Next year will mark my 20th continuous year living in Beijing—a city which, for all its faults, also does have quite delicious food on offer everywhere. I was born and raised in the US, but in my teens I took a very keen interest in what was happening in China, and realizing that I had an opportunity to witness up close the sweeping transformation of a full fifth of humanity, in college I shifted the focus of my study toward China, and spent a year living in Beijing after graduating. I then plunged into the study of modern Chinese history as a graduate student, and in all the time since, my fascination for history hasn’t diminished. Long ago I recognized that studying the history of one people or one nation is woefully inadequate to gaining any real, useful understanding: Comparative studies are really the only way to recognize patterns, to grasp their significance, and begin to comprehend the underlying reasons for divergence. For students of modern China, one of the touchstones for comparison has always been the Ottoman Empire and the emergence of modern Turkey, and I’m grateful finally to experience the sights and sounds of this place I’ve been reading about for so many years. I’m deeply impressed at the obvious success that Turkey has had in creating such a vibrant, cosmopolitan, and truly modern society that’s still deeply and meaningfully connected to tradition and history.

Comparisons are very relevant too today, and I’m particularly interested in looking comparatively at the development of technology and its impact on society in different countries. Unfortunately my knowledge of what’s happening in the digital world in Turkey is superficial, limited to stray observations, anecdotes, and some scattered readings, so please forgive any mistakes I may make.

But with China, and the Chinese Internet in particular, I’m very fortunate to have witnessed its development up close and in person, first as one of the early, frustrated users of the Internet—I think my family was among the first few thousand Internet subscribers in China, from early 1996—then as someone working in an Internet startup, then later as China bureau chief for a US technology magazine, and finally working in communications for leading Chinese Internet companies. In the time I’ve been involved in the Internet, the number of users has soared from under a million in 1999 to close to 700 million today. I’ve been privileged to watch as the Internet in China became really the crucible of contemporary culture, developing its own quite unique culture and becoming, at least for a while, a de facto public sphere in Chinese life—something that really had never existed previously at any scale. And I’ve had the opportunity to meet and get to know many remarkable entrepreneurs, some of whom now lead multibillion-dollar companies.

So this afternoon, as I share some observations about major trends in digital technology in China today and speak a bit about China’s rise as a digital superpower over the last 15 years or so, I hope that you’ll find in some relevant points of comparison with your own home markets—whether here in the region or further afield. I want to focus on two major developments that are already reshaping the digital landscape in China, and in ways that are arguably ahead of what’s happening even in the U.S. and may be of some interest here in Turkey.

One isn’t unique to China: It’s the Big Data Revolution that you’ve doubtless heard so much about, but in particular I want to give some examples about how Baidu is using big data and artificial intelligence to transform our business.

The other will I think be of even greater interest: What we in China call O2O—a name actually coined in the U.S., meaning “Online to Offline.” In the U.S. and in other markets, people refer to it as “on-demand services,” but I think “O2O” really gets at it better. Of course those of you from here in Turkey know Yemeksepeti.com, which was acquired by Delivery Hero for 589 million dollars earlier this year. It’s exactly the kind of O2O play I’m talking about, leveraging the power of networked smartphones to connect people with services, and it’s the marketing treasure from China (where there are many, many such companies) that I’m most keen to share with you. It’s fascinating because it offers a way to truly unlock the local layer, and unlock services—things that really weren’t possible to nearly this extent before—and because it promises to liberate Internet companies from their almost total dependence on advertising revenues. It’s also interesting in a comparative sense, as it is not something that has developed to the same extent in the U.S. and represents, in a way, the kind of “leapfrog” effect we’ve seen now at work a few times now in China.

A comparative approach to understanding the development of tech ecosystems has a lot in common, I think, with a comparative approach to understanding the emergence of institutional, legal, intellectual and political features—and at a deep level the two are  certainly related. Bringing people together at conferences like this, held I think very significantly at the historic juncture of Asia and the West, is a great opportunity to take in the enormous diversity and to learn from where each market seems to excel. I’m quite sure that historians in the future will find it as challenging and complex to explain the uneven development of digital technologies globally in the late 20th and 21st centuries as they do today in explaining the disparate forms of government or governing institutions, of rule of law, of different socio-political norms across different cultures and countries. It’s no easier to grasp why certain geographies have become leading centers of tech innovation and others have not than it is to understand precisely why the European Enlightenment centered on France, or why 800 years before it there was an age of enlightenment led by Persian and Turkic scholars in the great medieval Central Asian cities of Bukhara and Samarkand, or why England led the way in the Industrial Revolution.

When it comes to comparative approaches to Internet markets, probably the U.S. and China is a more meaningful comparison than, say, China and Turkey. These are after all countries and markets on a completely different scale: Turkey’s Internet penetration is about 47%, at about the same level as China’s (which is 49%), but that still gives China almost 20 times the number of Internet users. China’s enormous scale is among many factors that have made it possible for China to develop a quite different Internet ecosystem than most of the rest of the world, which with few exceptions is dominated by US-based Internet companies. In search, in e-commerce, in social networking, in microblogging, in online video and in just about every other sector of the Internet China’s market leaders are its own companies and not, say, Google, Amazon, Facebook, Twitter, and YouTube. And yet China’s Internet companies have been astonishingly poor at establishing any meaningful markets outside of China. Beyond a small community of Chinese living here in Istanbul, I wouldn’t think any of you who are from Turkey are particularly familiar with the Chinese category leaders in all of these areas.

For that reason I’ll take a few minutes to point out some of the salient features on the map of digital China. I suppose that there’s one that everyone’s aware of: The Great Firewall of China. That’s a metaphor that people understandably find completely irresistible. But its sheer ubiquity has led to some fundamental misunderstandings about the Internet in China. First and foremost, it has become in the minds of many a kind of “Iron Curtain 2.0,”to borrow a phrase. Here in this room, where everyone is doubtless aware of just how well-developed the Chinese Internet is, it might not be the case; but you’d be surprised how many people I meet who, because they equate the Great Firewall and the Iron Curtain, imagine a situation in China more or less like Eastern Europe before the collapse of Soviet communism: They perhaps envision people standing in virtual bread lines, clicking on boring state propaganda sites and nothing else. Then of course they’re shocked to hear that one single company, Alibaba, racked up $14.3 billion in e-commerce sales on “Singles Day,” November 11—compared to only $3 billion for all e-commerce in the entire US on Cyber Monday, 19 days later. They’re surprised at the incredibly vibrancy of Internet culture in China, and shocked to learn that four of the top 10 largest global Internet companies by market cap are Chinese:Alibaba, Tencent, Baidu, and JD.com.

Has censorship conferred competitive advantages on Chinese Internet companies? Yes—but probably not to the extent that you imagine. If you look at the period before the major US-based Web properties like Google, Facebook, Twitter and YouTube were blocked, actually they were all quite far behind their Chinese counterparts in terms of market share: Google, at the time that it announced it would no longer censor in accordance with Chinese law, had about 17% while Baidu had 80%; Facebook was a distant also-ran behind Renren.com, 51.com, and a site called Kaixin001. Twitter was far behind Fanfou.com and other Chinese microblogs, and YouTube was probably the 7th or 8th place video site in China behind a slew of video sites.

It’s probably beyond today’s scope to delve into all the reasons why China achieved what it did, but I think a few are worth noting: First, there was the relatively enlightened policies of the government, which spent when it needed to on infrastructure, was quite liberal in its interpretation of rules that would have barred foreign investment in media companies; and who deliberately courted US-educated technical talent and encouraged overseas Chinese to come back and get in on the opportunity. I think people are so used to thinking that the Chinese government is somehow “opposed to” the Internet that they forget how much has been done under each administration since Jiang Zemin to advance its development.

Kind of like the formation of planets, once pieces started falling into place together it exerted a kind of gravity. So with infrastructure and tech talent, you also attracted venture capital, and managerial talent, and all the other things that are necessary for a strong tech ecosystem to develop. The presence beginning in the mid-2000s of all that manufacturing, and especially if the high-end OEMs and ODMs, meant fast iteration of new hardware ideas and a lot of incentive for software companies to match and test the limits of hardware innovation. Of course there were attitudinal and cultural ingredients that were needed, too, and in the case of China a spirit of entrepreneurship was not lacking, the kind of acceptance of failure—the ability to wear failure as a proud badge, as you see in Silicon Valley—didn’t take root easily in a society still so wedded to saving face. Now, though, it really does seem to have taken root, and China now boasts a tech ecosystem that may yet rival Silicon Valley’s.

I would be wary of anyone who tells you breathlessly that China is flat-out more innovative in Internet (or, as you’re more likely to hear, even in mobile Internet) technologies than the U.S. But I would be even more wary of people who disparage Chinese tech companies and dismiss their innovation. The fact is, in mobile especially there really are some areas where the world can really learn from China. The fact is also that many of the opportunities for this kind of innovation arose, ironically, from what might be called backwardness: The relatively low rate of fixed-line Internet and PC penetration was an opportunity for mobile to leapfrog, as we’ll doubtless see happen in much of the rest of the developing world. The relative underdevelopment of China’s services sector, which is quite fragmented and has low rates of Internet penetration, has allowed for O2O—what the Americans call “on-demand services”—to really take off in China.

One of the things that I believe was really responsible for the extensive innovation we’ve seen in the mobile Internet in China is the absence of Google Mobile Services, the agreement with Android handset makers in most of the rest of the world requiring them to use Google’s own properties like search and maps on Android phones. In much of the world, mobile maps simply means Google or Apple Maps. Here in Turkey I understand that the Russian search giant Yandex has done a good job with maps, and it’s great to see you have choices, at least as long as their fighter planes stay out of Turkish airspace, but this level of choice is the exception for most of the world. In China, when Google withdrew many of its services in 2010, and when remaining mobile services became mostly inaccessible for Google a few years later, there was a great opportunity for other platform-level apps—and especially maps, which are fundamental to O2O services. Baidu wasn’t the only map provider in China, either: AutoNavi, which was acquired by Alibaba last year, is another major map provider, and Tencent—the third of China’s trifecta of major Internet powerhouses—has its own mapping service as well.

Anyone with even a glancing familiarity with the Internet landscape in China knows these three major Chinese Internet players, with a combined market cap of about 450 billion dollars. The three are known by the shorthand BAT: Baidu, Alibaba, and Tencent. Back in the days of the PC-centric Internet, sure, the three had a bit of overlap and certainly competed to an extent, but for the most part, Baidu as a search engine was mainly about connecting people with information, Alibaba as the giant on the e-commerce scene was about connecting people with products and with merchants, and Tencent, as the most powerful social player, was about connecting people with other people.Things changed rather dramatically beginning in 2012, though, when smartphone penetration began to increase dramatically and mobile Internet use became the norm. Suddenly, China’s Internet giants—as well as a slew of other companies—recognized the opportunity to connect people with services, and direct competition really heated up.

The transition from being PC-centric to mobile-first companies happened fast for China’s major Internet players. Baidu was criticized in some quarters for having started the transition a bit late, but even if that was once true, we’ve certainly made up for lost time. We quickly came to dominate in mobile search, in mobile maps, and in mobile app distribution. And in Q3, our last reported quarter, mobile accounted for about two-thirds of Baidu’s total search traffic, and about 56% of our total revenue.

Our position in mobile was never just about more ad revenue, though. It was about escaping, finally, that dependence on advertising that is the curse of so many Internet companies all over the world. Even Alibaba,which operates China’s largest online marketplaces, surprisingly derives more than half of its revenue from advertising. The upshot is that most Internet companies are competing for a rather small slice of the pie—a slice that’s called “online marketing budget.” And while that’s been sufficient to nourish a good-sized ecosystem to date, and while online advertising (and in our case, performance-based search marketing) will continue to be effective for the customer, useful for the consumer, and vital for us, we believe that our position in mobile opens up a huge new opportunity that we’re now investing very heavily in—the opportunity in services. Where search is terrific for lead-generation and creating brand awareness, what we can now do is close the loop from discovery to payment. We broaden the scope of what we do, then, extending from the familiar search model to include a transactional, commission-based model. In other words, we now want to be dipping our ladle directly in the revenue stream of the customers we’re driving business to.

We’ve long known that the majority of consumption is local, and the majority of consumption is services. This is true not just for China, but for every major market I’ve looked at. But the local and services layers have been notoriously difficult to unlock. IP address-based geo-targeting was always inaccurate and clumsy. Local businesses and services need to target by actual proximity, and need to reach people—and be discoverable by people—at the right moment, when they have that propensity to consume. A truly usable mobile Internet, with affordable devices that are location-aware and always on, has made this possible. It has already begun to transform life for consumers and for merchants alike in Chinese cities, and I believe that there are many parts of the world—and Turkey is very, very promising—where O2O will do the same.

Just how big is this opportunity? Well, if we’re only talking about vertical areas of services we’re now involved in directly—travel, hotel, restaurants, takeout delivery, entertainment (like movie theater tickets)—we’re talking a market of about $1.6 trillion dollars in 2015, likely to double in 4–5 years. And it’s a very under-penetrated market right now: Only 4 or 5% of restaurants in China are making use of the Internet now to reach customers.The more heavily penetrated segments, like movie theater ticketing, are already hugely promising. Some 60% of all movie theater tickets sold now in China are sold online, mostly via mobile devices. For consumers, the convenience and the savings are incredible. In literally under 30 seconds you can find the movie you want to see, find the closet theaters where that film is showing, deep discounts on tickets through Baidu or another provider, select your seats and buy your tickets. No one I know pays full price at the theater. For theater owners this is great: They fill their empty seats and generate much better revenue even if they’re offering deep discounts. A similar revolution is happening in dining, where restaurateurs are all getting on board with O2O platforms both to fill seats in their restaurants and to soak up excess kitchen capacity by offering takeout delivery—companies like you Yemeksepeti. A restaurant owner doesn’t need to create a website, or pay for bandwidth or rack space in an Internet Data Center. In fact a restaurateur doesn’t need to have a storefront: Many are doing great business now through delivery-only models, with just a kitchen and a partnership with an O2O food delivery provider like Baidu. O2O also presents tremendous opportunities to build CRM, real relationships with customers, through stored-value and VIP membership cards for instance.

Why is this happening on steroids in China, and what was so different about China? Part of this, it must be said, is the relative backwardness of the Chinese service sector.While in the developed West, a large percentage of hotels and restaurants are chains, with multiple outlets and back office staff and Internet-savvy marketing teams, China’s services are much more fragmented and don’t have the resources or the skills to leverage digital marketing on their own. Baidu and other companies who’ve led the way in the O2O space are in a great position, then, to work with these local services. And the rebalancing of the Chinese economy from investment-led to consumption-led growth, especially in services, augurs well for O2O’s prospects.

China’s large number of migrant workers, who move to the cities by the millions each year, can find great work, for instance, as a takeout delivery guy: They can make a good wage just following the pickup and drop-off instructions on their phones, just like Uber drivers—only unlike Uber drivers, they only need an electric bike, easily available for $300, and don’t need a drivers license. In America and in other developed markets, the high cost of delivery, and at least in the US case, the disturbing willingness with which Americans will get in their cars and drive even very short distances to their favorite restaurants, is a significant inhibitor to the development of O2O.

Urban density in China also works well for this model. We’re all familiar with the major metropolises of China, like Shanghai and Beijing, each with well over 20 million people. But there are over 250 cities in China with over a million people, and even someone who’s lived in China and studied the country for as long as I have would be hard pressed to name over 75 of those cities.

Of course if you’ve been following the development of the on-demand or O2O sector in China, you’re aware that there are a lot of subsidies being paid out—by Uber, in which Baidu is an investor, by Baidu Nuomi, our services-oriented e-commerce platform, by Baidu Takeout Delivery, and of course by all of our competitors. It’s true that we’re deploying substantial resources to win market share and to shape consumer behavior. We’re confident that we’ve already profoundly reshaped the market, and not just Baidu but the whole O2O sector still has tremendous momentum. All the subsidies we’ve paid out, the couponing that we’ve done, is really a small price to pay to open up a huge new market opportunity not just for us but for local businesses and services across the country. And as long as we’re spending that money to train core competencies—for example, the very sophisticated real-time, multipoint-to-multipoint logistics system for on-demand food delivery we’ve developed for Baidu Takeout Delivery—and to really create a compelling value proposition for merchants, we believe it’s money well spent. 

So this is the great Chinese marketing treasure I wanted chiefly to share with you today—a way to unlock the riches of the local and services layer through O2O services. Here in Istanbul, as my wife and I have explored the city over the last five days, we’ve seen no real impediments to this happening here, too: You have here an enormous, populous city with one of the highest restaurants-per-capita I’ve seen, if my gut is correct. You have a huge, high-quality, but also quite fragmented services sector. In tourism alone the opportunities are enormous. Yemeksepeti is great, from what I can tell, and my hat’s off to the entrepreneurs who created and sold it. But imagine an app that would allow you from a single interface to book the right hotel, hire a driver, a tour guide who speaks your native language, recommend restaurants ranked highly among trusted people in your social network, offer you great nightlife options and more. If you were a hotel operator, a tour operator, a restaurateur or a nightclub owner, you couldn’t afford not to be using such a platform. This already exists in China: If you read Chinese and you visit any Chinese city, you can already do pretty much all these things and more directly from search or from maps.

Services means of course more than just restaurants, food delivery, travel, and movie theater ticketing. The really big prizes are in major verticals like financial services including insurance, like healthcare, and education—and these are all areas into which the major Chinese Internet companies have made very aggressive moves. And in these areas, it’s not just the power of mobile, and maps, and location-awareness that Internet companies are able to effectively leverage. It’s the power of big data.

Search engines like Google and Baidu were of course dealing in “big data” long before anyone started using that already rather shopworn phrase. But it’s only been in more recent years that we’ve seen real breakthroughs in artificial intelligence. Whether you realize it or not, AI now touches the daily lives of many people, and will do so more and more. Since he joined Baidu in May of 2014 I’ve gotten quite close to Andrew Ng, Baidu’s chief scientist. Andrew is one of the major luminaries in AI right now, and particularly in one specific approach to AI called Deep Learning, which uses very large scale Deep Neural Networks to allow machines to actually learn. Some of the significant uses to which this has already been put include speech recognition and computer vision, but it’s going to revolutionize semantic intelligence and therefore machine translation, and bring about unforeseen changes in our abilities to process and extract insight from data.

If you’ve followed the AI field in recent years, you’ll see how hot deep learning is now, and you’ll also see that the world’s great deep learning talent has amassed at major Internet companies. Without going into too much detail, deep learning’s basic ideas have actually been around for decades—at least since the late 1980s and early 90s. But it’s been two things that have enabled the kind of breakthrough we’ve seen just in the very recent years: The rise of high-performance computing that has allowed us to build very large neural networks; and the massive availability of data. Andrew Ng, who headed the Stanford AI Lab and founded the Google Brain project, as well as co-founding the massive open online course platform Coursera, has led very impressive work in the time since he joined Baidu. He likens the AI systems themselves—the deep neural nets and the deep learning algorithms—to “rockets,” and the oceanic amounts of data to “rocket fuel.” Neither is much good without the other. It’s the large Internet companies—Google, Facebook, Baidu, Apple, Microsoft—who have both the rockets and the rocket fuel: They have combination of financial resources, hardware resources, and data resources that AI scientists really need. And we’re really achieving lift-off—with an impact that will really change the world.

As Andrew likes to point out, the great thing about deep learning as an approach to AI is that the bigger the neural networks—the more neural connections—the better the performance. Size actually matters. And the more data you put through a system, the better it performs too. With very little modification, we’ve been able to use a deep learning system we call Deep Speech to recognize speech in both English and in Mandarin, by feeding it thousands of hours of transcribed speech in both languages.

Marketing is probably the most obvious application of AI, and it’s one we worked on right away. A deep learning system can figure out pretty quickly the likelihood of a given search ad being clicked on, and this ability to predict click-through rate obviously makes not only for better ROI for the advertiser but also a better user experience and better revenues for the search provider. We’re already also using deep learning to add images to text ads, automatically finding relevant images from the customer’s website, cropping and resizing to generate optimal click-through—and that’s already creating significant value for us and for our advertisers. AI plays a significant role in recommendation across many of our products, including O2O platforms: We can pick out interesting and non-obvious correlations that show up consistently in the data, and use those to make better recommendations.Who would have known, for instance, that Chinese men who drive Audis are far less likely to eat western for than Chinese men of the same age who drive BMWs?

Besides smarter ad serving, recommendations, computer vision and speech recognition, there are other surprising applications of deep learning that we discovered when we turned our deep learning system into a platform for engineers across the company to access. “Platformization” is a really important concept at Baidu. It’s something of an obsession for Jeff Bezos, CEO of Amazon, and you may have heard him talk about it. We agree, and it’s something we implement aggressively.

Let me give you an example of “platformization” at work. At Baidu we created something called PADDLE: it stands for Parallel Distributed Deep Learning platform.Basically, it allowed R&D personnel from across the whole company to make use of our deep learning engine by giving them relatively easy APIs to use. In very short order we started seeing deep learning applied to a wide range of things: Our Infrastructure team used deep learning to predict crashes of hard disks in our data centers, so that with 85% accuracy they could identify disks that were going to crash 24 hours before they did. This actually saves us about millions of renminbi per day. Another team used deep learning to develop a cloud-based antivirus system that could detect malware before it had actually been observed and defined by traditional antivirus companies, with rather effectiveness and so far a vanishingly small percentage of “false positives.” We’re using it now in healthcare, too—for diagnosis from just a layman’s description of symptoms. 

I want to finish today by talking about one final thought on lessons from the Chinese Internet, one last idea that I think will be of value to people here in Turkey as you look to take your own valuable treasures out into the world.

When you work at a technology company—whether it’s in the Silicon Valley or Shenzhen, in Austin or Boston, in Beijing or Bangalore—it’s easy to get caught up in the obsessive culture of innovation: To be focused constantly on that next dazzling piece of tech, that thing you hope is going to catch fire with the elusive early adopters. We spend most of our workday interacting with educated, tech-savvy young people, and those of us in our forties—if you’re anything like me—are always worried that we’re out of touch with those young digital natives and their fickle, fast-changing habits. We train our sights too often on them, focusing on the shiny tech, not realizing that the treasures that sparkle brightest aren’t always the most valuable.

Living in a country like China, or anywhere else in the developing world, can remind you that there’s a whole lot more that should be driving innovation—that we shouldn’t be thinking just about putting the next wow-factor gadget or app into the hands of tech influencers. Technology ought to be put to work eliminating inefficiencies, democratizing access and putting the means of production and creation into the hands of the many, freeing us all from drudgery. And in China all around you you see examples of how those jobs are far from finished.

I believe that technology companies should be dedicated to solving these problems—at serving underserved people, and putting the information and the services they need within their reach—but not just because it’s the right thing to do. It is, of course, the right thing to do, and I do lament that it’s something that Silicon Valley seems to have moved away from a bit. So many of the hot new startups seem to me to be solving problems mainly for other affluent and educated people living in the first world.

In China, it’s different, and that’s good—because I also believe focusing on serving the underserved is the smartthing to do: Serving the underserved is an impetus to innovation.  I’ve come to realize that Chinese technology companies enjoy what I call The Straddler’s Advantage—the advantage of standing astride two very different markets: One that’s very much like the developed world, and one that still resembles the developing world. I believe that Turkey is in a similar situation.

The first tier cities, cities like Beijing, Shanghai, Guangzhou and Shenzhen, all have very high rates of Internet penetration—close to or even higher than the national average of OECD countries. There are sophisticated users, sophisticated customers, and the tech companies themselves are very much current when it comes to technology: They keep up, despite what you may have heard about China having an almost separate “Intranet,” with all the new companies and all the new technologies coming out of Silicon Valley and everywhere other innovative geography.

Outside of the major cities, it’s a different story entirely:Not only is Internet penetration much lower, but the level of sophistication of users is much lower as well. There are many impediments to using technology. Strong regional dialects or even different Chinese languages are spoken, and while the written languages may be almost identical, input method based on standard Mandarin pronunciation is difficult for some to master. People are more sensitive to bandwidth costs.

It’s not just a urban-rural divide, either: There’s a major age divide that has created a large underserved population in China, among people who never learned the romanization system that’s been taught in Chinese schools since the 70s and that the overwhelming majority of people now use to input Chinese characters on digital devices. A great many people in China over, say, 60, never learned that system, called pinyin, so simply entering text on a device becomes very difficult.

What’s needed, clearly, are simpler and more intuitive interfaces. But serving users like these isn’t a simple matter of taking existing technology and just “dumbing it down.” To solve problems of access you need to go smarter, in fact. It actually places even greater demand on you as a technology provider. Graphic User Interfaces opened up computing to billions of people who might never have mastered DOS. The touchscreen interfaces on our iOS and Android devices are so incredibly intuitive that I’m sure many of you with children or younger relatives remember seeing two- and three-year-olds familiar with those smartphones and iPads going up to framed photos and trying to expand them with their fingers. There’s still a need for more intuitive, more naturalistic interfaces. For inputting Chinese, it was really revolutionary when smartphones allowed writing of Chinese characters, and those systems have developed very well—though they’re hardly very efficient still, and much slower than someone who knows pinyin well. To me, the most natural, most intuitive interfaces are perfectly obvious: Voice and visual.

We’re doing some very exciting work in our Silicon Valley AI Lab in something we call Deep Speech, which I mentioned earlier: Using Deep Learning algorithms in very large neural networks to process huge amounts of recorded speech, so that the system can actually learn to recognize words spoken all sorts of different ways—in noisy environments, with all the hesitation sounds and the pause words and the false starts, stammers, and repetition that vex most of the current systems that have been deployed. We’ve gotten truly amazing results in both English and Mandarin.

But even assuming that we can get speech essentially solved, and solve it in multiple languages, we still have to make actual sense of what’s being said: We need to derive intent from the commands that are given, and for that we need real semantic intelligence. After all, it’s not truly intuitive to have to use Boolean algebra with excluder and connector words for all the search commands we read into a smartphone. No, instead, we need to think of the way your older parents or grandparents use search: They’ll type in full sentences, like “I wonder whether they still have that $1.99 breakfast special down at the International House of Pancakes.”

It means better computer vision—the ability to actually extract meaning from images. After all, all of our smartphones now come equipped with quite good cameras. There’s a little camera icon in the search bar on Baidu’s search app that will let you snap pictures of things and do searches based on inputs. Point it at a QR code and it will resolve to the corresponding page. Take a photo of a book cover and it’ll take you to online reviews and online book stores. Same basic thing if you point it at a CD. But then it gets really cool: Point it at a dress, a blouse, a shirt, a sweater, a jacket someone’s wearing or a handbag they’re carrying and it does a pretty good job of finding matching items for sale on China’s myriad e-commerce sites. Point it at a movie poster and it’ll show you nearest theaters showing that film, and let you pick your seats and buy your tickets in a couple of clicks. And of course you can find similar images to photos you take or upload. When you do that, the results you get aren’t just images with visually similar patterns and shapes and colors: We actually extract semantic information from pictures and give you results based on those. So if you upload, as I recently did, a photo of the Sultan Ahmed Mosque, results don’t just include pictures of the same mosque, but pictures from areas where it’s located, perhaps of the Hagia Sophia museum, and pictures of other Islamic religious architecture which may look completely different. There’s Deep Learning at work here, too.