Image source: Decoder

Z Highlights

• •Multimodal search: Google is creating a search superhero that can listen, watch, and read, allowing you to search not only for text, but also for video and audio.

• •AI's' fake intelligence ': Google CEO Pichai hinted that although AI is eloquent, sometimes it may just pretend to be smart, and true intelligence still depends on humans.

• •Who is right or wrong about AI generated content: Using AI to mass produce content without adding any value is not what users need. But there is a significant continuous effect, as users adapt over time.

• • The 'Mind Reading Technique' of Search: In the future, search may understand you better than your heart. You can easily find what you want with just one search, personalized enough to make you wonder if it has installed surveillance in your mind.

The relationship between language and intelligence

Nilay Patel: Sundar Pichai is the CEO of Alphabet and Google. Welcome to Decoder.

Sundar Pichai: Nilay, it's a pleasure to be here.

Nilay Patel: I'm glad to be invited to do an interview, and it feels like I talk to you every year at Google I/O, and we talk about all the things that you said, there's a lot of news to talk about AI. As you know, I'm particularly interested in the future of the web, so I really wanted to talk to you about this topic, but I'm going to start with simple questions first. Do you think language and intelligence are the same?

Sundar Pichai: Wow, that's not an easy question! I don't consider myself an expert on this, and I think coding languages do incorporate a lot of intelligence, probably more than people think. This largely explains the success of large language models, but my gut tells me that as humans, we have access to information in ways that go far beyond language, but I have to say that there is a lot more language than people think.

Nilay Patel: The reason I'm asking this question is that I'm seeing announcements on I/O about AI and what you're doing, I'm seeing your competitors and what they're doing, it's all language-centric. It was LLMs that really sparked the interest and investment boom in innovation and investment, and I doubt that the development of intelligence has kept pace with the improvement of language skills. Honestly, I don't quite think so. I've seen computers get better and better at language, and in some cases actually get dumber. I wonder if you see the same chasm.

Sundar Pichai: yes, that's a good question. Part of the reason we're making Gemini natively support multimodal processing - you're starting to see some signs now, but it's not quite into the product yet - through audio, video, text, images, and code, when we all have multimodal work in terms of input and output - we're using all of that to train the model - maybe in the next cycle, it's going to be more encapsulated than it is now, which is mostly text-based now. I think the continuum is going to shift as we get more information in this way. So maybe there's more to come.

Nilay Patel: I have a concept called "Google Zero", which is the result of my own paranoia. All the referral traffic that The Verge ever had went up and then disappeared, and now Google is the largest source of referral traffic for almost every website on the web. I can see that for many websites, "Google Zero" is happening. Their Google traffic has dropped to zero, especially for independent websites that are not part of some large publishing group. We reported on an air purifier blog called HouseFresh. There is also a gaming site called Retro Dodo. Both sites say, "Look, our Google traffic is down to zero. Our business is coming to an end. ”

Sundar Pichai: It's always difficult to talk about individual cases, and at the end of the day, we're trying to meet the expectations of our users. Users are voting with their feet, people are trying to figure out what's valuable to them, we're doing it on a massive scale, and I can't answer specific sites –

Nilay Patel: A lot of the little players feel hurt and they say loudly, "Our business is dying. That's what you say: "We're engaged, we're communicating." "But it's happening very clearly.

Sundar Pichai: I don't know if it's a general trend, I have to look at the data as a whole, so on a case-by-case basis, there's always someone who comes into a field and says, "I, as a particular site, I'm doing worse." "But it's like a single restaurant saying, "I'm slowly losing customers this year, people don't come here anymore," or whatever. This is not necessarily true, probably another restaurant opened next door and did a very good job, so it's hard to say. From our standpoint, when I look back at the last decade of history, we've provided more traffic to the ecosystem, and we've driven that growth. You may be making a secondary point about small websites versus more aggregators, and that's the second point you're discussing. Ironically, sometimes we make changes in order to actually send more traffic to smaller websites. Some of the sites that complain a lot are actually aggregators in the middle. So, should the traffic go to the restaurants that have created websites with menus, or to the people who write about those restaurants? These are deep-seated issues. I'm not saying there's a right answer.

Nilay Patel: But you're on the verge of turning the whole situation upside down, right? You're about to start answering these questions directly. And the source of the content in the future, I think you want the people who care most to publish that information directly, to be something that you synthesize.

Sundar Pichai: I agree.

Nilay Patel: But the motivation to do so seems to be getting lower and lower – at least on the web.

Sundar Pichai: I feel like it's the opposite, if anything, I feel like with AI overview, when you give people context, sometimes people just want a quick answer and then they bounce back. But in general, when we look at the user journey, when you give context, it also exposes jumping points, so they're more engaged, and actually, that's what drives growth over time. I look at desktop to mobile and have a similar issue. In fact, there was a magazine cover that I almost wanted to pull out that said, "The Internet is dead." "There was a Google Zero argument 10 years ago. But you yourself point out that we are still one of the biggest referrers, and it is no accident, because we have been very concerned about it for a long time.

I've looked at our experience, even last year with search to generate experiences, and I've constantly found that the approach we prioritize sends more traffic while meeting user expectations. We do this thoughtfully, and we actually change our approach. If there's an area that we feel we're not doing exactly right, we'll be cautious about exiting. But I think what pleasantly surprised us was that people were more engaged, which would lead to more growth in high-quality content over time.

There's a lot of debate about what constitutes high-quality content, at least in my experience, I value independent sources, I value smaller things, I want more authentic voices, and I think those are important attributes that we're constantly trying to improve.

Nilay Patel: You mentioned that people are going to click on more links in the AI overview. Liz, who is in charge of search, makes the same claim in a blog post. There is no public data showing that this is true. Will you publish that data? Will you show people that this is actually happening?

Image source: Decoder

Sundar Pichai: Looking at it as a whole, I think people rely on the value of this ecosystem. If people don't see value over time, website owners don't see value coming from Google, I think we're going to pay the price. We have the right incentive structure, we have a lot of individual differences, and some are which path the user chooses to take. This part is hard to figure out, but I do think we're committed to doing the right thing on a holistic level.

Nilay Patel: I read some industry publications from the SEO community this morning that were in response to the change, and one point that was pointed out was that in the search console, it didn't show whether the click was from a featured snippet, an AI overview, or just Google's regular 10 blue links. Will you show it separately? Would you commit to showing it separately so that people can actually audit, verify, and measure whether the AI overview is actually sending that much traffic as you said?

Sundar Pichai: That's a good question for the search team. They think about it more deeply than I do. I think we're constantly trying to provide more visibility, but at the same time we want people to create good content. We're trying to rank and organize it, so I think there's a balance here. The more we elaborate, the more people design for that. There's a trade-off there, so it's not clear to me what the right answer is.

Ethical issues in the use of user-uploaded data

Nilay Patel: You're detailing the trade-off between what people are making, and that's been happening for a while. I think that regardless of whether you think this homeostasis is good or bad, it is at least a homeostasis. Now, that state is changing – and AI is clearly changing it.

The 10 blue link model, old homeostasis, relies heavily on a swap: "We're going to let you index our content, and we're going to have featured snippets all exposed to you." In return, you'll contribute traffic to us. This forms the basis of what you might call a fair use argument. Google will index these things and there won't be much payment in between.

In the age of AI, no one knows how it will evolve. There are some big lawsuits going on. Google and OpenAI are trading for training data. Do you think it's appropriate for Google to pay for more data to train search results? Because those AI snippets are not the same as the last 10 blue links or anything else you've ever done.

Sundar Patel: To be very clear, there's a myth that Google search has always been 10 blue links — I look at our mobile experience — for many, many years. We've got the answers, we've allowed you to refine the questions, we've got featured snippets, and so on. The product has evolved significantly.

That being said, as a company, even though we've done Google News impressions and licensing agreements for the sake of AI. We clearly believe that there are cases for equitable use in the context of beneficial, transformative use. This one is indisputable. I think there is dedicated incremental value to our model, and we will look to collaborative partnerships to make that happen. I do think we're going to approach it that way.

Nilay Patel: Let me put this another way to you, I'm not going to do too much fair usage analysis with you, I promise, as much as I like to do that. There have been some recent news reports that OpenAI trained its video generation product, Sora, on YouTube. How did you feel when you heard the news?

Sundar Patel: We don't know the details. Our YouTube team is following up and trying to make sense of it. We have terms and conditions that we want people to adhere to when they build a product, so that's how I feel about it.

Nilay Pichai: So you feel like they might have violated your terms and conditions? Or if they did, it was inappropriate?

Sundar Pichai: That's right.

Nilay Patel: The reason I'm asking this question is — and it's a more emotional question — well, maybe that's inappropriate. And what OpenAI says is basically "we've trained on publicly available information," which means they found it on the web.

Most people don't have the opportunity to make that kind of deal. They don't have YouTube's licensed team of professionals to say, "We have terms and conditions. "They don't even have terms and conditions. They just put their stuff on the internet. Do you understand why, emotionally, the creative community's reaction to AI – feels like you might feel about OpenAI training on YouTube?

Sundar Pichai: Absolutely, whether it's a website owner or a content creator or an artist, I can understand how emotional this shift can be. Part of the reason is that you see us at Google I/O, even when working on a product like music generation, we do take an approach where we make tools for artists first. We don't have a universal tool to create songs for anyone.

The approach we've taken in many of these cases is to put the creator community at the center as much as possible. We do this on YouTube for a long time. Still, we're trying to figure out the right way to do it.

But it's also a moment of transformation, and there are other players in the ecosystem. We are not the only participants in the ecosystem. But, to your previous question, yes, I understand people's perception of it. I am absolutely very sympathetic to people's perception of this moment.

Nilay Patel: They feel like it's a solicitation — they put their work on the internet, the big companies come in, take it for free, and then make a product that they charge $20 a month, or elevate their creative work, and mix for other people. Make it feel like a claim thing, with little value coming back to them. That's what I'm really asking: how do you bring value back to them? How do you bring the incentive back to small creators or independent businesses that say, "Look, this feels like a take"?

Sundar Pichai: The whole reason we've been successful on platforms like YouTube is that we've been trying to answer that question. You'll continue to see us think deeply about how to do this well. I think that the participants who end up doing better here will have more winning strategies over time, and I sincerely believe that.

In everything we do, we have to address this. Whenever you operate a platform, this is the foundation on which you can build a sustainable long-term platform. With this AI moment, over time, there will be some players who better support their platform through content creators, and those who do it better will emerge as winners. I believe this is a creed of these things over time.

The impact of AI-generated content

Nilay Patel: I think one of the particularly interesting contrasts with YouTube is that people have described to me a lot that YouTube is actually a licensing business. You're licensing a lot of content from creators, and you're obviously giving them back through the advertising model there. The music industry has a huge licensing business with YouTube, and it's an existential relationship for both parties. Susan Wojcki (ZP note: Susan became CEO of Youtube in 2014 and announced his step down in 2023) used to describe YouTube as a music service, which confuses people until you see the data.

Universal Music is very angry with the AI on YouTube. YouTube reacts. It builds a bunch of tools. It writes a constitution about what AI will and won't do. There is outrage about web search generation experiences or AI overviews. Google's reaction was different. I wonder how you balance this circle?

Sundar Pichai: It's far from reality. Generally speaking, when you look at how close we approach the search build experience, even in moments like this, where we spend time testing, iterating, and prioritizing approaches, and the way we've been doing that over the years, I definitely disagree with the view that we don't listen. We care and are willing to listen. People may not agree with everything that we do, and when you run an ecosystem, you're balancing different needs, and I think that's the essence of what makes a product successful.

Nilay Patel: Let me talk about the other side of the question, there's a voice that people are always going to manipulate search, it's always going to happen, it's a chicken-and-egg issue.

The other thing I'm seeing is that the web is being inundated with AI content. There was an example a few months ago where some shady SEO figures said, "I'm stealing a lot of traffic from my competitors." I copied their sitemap. I fed it into an AI system and had it generate copy for a website that matched their sitemap, and then I built this website to steal a lot of traffic from my competitors. "I don't think it's a good result. I don't think we want to motivate that in any way.

Sundar Pichai: No, no-

Nilay Patel: It's going to happen on a large scale, and the internet that we experience is going to be more and more synthetic in some important way. How do you build a system that creates synthetic content for people on the one hand, and ranks it on the other so that you get only the best? Because at some point, for a lot of people, the defining boundary was, "I want something that humans make, not something that AI makes." ”

Sundar Pichai: I think your question has multiple parts. First of all, how do we distinguish between high quality and low quality? I actually think of it as our mission statement, and that's what has defined search for years.

I actually think people are underestimating... Whenever you go through these disruptive platform shifts, you go through phases like this. I saw that team put so much into it. Our entire Search Quality team has been working on our ranking system over the past year, among other things, to better understand what high-quality content is. If I think about the next decade, those who can do it better, I think, will win in the end.

I think you're right about people... People will value the experience that humans create. I hope the data will prove it. We have to be careful every time we face new technology. There are filmmakers, and if you talk about CGI in films, they can be very emotional, and there are still some respected filmmakers who never use CGI to shoot films. But there are also people who use CGI and make great movies. So you might use AI to lay out and enhance the video effects in your video.

But I agree with you, I don't think using AI to mass-produce content without adding any value is not what users need. But there's a big continuum, and users are adapting over time. We're trying to make sure we're doing it in a responsible way, but we're also listening to what users think is high quality and trying to find the right balance. This continuum will look different in a few years than it does today, but I think I think the essence of search quality is this. Am I confident that we will be able to handle it better than others? I think that's the definition of what we do.

Nilay Patel: For listeners, there's been a lot of subtle bashing of OpenAI today.

Can I put it into practice? I just did this search. It's a search for "best Chromebook". As you know, I once bought my mother a Chromebook Pixel. It's one of my favorite tech purchases I've ever made. It's a search for "best Chromebook". I'm going to click "generate" at the top and it's going to generate the answer, and then I'm going to do something terrible, and that is, I'm going to give my phone to the CEO of Google, and it's my personal phone, don't look through it.

You look at that – it's the same generation as what I've seen before, and I asked it to give me the best Chromebook, and it says, "Here's some of the things you might be considering." "And then you scroll down the screen, and it's just some Chromebooks, and it doesn't say they're the best Chromebooks. And then there are some titles, some of which are Verge's, "Here are some of the best Chromebooks." "It feels like an exact thing that an AI-generated search can answer better. Do you think it was a good experience? Is this a signpost or an end?

Sundar Pichai: I think, you showed me a query that we didn't have auto-generated AI.

Image source: Decoder

Nilay Patel: There's a button there that says, "Do you want to do this?" ”

Sundar Pichai: Let me retort, there's an important distinction here. There's a reason we're giving a view that doesn't generate an AI overview, as a user, you're initiating an action, so we're there to respect the user's intent. When I scroll, I see the Chromebook. I also see a whole set of links that I can click on and they tell me all the ways you can think of a Chromebook. I see a lot of links. We don't show an overview of the AI in this case. As a user, you're generating a follow-up question. I think that's right, we respect the intentions of our users, and if you don't, people will go somewhere else.

The future of Google search

Nilay Patel: But what I'm saying — I didn't write, "What's the best Chromebook?" I just wrote "Best Chromebook – but something that claims to be the answer, not on that page." The leap from "I had to press the button" to "Google pressed the button for me and then said what it thought was the answer" was very small. I wonder if you think a page like today is the end of the search experience, or if it's a signpost where you can see a better future version of the experience.

Sundar Pichai: I think it's hard to fully predict how these things are going to play out. Users are constantly evolving. It's a more dynamic time than ever. We're testing all of that, and that's where we're not triggering the AI overview, because we feel like our AI overview isn't necessarily the first experience that we want to provide for that query, because behind that might be a better first impression of the user – all of which are quality trade-offs that we're making. However, if a user requests a summary, we will summarize it and provide a link. I think that seems like a reasonable direction to me.

Nilay Patel: I'll show you another example of it auto-expanding. I only have screenshots of this, and I don't think I've chosen it entirely. This is Dave Lee of Bloomberg, who did a search. He got an AI overview, and he just searched for "JetBlue Mint Lounge SFO". It just says the answer, and I think that's good, that's the answer.

If you swipe it, you'll see the website it got from. It's a verbatim rewrite of that website. That's what I'm talking about. An AI-generated overview of that answer, if you just look at where it came from, is almost the same sentence as the source. I mean. At some point, the better experience is the AI overview, which is what is present on all the sites below. This is the same information.

Sundar Pichai: About search – we deal with billions of queries. You can definitely find a query and give it to me and say, "Can we do better on this query?" "Absolutely. But in many cases, part of the reason people respond positively to the AI overview is that the summaries we provide clearly add value and help them think about things they may not have thought about. If you add value at that level, I think people will notice over time, and I think that's the standard you're trying to achieve. Our data shows that over 25 years, if you haven't done something that users find valuable or enjoyable, they'll tell us right away. Again and again we see this.

With this shift, everything is going in the opposite direction, and it's one of the biggest quality improvements in the product we're pushing, and people value that experience. There is a common assumption that people don't know what they're doing, and I strongly disagree. People who use Google are smart and they understand what is being done. So, for me, I can give a lot of examples of how I use AI as a user at a glance. I'm like, "Oh, that's providing context. Oh, and maybe there are these dimensions that I didn't even consider in my original query. How do I expand it and see it? ”

Nilay Patel: I saw OpenAI show GPT-4o, Omni a few days ago. It looks a lot like the demo you showed on I/O. This concept of multimodal search, where you have this character that you can talk to—you have Gems, the same idea—feels like you're in a race for the same results for the search experience or the agent experience, do you feel the pressure from that competition?

Sundar Pichai: It's no different than Siri and Alexa, when you work in the tech industry, I think we all felt the relentless innovation a couple of years ago, and all of us were building voice assistants. You can ask the same version of this question: what is Alexa trying to do, what is Siri trying to do? It's a natural development, and I think you have a new technology right now, and it's moving at a rapid pace. I think it's been a big week for technology because there's a lot of innovation, like on Mondays and Tuesdays and so on. That's how I feel, and I think it's going to be the case for a while.

Nilay Patel: You'd rather be in a place where the underlying technology is evolving, which means you can radically improve the experience you're launching.

Sundar Pichai: You can also move forward quickly at any time. Many of us have a vision of what a powerful assistant can look like, but we can't achieve that goal due to the limitations of the underlying technology. I think the technology we have now is better able to serve that goal. That's why you see progress again. I think it's exciting. For me, I look at it and say, "We can really make Google Assistant better." You see this with Project Astra. When I use it, it's just incredible magic for me, so I'm really excited about it.

Nilay Patel: This brings me back to the first question I asked: language and intelligence. To make these products, I think you need a certain level of intelligence core. Do you have a measure in mind, "When it's good enough, I can trust it"?

On all of your presentation slides and all of your OpenAI presentation slides, there's a disclaimer that says "check this information," and for me, when it's ready, you don't need that anymore. You don't have "Check this information" at the bottom of the 10 blue links. You don't have "Check this information" at the bottom of the featured snippet.

Sundar Pichai: You've touched on a deeper problem, and the illusion problem is still an unresolved problem. In some ways, this is an inherent characteristic. This makes these models very creative. That's why it was immediately possible to write a poem about Thomas · Jefferson in Nilay's style. It can be done. It's very creative. But LLMs aren't necessarily the best way to get factual, which is part of the reason I'm excited about searching. Because in search, we're introducing LLMs in some way, but we're reinforcing it with all the work that we've done in search and layering it with enough context so that we can provide a better experience from certain angles. But I think the reason you see those disclaimers is because of their inherent nature. Sometimes it still goes wrong, but I don't think I should underestimate its usefulness because of that. That would be the wrong idea.

Google Lens is a good example. When we first introduced Google Lens, it didn't recognize all objects very well. But year after year, the curve has become very dramatic and users are using it more and more. Now we've made billions of queries with Google Lens. This is because the underlying image recognition, coupled with our understanding of knowledge entities, has expanded significantly over time.

I think it's a continuum, and again, I think users vote with their feet. Fewer people use Lens in the first year. We also didn't put it everywhere because we were aware of the limitations of the product.

Nilay Patel: When you talk to the DeepMind Google Brain team, is the solution to the hallucinations on the roadmap?

Sundar Pichai: It's Google DeepMind.

Sundar Pichai: Are we making progress? Of course, when we look at factual metrics year after year, we definitely make progress. We're all making it better, but it's not worked out yet. Are they working on interesting ideas and approaches? Yes, but time will tell. I would think of LLMs as one aspect of AI. We're looking at AI in a broader way, but it's an area where we're all trying to push for more progress.

Nilay Patel: In five years there will be a paradigm shift in this technology, and I have a hunch that we will definitely go through this phase. For you, what does the best web version look like in five years?

Sundar Pichai: I would like the web to be richer in terms of modality. Today, I feel that the way humans consume information is still not fully encapsulated on the web. Today, things exist in very different ways – you have web pages, you have YouTube, and so on. But over time, I want the web to be more multimodal, richer, and more interactive. It has more statefulness that it doesn't have today.

I think that while it's fully acknowledged that people may use AI to generate a lot of spam, I also feel that every time there's a wave of new technology, people don't quite know how to use it. When mobile devices came, everyone crammed web pages into mobile apps. Then, later, people evolved into true native mobile apps.

The way people can actually solve new things, new use cases, with AI is yet to come. When that happens, I think the network will become richer as well. So: dynamically compose the UI in a way that makes sense to you. Different people have different needs, but today you're not putting that UI together dynamically. AI can help you do this over time. You can also use it wrongly, wrongly, and people may use it in a superficial way, but there will always be entrepreneurs who figure out a really good way to do it, from which great new things will be created.

Nilay Patel: Google is a huge incentive for web development through search, Chrome, and everything you do. How do you ensure that these incentives align with these goals? Because probably the most important thing here is that the web ecosystem is in a moment of change and Google has a lot of trust to build and rebuild, how do you think about making sure those incentives are directed towards the right goals?

Sundar Pichai: Look, not everything is under Google's control. I wish I could influence one of the hardest experiences I have today when visiting a website as a user – you have a lot of places where you need to accept cookies, dialog boxes, etc. So I'd say there's a lot of things outside of that, you can survey 100 users.

But what kind of incentives do we want to create? I think it's a complex question of how do you reward originality, creativity, and independent voice, at any scale you can, and give it the opportunity to thrive in this content ecosystem? That's what I'm thinking. That's what the search team is thinking. But I think that's an important principle, and I think it's important for the network, and it's important for us as a company.

Nilay Patel: Awesome Sundar, thank you so much for taking the time to participate in Decoder.

Sundar Pichai: Thanks, Nilay.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com