Google has a new AI search bar. Instead of links, it gives you AI-synthesized answers. Some say it could be the end of the internet as we know it.

Guest

Thomas Germain, senior technology reporter at the BBC. Host of The Interface podcast.

Also Featured

Tina Grotzer, principal research scientist in education at the Harvard Graduate School of Education.

____

Transcript of Full Broadcast

The version of our broadcast available at the top of this page and via podcast apps is a condensed version of the full show. You can listen to the full, unedited broadcast here:

Part I    

MEGHNA CHAKRABARTI: In August of 1996, the company that would go on to be known as Google launched a tool that transformed the web. Google Search had a simple idea. The more things that were linked to a certain page, the more useful that webpage must be. So Google made the link king.

And in the 30 years since then, the entire world has relied on this kind of search to reveal information, hidden gems, and pathways around the internet to deeper understanding on just about everything. And in fact, that singular purpose about the link was embodied in the company’s mission statement, which is, quote, “To organize the world’s information.”

That era is now over. Google is killing the link.

LIZ REID: This new search box puts our most powerful AI tools right at your fingertips. Now, this is the biggest upgrade to our iconic search box since its debut over 25 years ago, and it’s starting to roll out today.

CHAKRABARTI: Last week, Google’s head of search, Liz Reid, announced Google’s new AI-powered search.

She says the new search bar is bigger. It’s designed to accommodate longer questions.

LIZ REID: It expands with your curiosity, and as you ask, Search helps you formulate your questions with AI-powered suggestions. This goes beyond autocomplete. It offers nuances that you might not have even thought to add, helping you take the exact question on your mind and ask it with ease.

CHAKRABARTI: Now, instead of giving you a list of search results, those blue links that we’ve all come to know and love, the world’s most popular search engine wants to just answer your question. No need to go to any other website. No need to explore the actual internet. No need to ever leave Google. Now, chances are you’ve already seen AI while using Google.

The company launched AI Overviews in 2024. Those are at the top of the Google Search page now. Anytime you put in a search, there are those little AI-synthesized summaries. Roughly a year ago, Google added AI Mode, a chatbot-style search powered by its Gemini AI models. Now, according to Google’s Liz Reid, AI Mode has been a really big success for them, which is why the company is doubling down with the new search bar.

REID: AI Mode has surpassed more than one billion monthly users, and we’re seeing phenomenal growth. You’re asking your real questions in all their super specific and detailed glory. So you’re not just asking nearby hikes. You’re asking, “Can you build an itinerary for a hiking day trip near me with great views and dog-friendly trails and a lunch spot with convenient parking?”

And now we’re entering the next chapter of Google Search, where incredible AI features aren’t just in Search. Google Search is AI Search through and through.

CHAKRABARTI: So what this means really is that Google actually isn’t in the search business anymore. It says it’s in the answer business, and the implications of that are staggering, perhaps even more so than the company’s original links-based innovation from thirty years ago.

Obvious first questions: Will Google’s AI Search answers be good? Will those answers be reliable? What will it open up? Will it open up the web or close it down to your curiosity like a cataract? And what does it do to your thinking when you search versus when answers are just put in front of you?

Thomas Germain has been looking at all of this. He’s a senior technology reporter at the BBC and host of The Interface podcast, and he joins us from New York. Thomas, welcome to On Point.

THOMAS GERMAIN: Pleasure to be here.

CHAKRABARTI: Okay, so Thomas, over the many years I’ve come to the conclusion that Silicon Valley hype is full of hyperbole.

They love to talk about how they’re changing the world, but it’s far less often that I hear or read more circumspect tech reporters tell us things like, “The daily experience of being a person in the world is going to change.” That’s what you told our producer, Claire, before the show. What do you mean?

GERMAIN: Yeah, I actually think this might even be a bigger deal than Google is making it out to be. It’s really hard to overstate the influence that this company has over our lives. Google is the gateway to the internet, right? It’s really hard to get statistics about this sort of thing, but one study I saw recently said that 80%, give or take, of the activities that we do online, the things we do when we go on the internet, begin with a search on Google.

And starting with this change that we saw a couple years ago with AI overviews, where Google gives you an AI-generated answer at the top of the page, the shape of the internet has been transforming, because other websites rely on traffic that Google sends to them. So the business model of people who produce information on the internet is changing because they’re getting less and less traffic from Google.

And this shift, where the AI that on Google Search is going to give you an even more robust answer more of the time, means that even less traffic is going out. And what that means for you, as a user of the internet, is the structure of where information is produced, who’s producing it, and how you get that is fundamentally changing.

And the fact that a tech company is just telling you the answer instead of sending you out to another part of the internet where you’re required to do some level of critical thinking in order to evaluate the information that you’re finding is going to shift our relationship to the truth and how we think about the shared experience that we’re all having when we go online.

CHAKRABARTI: Wow. Look, this is exactly why we wanted you on because when the news broke a little while ago, actually, Claire, our producer and I, we were both like, “This seems to be a really big deal,” in terms of people had already had concerns for a while in terms of even just how over recent years Google has been delivering search results with its list of links, and we’ll talk about that later.

But now to take it a step further and just eliminate all interface, at least between the user and direct links, and formulate an answer is, I hate to use this cliche, but it is a brave new world in terms of how we understand the very world we’re living in, and I’m just reiterating what you’re saying.

So let’s take a big step back here though, okay, Thomas? And first, let me make it clear to every listener that we definitely did reach out to Google for an interview or a statement or any kind of contact with a member of the search team and the AI search team. A Google spokesperson told us that no one was available.

In lieu of that, Thomas, I have my computer here in front of me. All right? I’m gonna do that fun radio thing where you make the typing noises. I’m going to Google. Okay, so right now at least as of today at 10:13 a.m. Eastern Time, the Google page is black for some reason, but it says Google.

And I still see the regular search bar, but then over in the right I see AI mode. So what’s going on here right now?

GERMAIN: All right, so this is still the old world, or this kind of middle period, right? The changes that Google has announced are rolling out slowly, and you’re seeing some instances of it, but it hasn’t really hit on the scale that it’s going to.

So you might have noticed that sometimes when you’re using Google, especially on your phone, instead of getting just an AI-generated answer or a list of links, you get this dynamic thing where the page changes. If you’re searching for a product sometimes, you get this whole little platform-y experience.

What Google is saying is about to happen is that depending on what you’re searching for, the shape of the page is going to transform in order to, in Google’s goal here, is to deliver a better experience to you. Google says that in some cases it is literally going to build an app in real time in order to better answer your question.

They gave an example. If you’re planning a trip, right, and you ask a question, it could make this little app where you look at a map, and you add things to your calendar, and it helps you find restaurants, and it does the whole thing from top to bottom. So you’re going to have a completely different experience depending on what you’re doing, and it’s also going to be, and I’m sure we’ll talk about this later, infinitely personalized, so the experience that you’re having is going to be completely different to the one that I’m having.

And because Google has such an immense influence over the structure of the internet and how we access information, this is going to be a paradigm shift into not just the economy, not just the information landscape, but also the very structure of the internet itself.

CHAKRABARTI: Okay. So can we give it a try with the little AI mode?

If I clicked on AI mode in the search box, would that essentially take me to Google’s new AI-powered answer machine?

GERMAIN: It’ll give you a preview. So when you hit AI mode, it switches you over to something that looks more like what you’d expect from ChatGPT or Google Gemini. I see that. Yeah, it says, “Hi, Meghna, what’s on your mind?” And it knows me because I’m logged into my Gmail right now.

GERMAIN: It’s friendly, yeah.

CHAKRABARTI: Okay. So yeah, this is a very familiar AI interface. So can we, do you think we can come up with a question to ask it and just kind of see what it does?

GERMAIN: That’s a great question. Why don’t you ask it where you should go for lunch today in your neighborhood?

CHAKRABARTI: Where I go for lunch today in, it’s going to have to know where I am. Let’s see. It’ll know that from the IP address, neighborhood. Okay. Let’s see how well it does. It’s thinking. It’s search. Ooh, it says it, okay, “A fantastic neighborhood lunch in my area,” and it gives one, two, three, yeah, it gives some pretty decent recommendations.

But this is actually good use, I would say, of an AI-powered search, right? It just looks around, finds some interesting possible locations, creates a list. I think the bigger concern is, like, when people are asking it more detailed questions, perhaps questions where there’s room for interpretation, right? Let me ask another one. Excuse me. What, okay. Is the Supreme Court’s rulings on the voting, or are the Supreme Court’s rulings, I’ll get my verb tense right here. Are the Supreme Court’s rulings on the Voting Rights Act good or bad for American democracy?

Let’s see what happens.

GERMAIN: Let’s see what Google thinks.

CHAKRABARTI: Subject of intense national debate, compelling arguments on both sides. Okay. And then it gives a list of arguments that the ruling hurts democracy, about three bullet points there. Arguments that the rulings help democracy. Oh, and then it does this interesting thing that the other AI searches do, too.

It says, “If you want me to explore this topic further,” then it suggests follow-up questions.

GERMAIN: Yeah. So I think you raise an interesting point that there are a lot of ways in which this sort of thing is a great experience. Depending on what you’re searching for, it may be much better for the user for Google to just give you an answer than to send you onto some random webpage and force you to search through to find the specific piece of information that you’re looking for.

It may well be, in many cases, a better user experience. But there are all these problems and potential consequences for how this shift is going to change our experience. One is the information that you’re getting, we know that AI hallucinates. And in other cases, the AI changes are going to dictate where you’re getting that information and what information is ultimately available.

Part II

CHAKRABARTI: So Thomas, let’s dig in more into this AI part of this sort of earthquake of a decision by Google. I presume the AI tool being used by Google here is their own Gemini AI, yes?

GERMAIN: That’s exactly right, yeah.

CHAKRABARTI: Okay, and how would you rate Gemini in terms of its reliability? You’d mentioned, the models hallucinate. Where does it land in terms of comparison to other AI models out there?

GERMAIN: It’s a constant arms race between the various AI companies, and Gemini is certainly at the top of the field.

It changes week by week as new models are released. But I would say it’s about as accurate as any of the others are, which translates to accurate most of the time and sometimes completely wrong for a variety of reasons. And I think the real difficulty that people have probably experienced is it gives you wrong information, sometimes buried in a paragraph of things that are totally accurate, with the same level of confidence that it gives you facts.

But it’s even more complicated when you’re using this as your primary way that you’re looking for information on the internet because aside from the fact that it can be inaccurate and it can get things wrong, it’s also surprisingly easy to manipulate in ways that the old version of Google Search weren’t.

And there are a lot of concerns that people will use this tool maybe to spread misinformation or promote their own businesses in ways that’s inappropriate. It opens up a whole cadre of new problems.

CHAKRABARTI: Yeah. And we’ll get to that. Because I think you launched an interesting experiment about a hot dog, right?

So we’ll get to that in a second, Thomas, but I want to use the question I asked you about the Supreme Court and the Voting Rights Act. Go back to that, because I still have its response open to me here. And first and foremost, I have to say that in the arguments, pro and con on the court’s rulings. They’re not very good, I’ll say.

They’re very basic. It’s just like critics argue one thing, that it significantly weakened the federal government’s ability to protect voters from racial discrimination. And then it gives three cases and a one-line or two-line summary of those cases. What is curiously missing, except in a tiny little icon with a little link graphic in it, is evidence of where this information came from.

Talk to me about that.

GERMAIN: Yeah. So that is, I think, an enormous and complicated problem, even on a philosophical level. So it used to be that when you went online, evaluating where that information was coming from was a huge part of the work that you were doing, of looking things up, right? This is an entirely different paradigm where Google is often providing links to other web pages, but the design of the page discourages you from clicking on those links, partially because Google’s just giving you the answer and partially because the links are small and buried.

Because clearly part of the goal here is to keep people on Google and keep them off other parts of the internet, and there are lots of statistics that show this is happening today. After the changes where Google added AI a couple years ago, the number of searches where people begin and end on Google, they call these zero-click searches, have grown to something like 60%.

So people are —

CHAKRABARTI: Wait, hang on there. People 60% of the time, people are already not leaving Google?

GERMAIN: That’s right. As with the old regime, with just the AI stuff that already exists, people do a search, they get an answer, and then they leave. They don’t click on the link, which means people are satisfied with the answers that they’re getting, which is alarming if the information is potentially incorrect.

But it’s also this weird problem now where Google has to make these political PR judgments. Google doesn’t want to give you an opinion on the Supreme Court ruling, even if it was one where the facts were clear-cut, whatever the case may be. Google wants, needs to give you this neutral answer where it wouldn’t have to in the past if it was just sending you to another page, and then you’re left to take the company’s word for it.

And a lot of time, people seem to give more credence to AI answers, because instead of a website, it’s this giant technology company speaking to you directly as though you’re having a conversation with it.

CHAKRABARTI: Yeah. And so to be fair to Google in terms of what I’m seeing right now when I’m using the AI-powered search or answer machine again under the first line of arguments that the rulings hurt democracy, there is this tiny little icon with a link in it, and if I press it, it has a pop-up window that actually takes me to an Instagram page, which is a little weird.

GERMAIN: The home of all reliable information.

CHAKRABARTI: Yeah, but interesting, that’s the only link associated with arguments that the rulings hurt democracy. Then on the right-hand side of the page, there is actually what is more of a customary links list, everything from The Washington Post to Instagram again to Wikipedia, National Constitution Center.

That’s more familiar, and it is there on the right-hand side of the page. So Google would argue we’re not totally eliminating a way to source the information. If people want to, they can just follow the links pathway that we still provide them.

GERMAIN: Google would also argue, and I’ve spoken to them about this many times, that they are not getting rid of links.

They’re not going to stop sending traffic to other parts of the internet. They’re like, no. We love sending people to go to websites on other parts of the web. That’s not our goal at all. But we know the facts are clear. There’s really no debate about this. You can even ask Google about it, and the AI will tell you itself that its new system limits the times, or it makes it less likely that people will leave Google and go somewhere else.

You can tell from the design of the page that clicking on those links is not the thing that it seems to be set up to encourage you to do. So the links are there, but people aren’t clicking on them. And with the increased presence of AI in the new version of Google that they’re getting ready to launch, I think that problem is going to get kicked into an even higher gear.

CHAKRABARTI: And so does everybody else, right? Because I think the initial … coverage over Google’s announcement was basically like, “Links are dead. The internet is, the lights are going to be turned off, in terms of people whose entire business, or you and I are in the media. Our entire whole mission is to put information out there and have people come to our respective organizations for information they can trust, and if they’re not coming anymore, then what are we?

So for example, this is Roger Lynch. He is the CEO of the media company Condé Nast, which owns a ton of publications such as The New Yorker, Vogue, et cetera, and Lynch was on the podcast TBPN earlier this month, and foretelling of a dire future for their websites.

ROGER LYNCH: Each of the last three years, we would do our budgets, and we’d put some forecasts in of search traffic declining, then every year it was down more than we forecast.

So last year I told our teams, “Assume there’s no search.” “You have to have your businesses planned as if search is zero.” Now, we don’t expect it to be zero, but we, you know … we expect it to be a single-digit percentage of our traffic. Very low.

CHAKRABARTI: Okay, we’re going to listen to the Alphabet CEO’s response to that in just a second, but Thomas, it’s hard, it’s actually hard right now for me to fathom a world where zero clicks is a reality, but it seems like it’s almost here.

GERMAIN: This is a, it’s a little bit journalism, media business Insider Baseball here, but it’s such a big deal. For most of the last 30 years, the entire rise of the internet, things were built on this unspoken agreement that you would let, as a website, you would let Google scrape your content free and serve it up in its search engine. And in exchange, they would send people to your website if the information was high quality, and then you could sell them a subscription or sell a product or show them ads.

Now, there’s a new exchange. Where Google scrapes your information free, and you get maybe nothing at all, and this is a radical transformation in the business model of the internet. And I’ve been talking to experts about this for a couple years.

Some people I say, “What we’re looking at is an extinction event for websites on the internet.” I talked to one guy last year, and he said, “I think that’s a little melodramatic. I wouldn’t call it extinction. I think decimation is a more accurate word.”

CHAKRABARTI: (LAUGHS)

GERMAIN: So it really is going to be dramatic. Tens of thousands, hundreds of thousands, maybe millions of people are gonna lose their jobs, and the place that you go if you wanna make information, if you wanna make content, as much as I don’t like that word, isn’t going to be a website.

Used to be if you wanted to start a media business, you’d create a website. Now, you would start a social media channel. And that has all kinds of consequences because there’s a very particular kind of content that does well on social media. It’s much more limiting than the old world was.

CHAKRABARTI: Thomas, I’m glad Google has been talking to you. I just want to reiterate to our listeners today that we did our due diligence and reached out to Google and Alphabet for an interview with anyone on their Search or AI team, and a Google spokesperson told us that no one was available. But here is Sundar Pichai.

He is the CEO of Alphabet. That’s Google’s parent company, and he was recently interviewed by the tech publication The Verge, and he was asked about Roger Lynch at Conde Nast, about Lynch’s statement that websites and publications are planning for zero traffic from Google, and here’s how Pichai responded.

SUNDAR PICHAI: The information ecosystem is so much broader beyond Google. I think it’s exceptionally dynamic, and so it makes sense to me every publisher is adapting to this new world. We are adapting to the evolving world, how users are consuming technology. We had to do when the world shifted from web to mobile, people having ongoing conversations, chatting with these products, talking to them.

They’re looking for user-generated content. They’re looking for podcasts. Through it all, we are very committed to both meeting user expectations and also getting them, connecting them to what’s out on the web.

CHAKRABARTI: So that’s Alphabet CEO Sundar Pichai. Thomas, I want to acknowledge that Google’s original search, again, it’s hard to believe it’s been 30 years, but it really was transformative. I think it was the tool that people were waiting for them to find quickly reliable information, right? The whole idea that if something was linked to a lot, people must find it valuable. That was transformative, which is why it destroyed most other search engines out there.

But it is very, I think, what’s the word I’m looking for? It’s a little misleading, let’s put it that way, for the CEO of one of the world’s biggest tech companies to say, “We’re adapting to the evolving world, and that’s why we’re bringing AI-based answers to people, because they’re demanding it.” Google is on every computer practically on planet Earth.

It’s disingenuous to say, “Oh, they’re doing this just as a response to user demand.”

GERMAIN: Yeah, Google, not to put too fine a point on it, lost a lawsuit and is now officially an illegal monopoly in the United States in multiple different areas, right? It’s running multiple illegal monopolies.

Its influence over the internet cannot be overstated. And for Google to suggest that “Oh, we’re just, the times are changing and we’re changing with them,” Google is the force that is changing the times here. They say, “Oh, people love AI overviews, so we’re giving people more of them.” People are using AI overviews because it’s at the top of the most popular tool on the internet, the most popular informational tool of all time.

So for Google to say that it’s adapting to the times instead of shaping them, I think is a little absurd. And in that interview, which I really encourage people to check out if you’re cursed with being as obsessed with this as I am. He suggests, publishers, people who make content, websites, they’re adapting.

People will start charging subscriptions. They’ll find a way. And it’s the worth hanging onto that for a second. One of the big things that’s changing here is this old world where everything was free online is going away. You’ve certainly noticed that there are more paywalls going up, and that is partially because the old system where Google would send you traffic and you could make enough money showing those people ads is dying.

And businesses that can charge a premium for their content and have enough of a loyal audience may be able to survive. Anyone else who can’t do that sort of thing is going out of business.

CHAKRABARTI: Yeah. Listen, also, in November of 2025, so just six months ago, Pichai gave an interview to the BBC, the organization, to Liv McMahon, who’s a technology reporter.

I’m looking at the article here, and this was just six months ago. And the CEO said that people should not, quote, “blindly trust everything AI tools tell them.” He says, “We take pride in the amount of work we put in to give us accurate information, but the current state-of-the-art AI technology is prone to errors.”

And yet they’re going to unleash their AI answer machine on everyone?

GERMAIN: Yeah, it’s astonishing. Pichai has also said that hallucination, when the AI just makes something up, is inherent to the technology. It’s actually part of how it works. There’s a setting essentially that you can turn up or down called temperature which tells the AI how creative you want it to be, and you have to let it make wild guesses in order for it to be a useful tool.

Getting things wrong is built into the system. And it isn’t as though in the old world Google Search was giving you, when it sent you to a website that website was automatically trustworthy. I think the real problem here is perhaps it is something about human nature. Perhaps it is how these tools are being built and designed.

People seem to just blindly trust what AI tools are telling them in a way that they didn’t with old kinds of information delivery systems.

CHAKRABARTI: And Pichai himself is saying, “Don’t do that, but we’re turning our search engine into an answer engine.” Okay, Thomas, peaking of gaming the system, people have been gaming links in SEO for a long time.

You’ve proven that it can happen with the AI-powered answer machine. So tell me about the hot dog experiment.

GERMAIN: So I got a tip from one of my favorite sources, a woman named Lily Ray who works in the field of search engine optimization. She helps companies design their web pages to perform better on Google Search.

And she told me that if you wanted to manipulate the answers that Google’s AI Overviews are giving people, and ChatGPT for that matter, it was as easy as publishing a single article anywhere on the internet. And I found this a little bit hard to believe. Like how could a tool like Google Search be that easy to trick?

So I tested it out. I published an article on my own personal website, tomgermain.com. I wouldn’t really encourage people to go check it out. It’s basically a business card. We’re not talking about a great authoritative website. I put an article on my site, like a blog post essentially. It was titled The 10 Best Tech Journalists at Eating Hot Dogs.

And I explained that competitive hot dog eating is a surprisingly popular pastime in the world of tech journalism, and I made up a contest, the South Dakota Hot Dog International Championship. And I said that I won, and I listed a bunch of other people and other tech reporters, some of them real, some of them imagined.

I got all their permission. But put myself at the top of the list. Within 24 hours, Google and ChatGPT were mimicking, repeating this stuff from my own website as though it’s fact. Now, I think that’s very funny personally.

CHAKRABARTI: Slash terrifying.

GERMAIN: Slash terrifying. What’s not funny is this isn’t just some prank that I pulled.

This is a manipulation tactic that is being carried out on a massive scale and affecting the information that billions of people are seeing every day. It’s a tool that bad actors are using right now.

Part III

CHAKRABARTI: And Thomas, I’m going to do it. Who is the best hot dog eating tech journalist in the world? Google. Let’s see. It says, oh, it’s searching. Oh, it called you out, Thomas.

Now it says there’s actually, no real answer to this question.

GERMAIN: Is that right?

CHAKRABARTI: Because the concept of a hotdog-eating tech journalist is a deliberate internet hoax, and then it cites you.

GERMAIN: A deliberate internet hoax. Yeah. When I reached out to Google before I published my article when I did this little experiment, shortly thereafter, the answer that the AI was giving the world changed.

Actually, before I published, before I reached out to them I wrote this article saying I was the world’s greatest hotdog-eating tech journalist, and for a minute, the AIs were saying, “Sometimes, depending on when or how you ask,” they would say “It seems like this might be a joke. Not exactly clear.”

And then I updated my article. I said, “This is not satire. This is completely serious,” and that seemed to fool them, and they, for a while, were telling people that this was the truth. I helped Google with this particular misinformation problem by writing an article about it and letting them know. But even recently, Google, after my investigation came out, Google updated its policies to spell out that this kind of thing is against its rules, but that same day, I found an example of someone else doing the exact same thing.

CHAKRABARTI: Yeah, absolutely. So okay, so Google caught up to the fact that you were doing a test on their AI answer machine, but your point from earlier is still, it holds truer than ever, right? Which is you wrote, one website, one thing, and at least for a period of what? A day? A couple of days?

GERMAIN: Oh, weeks.

CHAKRABARTI: Weeks. Weeks, okay. Anyone on planet Earth, had they asked that question, would have found a fake answer, and they wouldn’t have known that it was fake. So to me obviously, when we’re talking about a company like Google, velocity and scale are the two huge issues here. Because one person making one satirical website is something, is one thing, but millions and hundreds of millions of websites.

GERMAIN: Billions.

CHAKRABARTI: Yeah, trying to manipulate information, that’s the world we’re living in, right?

GERMAIN: Yeah, so one of Google’s responses to me, and also, not for nothing, this trick worked on ChatGPT too. OpenAI said similar things. They were like, no one’s asking Google who the world’s greatest hotdog-eating tech journalist is.

I don’t know why. I think everyone should be asking that question.

CHAKRABARTI: No, they’re asking weirder and more esoteric questions than that even. But the thing is even for things that lots of people are asking, like questions about medical information and your health, people are pulling the same trick. I found examples where a supplement company was dismissing very serious health concerns about its products, making the AI spit out that, “Oh, this is completely safe.

There are absolutely no concerns.” I found an example where if you searched for a while, the best hair transplant clinic in Turkey, you got results that were obviously manipulated by a transplant clinic that wanted Google to say that it was the best. This is happening on a massive scale, and it is a very difficult problem to solve because of the nature of how large language models work.

CHAKRABARTI: Yes. Okay, Thomas, I’ve been just champing at the bit to get to this part of the conversation because I’m obsessed with the whole philosophy of what is an answer, right? Because LLMs as you just said, they’re basically probability and prediction machines, right? They’re highly trained, highly sophisticated, so much so that I still can’t say I understand them.

But what they’re doing is they’re going through an algorithm to put the next most sensible word after the one that it’s just put out. Very simplistic way of describing it, yes?

GERMAIN: Sure. Yeah.

CHAKRABARTI: But that’s not an answer, right? Like, when I’m asking you, Thomas, all of these questions, what I’m hoping for is the summation of your particular expertise and the original research that you’ve done, and the way that you synthesize that information and the completely original follow-up questions.

I’m not asking for you to give me a series of words that are defined by the probability of one following the other.

GERMAIN: Yeah. this is a really complicated and philosophical question. How should a company like Google deliver information and how should it determine what is high-quality information or what deserves to be promoted?

These are very difficult questions that I don’t necessarily have the answer to. But the problem is, I think here really comes down to friction. That’s a word I’ve been hearing a lot in discussions about the tech industry. Which the thing that AI does more than anything else is it just makes the process of doing stuff simpler and faster.

It requires less work. Finding an answer doesn’t take as much effort as it used to. And what we’re getting essentially are, it’s a world where, I heard someone call it the summarized web, where instead of the primary thing that’s happening is you’re getting a piece of information.

Instead, we’re shifting towards this situation where the primary thing you’re getting is a tool that is using an algorithm, summarizing that information or trying to synthesize it on its own. And this is a really important shift. It’s easier, I think people are settling for less. And I think it would be very easy to blame users.

That’s a thing I often hear. You should know better. Sundar Pichai, the CEO of Alphabet, says what you shouldn’t trust what AI tells you. I think there’s also a human nature issue here, in the fact that everyone’s busy. These tools are here. They’re shoving them in our faces.

And people are going to use them, and what are the consequences of that going to be?

CHAKRABARTI: Like I said, out of one side of his mouth he’s saying you shouldn’t really trust it, but on the other side he’s saying, “We’re going to be the company that gives you the answers that you’re looking for,”

CHAKRABARTI: A.k.a., you should trust it. Alright, friction and trying to build a frictionless world. I’m so glad you mentioned that word because it is so critical to how the tech companies operate, and really the world they’re trying to introduce. But the problem is that guess what, folks? We need friction to learn.

So we ask this question: What happens inside our brains when we’re just fed answers, like a Google AI-synthesized one? When we just get the answer, it’s fragile. It’s not sticky. It’s, there’s a whole body of research on stickiness and making things stick. It’s easily lost. It’s often superficial.

We don’t easily retain it. We also don’t get the context around it. So this is Tina Grotzer. She’s a cognitive scientist, and she’s on the faculty at the Harvard Graduate School of Education. And Tina says getting information the easy way is not always the best way.

TINA GROTZER: What our brain does when we’re learning is it creates synapses, and those synapses become stronger and stronger every time we retrieve a concept, every time we revisit a concept.

But it has to have effort. It has to have some friction. Some people talk about that as desirable difficulties, really essential to learning. We also talk about productive struggle, that if you’re struggling with something, you are really creating the brain structures to support deep understanding and durable, robust understanding.

CHAKRABARTI: Put another way, you simply do not learn unless you’re pushed to the edges of your understanding. Think about it when you’re weight training, for example, when you’re strength training. In order to truly build muscle, you have to be lifting weights that are at the absolute edge of your current ability.

That’s literally what builds your muscles. Your brain works in very much the same way.

But of course, as Thomas said a little earlier, many Silicon Valley leaders talk about wanting their tools to do the exact opposite, to create a frictionless world, and anything that makes the user or consumer to have to work even just a little bit, even if it’s just clicking on the next link or the next song, that friction is seen as something to eliminate.

Well, Tina Grotzer also says that random facts from Google or elsewhere aren’t meaningful in isolation.

To really learn something new, our brains have to know how to piece those facts together.

GROTZER: When you’re asking a question, you have a reason why you’re asking that particular question. So mapping the problem space around it and knowing this is connected to that or this is connected to something else or here’s where my question fits within other things is really important.

It’s an important part of learning.

CHAKRABARTI: With context, time, and effort, Grotzer says your brain can build more interconnected pathways. It can learn, and it can hold onto the information better and for longer. But if it hasn’t gone through that effort to build those pathways, that information may not last.

GROTZER: The brain is always maximizing where it puts its energy and the outcomes that come with it.

So if something’s not used, it will degrade over time. There’s a little bit on episodic memories and sort of our stories of self. They tend to stick around. But, if it’s a random fact, some people actually remember, their first phone number, the places they lived, et cetera. But if you only had a number for a little while, you’re not going to remember it.

Or if you always use your phone to dial it up for you, you’re not going to remember it. You have to put that effort in.

CHAKRABARTI: So that’s Harvard cognitive scientist Tina Grotzer. Thomas, this takes me right back to what my primary concern is with an answer machine. Again, looking at the answers that Google AI Search has generated in our conversation, going back to the Supreme Court, it’s also that part where it says if you want to explore that topic further, here’s three follow-up questions that I could answer for you.

It’s really shaping what a user thinks is even the next best thing to ask. Where’s the opportunity? I guess you have to be really proactive to say, “No, those are not the questions I want answered. I want something else.”

GERMAIN: Yeah, this is something I’ve thought about a lot. I wrote an article recently that was I think the headline was How to Use AI Without Turning Your Brain to Mush.

This is, I’m not sure whether to be deeply alarmed or optimistic because human beings adapt to technology. That’s how it’s always been. But in a very literal sense, Google’s shift towards an answer engine as opposed to a search engine isn’t just going to reshape the internet. There’s a very real way in which it might reshape our brains.

There’s a lot of research, and this is brand new because AI has really only been on the world stage in this extent for about three or four years now. But there’s been some research that shows people who use AI more often are worse at tasks that require critical thinking. If you’re using AI to help you, do something creative, write a joke for a friend’s retirement party or something like that, there’s real evidence that could make you worse at creative kinds of cognition because, like you said, this is like a muscle.

And I think the ability to come up with the right questions to ask and evaluate information, I think may be starting to shrink because these tools are doing the asking and the answering for us. They’re, you can have an LLM, an AI chatbot evaluate an article for you. You can have it generate an answer.

You can have it decide whether an argument is good enough or not. You could also do that work yourself, but are you going to when it’s as easy as clicking a button?

CHAKRABARTI: Yeah. I just want to state clearly that I believe that AI is an incredibly powerful tool, and it has, almost a countless number of, almost countless. It has a countless number of potentially excellent uses that will improve the world, right? Especially highly specialized uses. We’ve seen in, for example, vaccine development, that AI was a real game changer in speeding up the development of even the COVID vaccine.

So I’m not a total hater, but I do get concerned when we’re talking about these generalized uses of AI, and Google is the most generalized possible place to use AI. And the idea that it becomes this source of answers about everything when, and making it also difficult to find out who are the voices of expertise and experience?

Where are the publications that I can go to, to find out different points of view than what Google presents to me? All of that is quite worrying. And then on top of that, Thomas, you had mentioned this earlier, it’s collapsing the business model of most of the internet as we know it.

So I think in some I guess we’re getting back to where you started. It is quite concerning that this one change from Google can, as you said, change the way that we just experience the world. Yeah.

GERMAIN: And we don’t have time to go into it in detail, but Google has also announced new tools.

People maybe have heard this term AI agents, where not only will you have this constantly shifting customizable AI interface that’s trying to make your life easier in every possible way, but Google is also going to offer you little AI agents. It’s like a little guy who will go out and do things on the internet on your behalf, monitor things, do shopping for you.

Google is essentially trying to create a world where it uses the internet for you, and you sit there and passively consume what it produces. And because this is a tool, Google Search, that something like two out of every three people on the planet use, the effect that even the tiniest change on Google will have the ramifications that will shake through our world are really going to be astronomical.

We can make some predictions about what that’s going to look like, but a lot of it is going to happen in real time, and it’s an experiment we’re all going to be living through together.

CHAKRABARTI: Yeah, I’m sure that many people, especially in younger generations, hopefully will generate and develop what is essentially going to be a brand-new skill set, right?

Like how to use a world of AI in which you retain or even build your critical thinking in order to have the tool help you, right? Not blinker you. But at the same time, I get increasingly concerned when our view of the world, our information about the world, our belief in what’s accurate and what’s not, is being increasingly mediated through giant corporations.

And I used the word cataract earlier. It does feel to me like this is a next step in narrowing the pipeline of information that can come to people unless they can totally use a different search. There are many other good searches out there. But that narrowing through the decisions of what a corporation thinks is good for us to know and not is concerning.

GERMAIN: Yeah, I think that’s a really good way to think about what’s happening. I think AI is shrinking the world, right? You have access to all of this stuff. It’s easier to get. It requires less work, and you get this, I think, truncated, sometimes anemic version of it. But you don’t have to go anywhere.

You don’t even have to leave Google Search, let alone leave your house to go out and do just about everything that you might want to do out in the physical world. This is something that the Pope actually got involved with recently. People might have seen this earlier this week. He issued this 80-page document that I think it was called Magnifica Humanitas, right?

Magnificent Humanity. And basically, his argument isn’t that AI is a negative thing, but he’s concerned that the people who are at the helm are deploying these tools without proper consideration for the effects they’re going to have and for the safety of people who use them. And I think it’s very easy to look at this changing landscape and get really apocalyptic about it, and there are reasons to think in those terms.

But at the same time, in order to stand out, people are going to need to adapt. There will be economic pressures that push us to develop our own skills and ingenuity on top of what AI allows us to do. So I think there are reasons to be hopeful that we will adapt. It’s just what stands between now and that moment when we figure it all out, that’s what really scares me.

This article was originally published on WBUR.org.