starting point is 00:29:56 We have children in Michigan who are living through their second school shooting in under a year and a half. If this is not a wake-up call to do something, I don't know what is. And the Times reports that China is seeking to cast the aggressive U.S. response to its spy balloon, which the U.S. shot down on February 4th, as a sign of American recklessness and decline. That strategy was reflected in an editorial that ran in the People's Daily, a Chinese government-controlled newspaper which described the U.S. actions as, quote, irresponsible and hysterical. Today's episode was produced by Asta Chaturvedi, Carlos Prieto, and Claire Tenesqueter, with help from Mujzadi. It was edited by John Ketchum and Michael Benoit, contains original music by Marian Lozano and Dan Powell,
Starting point is 00:30:54 and was engineered by Chris Wood. Our theme music is by Jim Brunberg and Ben Lansford of Wonderland. That's it for The Daily. I'm Michael Barbaro. See you tomorrow.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.