Image source @unsplash
As a map controller, I recall the surprise and excitement after finding my birthplace in Google satellite maps. I still sigh that technology has brought the minds of every ordinary person - just one computer can have a "God's Perspective" and you can travel around every corner of the world by sliding the mouse gently.
However, the last time I spent time, I visited world-renowned attractions and streets through Google satellite maps and street view maps. It was more than ten years ago.
For Alphabet, Google Maps is still the most core product besides search, Android, Google Store and YouTube. According to analysts forecasts, Alphabet's advertising revenue from Google Maps is likely to reach $3 billion to $4 billion in 2018, while it may still grow at a rate of 25% to 30% per year in the next three years.
On February 8 this year, Google Maps celebrated its 15th birthday. For this anniversary, Google maps have undergone major revisions and upgrades on iOS and Android. At the same time, Alphabet and Google CEO Sandel Pichai published a blog post specifically for this purpose, saying that in the future, Google Maps will be an important position in the company's AI First strategy.
Sandel's statement seems to be conveying two layers of information: the first layer is that in the five years since Google announced its investment in "AI First", Google Maps has not become an important base for AI technology support; the second layer means that Google Maps will focus on the application of AI technology in the future.
Since Google announced its full embrace of AI in 2016, all Google's series of products have achieved AI upgrades through machine learning and deep learning technology. Compared with the highly anticipated star products such as Waymo, Google Assistant, Gmail, Google Translate and other star products, Google Maps does seem a little bland and slow in the application of AI technology.
slowness does not mean doing nothing. As Google Maps, which covers 1 billion users around the world, the AI of its products has been underway. Being plain does not mean not worthy of attention. The idea of Google's AI-based products is to gradually penetrate AI technology into its products to improve user experience.
In the field of technology, we always overestimate what can be done in a short period of one or two years, and always underestimate what can be done in five or ten years. This is true for search engines, driverless cars, and so is Google Maps. In the long cycle of five years, some of Google Maps' technical depth deserves our attention again.
Google Maps, Born AI?
Time goes back to 2016. At Google Event 2016, Sundar Pichai, then CEO of Google, announced that Google's overall strategy shifted from "Mobile First" to "AI first". As we all know, Google is well prepared to transform its AI strategy.
Google began to plan machine learning as early as 2010; in 2012 it released the "Knowledge Graph" and built the super-large neural network system "Google Brain"; in the following two years, it included deep learning company DNN Research and the leader of British artificial intelligence DeepMind; then, Google's AI achievements were like a blowout. In 2015, Google's open source deep learning framework TensorFlow was successively applied to more than 50 Google products; in 2016, AlphaGo developed by DeepMind emerged, bringing AI into the public's vision.
We know that the development of AI requires the support of algorithms, computing power and big data. Among them, the map itself has a natural fit with AI.
On the one hand, the map products are mappings to the real world. Many people’s offline movement trajectories will be reflected through this product and accumulated into data. The same is true on Google Maps. Whether it is the billions of user travel records every month, uploaded location label data, PB-level satellite map data and street scene image data taken globally, they can be called "rich mines" in big data.
On the other hand, the function of the map itself can also be applied by a large number of AI-related industries. For example, smart cities cannot do without using map products to track traffic and people flow. Autonomous driving also requires combining high-precision maps with radar sensors.
Therefore, for Google, which is widely involved in AI, maps naturally become a good technical test site.
The most typical one is the utilization of Google Street View data.Google Street View was originally a special project of Google Maps. It was taken by a dedicated street view vehicle, and then the 360-degree real-life photos were placed in Google Maps for users to use. Soon when Google began to get involved in AI, especially when Li Feifei, who had trained on imageNet, joined the , , these huge amounts of data, images from the real world, had a place to use.
, for example, uses a reinforced deep neural network to scan thousands of street scene images in California and then convert them into professional-grade photos. For example, use face recognition and OCR recognition functions to code the license plate numbers and faces in street scene images to achieve the purpose of privacy protection.
Under the same method, detailed information such as street numbers, company names, traffic speed limit signs, etc. on the roadside are also continuously extracted from the pictures, and new addresses are appropriately created and located on Google Maps. The model can also be applied to name recognition on the outdoor wall of the merchant. Through this function, merchant changes can be continuously updated more accurately.
In other words, AI has been optimizing the underlying technology of Google Maps.
(no real address information is required, the system automatically recognizes the merchant name "Zelina Pneus")
In addition, at the Google I/O 2018 developer conference, Google Maps demonstrated the walking navigation system (VPS) based on AR technology for the first time. The AR system actually uses Google Lens products based on image recognition and OCR technology. It can capture and recognize landmark buildings through the mobile phone camera to achieve navigation. Just click the "start AR" button, and Google Maps will display the video image overlapping the forward arrow and the real shot of the camera to the user. Behind the flexible and simple AR navigation on the front end of
, the backend requires a large amount of street view data computing support; if the exact user location cannot be obtained through GPS, Live View will also use machine learning to compare the scenes captured by the user's camera, and then compare them with billions of street view images for repositioning.
In addition to images, Google also uses big data recommendations and predictions to provide Google map AI-based service experience directly for users. Google Maps provides AI-specific recommendations for commuting efficiency optimization and hybrid travel mode.
For example, based on real-time calculation and prediction of real-time road conditions and bus information for each time, it can help the commuter to tailor the route to and from get off work every time; for non-self-driving hybrid commuting mode, Google Maps will comprehensively calculate the time of walking, riding time and bus arrival time, and give commuting suggestions for users to make decisions.
In addition to performing intelligent recommendation routes, Google Maps adds more merchants and new addresses to the map through the combination of AI and satellite images. It can also display the opening and closing time of the target store selected by the user and the average length of time the customer stays in the store to remind the user to leave at the appropriate time to avoid not finding a parking space or encountering congestion on nearby roads.
This is closely related to the multidimensional data provided by users and merchants that Google Maps has collected for a long time. Although these new AI functions are considered by many industry insiders to follow the footsteps of similar APPs in China and lack novelty. But the details of AI algorithm recommendations are very remarkable.
For example, in 2018, Google Maps has launched real-time bus delay forecasts based on machine learning in hundreds of cities around the world. In addition to relying on real-time data on travel time and bus location provided by bus agencies as the basis for judgment, the model created by Google Maps will also combine factors such as speed, time, stops, and road congestion to make real-time predictions. Combining travel data and machine learning models, they can also predict the congestion of buses or trains in 200 cities around the world. The detail of
is very practical. When you're still a few hundred meters away from the bus stop, maybe Google Maps will tell you if it's worth walking quickly to catch a shuttle bus that will take a few minutes to get in.When you learn that there is an empty car with seats immediately behind, will you give up the crowded bus in front of you?
In addition, the system is based on machine learning. By comparing a large amount of data for users' personal preferences, it can customize personalized information for different users, such as telling you what new shops are opened near your residence, recommending nearby food to you, etc. Users can also quickly share the food on the map with friends and mark it in real time on the map. Above the
data, the "unstunning" AI upgrade path can be seen above
. The massive data of such as image data, user behavior data, etc., have become a powerful moat for Google Maps to remain strong. Backed by massive data, Google Maps’ AI upgrades show some more obvious characteristics.
is first of all a heavy dependence on Google Image technology. From the above AI evolution path of Google Maps, it can be seen that Google is the first to apply machine learning and other technologies to process huge map image data, including the identification of important information such as road traffic signs, street names, and shop names. On the one hand, it is derived from Google's own leading advantage in image recognition technology, and on the other hand, it is also Google's priority for precise travel map data.
is followed by Google Maps’ in-depth mining of user travel data and local service data. Using AI algorithms, a large amount of data hidden under user local searches can generate better travel service recommendations and more commercial value. Google found that in addition to travel navigation, exploring local services has also become a heavy tool for users to use. At present, Google Maps’ “Search” tool can meet users’ booking hotels, ordering cars and meals, querying travel routes, and even meet dozens of different types of services like Meituan in China.
For example, this time, the new version of Google Maps is upgraded, mainly to update the five tabs on the homepage: Explore, Commute, Save location, Contribution and Updates (Explore, Commute, Saved, Contribute and Updates).
From these newly added functions, it can be seen that Google Maps will pay more attention to users' localization services and users' active contribution to geographic data. The recommendation of complete travel data and AI algorithms allows Google Maps to respond to the needs of more and more diverse users in real time and provide more travel services.
In addition, we can also see Google Maps’ ambitions to integrate global city management. Although we rarely hear Google participate in the so-called "smart city" program. But Google is already planning this huge project in depth.
For example, Google Maps has begun to use machine learning algorithms and satellite images to map infrastructure such as complex buildings around the world. According to Google, 110 million buildings were added in the first half of 2018, using algorithms alone. It can be foreseen that in the future, what Google will give will no longer be a two-dimensional geographic schema, but more likely to be a three-dimensional 3D map world.
Through the above observations, Google's layout in AI technology pays more attention to the mining and processing of underlying big data, and pays more attention to the improvement of service details experience. This has made Google Maps almost "lost" in technology news in recent years.
, as a product that has provided Google AI for many years, coincides with the 15th anniversary of such a node, but there are no key updates in voice interaction, and no related news about Internet of Vehicles and autonomous driving. Such detailed technical improvements will inevitably make people feel a little overwhelmed. It seems that Google Maps' technology upgrade has also reached congested sections, and the speed is a bit unsatisfactory.
AI is not full: Google Maps' "change and unchange"
Google Maps has another obvious change in this version upgrade, which is the update of the icon. The new icon removes the classic map style and changes it to Google-toned map pins. Google's internal explanation of this change is not only to help users realize travel navigation, but also to help users discover where they want to go.
In fact, although from a business perspective, the 15th anniversary technology update may not be sexy enough. But from another perspective, Google Maps is still doing its old business: helping users achieve a better travel experience and provide better public services.
For example, Google Maps is predicting bus traffic delays in real time through machine learning. By predicting the delay of buses in advance, even if it is just a few minutes of error, it can be of great help to users. For example, enhancing data connection to public transportation allows users to submit more detailed information about public transportation, such as carriage temperature, wheelchair accessibility, or whether there are women's special carriages, etc., in order to make more humane and thoughtful travel recommendations for users.
and some more complex public services. Combined with prediction models such as disaster weather and epidemics, predict and promptly notify local users of some unexpected situations to avoid accidental disasters such as hurricanes, tsunamis, earthquakes, and infectious diseases.
Finally, it is worth mentioning that an artist user expressed his 15th anniversary of Google Maps in a "hacking" way.
This German artist named Simon Wechkert dragged a small cart on the road with 99 mobile phones driving in navigation mode. As a result, he successfully mistakenly thought that nearly 100 cars were driving on the road, and then caused severe traffic jams on the entire road. To this end, he also chose to conduct experiments before the Google Berlin headquarters building before the anniversary celebration. Wechkert's obsession is to verify that Google Maps' judgment on congestion is still very simple.
Google responded to this: This creative behavior is highly encouraged, and we are willing to accept user opinions and make Google Maps better.
I don’t know when this small problem will be solved on the congestion road of technology upgrades.
Editor: Tao Zongyao
