This article is from WeChat official account:Smart things (ID: zhidxcom), author: Xinyuan, from the title figure: vision China
Ten years ago, in February 2010, Siri’s first virtual personal assistant application Siri appeared in the Apple Store.
Two weeks later, Siri co-founder and CEO Dag Kittlaus (Dag Kittlaus) received a call: “Hi , This is Steve Jobs, can you come to my house tomorrow?”
▲Apple founder and former CEO Steve Jobs
Steve Jobs! The man who made the iPod and iPhone legends! Kitlaus immediately agreed to the invitation. The next day, they had a three-hour conversation in front of the fireplace at Jobs’s house. Jobs made it clear that he wanted to acquire Siri.
The founding members of Siri were very flattered, and then refused without hesitation. Because they have just raised a new round of funds, they have never considered being acquired.
The rejected Jobs was not discouraged and continued to paint a grand vision to the founder of Siri-Siri can become the core content of Apple’s strategy across multiple devices, and its impact on the world is far greater than that of an independent company. Jobs’ persistence caused the Siri team to change its mind. In April of the same year, Siri sold Apple for $200 million.
This is considered to be the last big thing Steve Jobs did during his tenure as Apple’s CEO. Siri, as Apple’s voice assistant, passed away on the second day of its debut at the 2011 Apple conference.
This event is also regarded as an important milestone in Apple’s acquisition of AI companies. Since the acquisition of Siri, Apple has successively acquired no less than 28 AI companies. It is the American technology giant that has acquired the most AI companies to date.
Even if the global economy is not prosperous due to the impact of the epidemic in 2020, Apple still has not stopped buying and buying, 1-5During the month, four successive acquisitions of AI companies were completed.
Which AI companies does Apple prefer to acquire? How do these companies integrate into Apple’s product system after being acquired by Apple? How did these acquisitions affect Apple’s AI layout?
This article will take a deep look at Apple’s acquisitions in the AI field over the past ten years, and see how Apple used capital to build an AI empire, and where it might go next.
▲The AI company acquired by Apple from 2010 to July 2020 (Smart Things Statistics)
Apple’s AI acquisition: a firm move, most of the transaction amount is undisclosed
Compared with other American technology giants, Apple is not the favorite company that relies on “buy, buy, buy” plump wings.
So far, Apple has initiated only over a hundred acquisitions, which is far behind Google’s over 250 acquisitions.
But when it comes to the number of AI companies acquired, Apple has topped the list in recent years. According to data from the research organization CB Insights, as of 2019, US technology giants such as Apple, Google, Microsoft, Facebook, Intel, and Amazon are the most active acquirers of AI startups.
▲Changes in the number of AI companies acquired by Apple from 2010 to July 2020
From 2010 to present, Apple has made about 28 public acquisitions in the AI field, covering different AI tracks such as machine learning, machine vision, speech processing, data optimization, emotion recognition, and autonomous driving, and has contributed a lot to Apple. AI products and services.
From 2010 to 2015, Apple’s AI investment attitude was relatively cautious, with only 5 AI acquisitions in total. At that time, Google was the US technology giant that acquired the most AI companies. Since 2015, Apple has been catching up in the number of AI acquisitions and successfully surpassed Google to rank first.
From a geographical point of view, 14 of the AI companies acquired by Apple were American companies, and one was from Canada; there were 11 European AI companies acquired by Apple, of which 5 were founded in the UK; 2 were acquired by Apple Asian AI companies are from India and Israel.
▲The distribution of AI companies acquired by Apple from 2010 to July 2020
From the perspective of funding, Apple loves mysticism, and most of the investment amount is not disclosed.
Compared with the $3 billion acquisition of Beats in the hardware field and the $2 billion acquisition of Intel’s modem business, Apple’s publicly acquired acquisitions in the AI field are less, and the highest one is the acquisition of Shazam, which is $400 million.
In the acquisition case with public transaction value, the transaction value of the three US AI companies, Turi, Lattice Data, and Xnor.ai, was US$200 million, and the rest did not exceed US$100 million.
▲2010-July 2020 Apple’s acquisition of AI company transaction volume
These AI acquisitions of Apple have a common feature. Once it perceives an opportunity to strengthen or supplement its core products, Apple will not hesitate to extend an olive branch.
From the perspective of landing applications, Apple’s AI acquisitions are mainly concentrated in two major applications: voice assistant and machine vision.
Nine of these acquisitions were aimed at improving the IQ of Apple’s voice assistant Siri, and 11 were aimed at improving privacy security or user experience in the field of machine vision.
In addition, Apple has 8 acquisitions to optimize and improve its seven types of segmented applications.
9 acquisitions around upgrading the Siri experience
As a “carrying madman” in the era of mobile Internet, Apple acquired Siri in 2010, directly igniting voice assistants. Siri has always been one of the most well-known voice assistants in the world.
From the early days of being labeled as “artificial mental retardation”, to now as long as the Internet is able to communicate with Apple users, Siri has become smarter and smarter. In addition to ten years of hard research and development by Apple engineers, it is also inseparable from Apple. 9 important acquisitions.
The protagonist of the first important acquisition is naturally Siri itself.
In February 2010, the Siri application landed on Apple’s App Store and became popular. Jobs initiated an acquisition invitation and subsequently acquired Siri for US$200 million. The following year, Siri made its debut on the iPhone 4s, far ahead of other companies such as Google at the time.
▲Siri application released in 2010
Siri is Norwegian, meaning “the beautiful woman who will lead you to victory”, but it is said that Jobs did not like the name.
Siri has not only become the first AI function to appear on Apple devices, but it also complements the search engine. The user issues voice commands, and the smarter Siri can quickly execute it. Since then, Apple has successively acquired two British speech recognition companies Novauris Technologies and VocalIQ in 2013 and 2015, all in order to continue to optimize and improve Siri.
However, although Siri was launched early, its initial performance was not good enough. Since 2016, the usage rate of Siri has declined. Amazon’s voice assistant Alexa and Google’s voice assistant Google Assistant have come in behind with the hot sales of smart speaker products.
Affected by Apple’s closed culture, the use of Siri is limited to Apple devices, instead of extensive cooperation like Amazon and Google’s voice assistants, which limits the development space of Siri.
This development route is at odds with the idea of the Siri initial team. By 2016, all the Siri teams acquired at the time had basically left Apple. It was not until the 2016 WWDC conference that Apple launched the open platform SiriKit to developers for the first time.
Also in September of this year, Apple acquired Tupl, an Indian machine learning companyejump, this company is good at using machine learning to process a large amount of complex data flow information, which is used by Apple in Siri and CoreML functions to optimize data management.
In the second year, Apple acquired Init.ai, an American startup with only 6 employees. The special thing about this transaction is that it does not involve any intellectual property rights. Apple only digs all its AI team to engage in Siri research and development.
Then in February to March 2019, Apple acquired the voice application startup PullString and the machine learning startup Laserlike.
The two are both American companies. The former was founded by former Pixar executives and can help users develop customized conversational interactive functions; the latter is mainly for personalized interest search, which can help Siri display news, music, and music according to user preferences. Video etc.
In 2020, Apple bought two more AI teams, Voysis and Inductiv, for Siri.
Voysis, headquartered in Dublin, Ireland, previously focused on improving digital assistants in online shopping applications. Its technology can help Siri improve natural language understanding, and it can also provide Apple developers with the Voysis platform.
The Canadian machine learning startup Inductiv is good at using AI to automatically identify and correct data errors. Its engineering team has joined Apple and has participated in multiple projects including Siri, machine learning, and data science to provide cleaner data. set.
Machine vision acquisition focus: convenience and privacy first
In the same year that Siri was acquired,Apple acquired Polar Rose, a Swedish facial recognition developer with only 15 employees, for approximately US$29 million.
This company is small but sophisticated, and it develops a face recognition application that can instantly display the person’s social information when it recognizes a face. This technology can not only be used to mark photos, but also participate in the security function of future iOS devices-face recognition unlock (Face ID).
Security and privacy have become the top priorities of all Apple machine vision-related acquisitions.
In addition to Polar Rose, Apple has three other acquisitions focused on privacy security.
One of them, in September 2015, Apple acquired Perceptio, a company that uses AI to automatically classify photos. This company mainly develops technologies that can perform local calculations.
Three years later, Apple acquired Silk Labs, an American smart home startup with only 12 employees. This company is mainly doing deep learning based on the local device, and emphasizes the privacy of key data acquisition and data anonymity, which is in line with Apple’s emphasis on data protection.
Then in January of this year, Apple acquired Xnor.ai, a Seattle AI startup that was just established for three years. Its image recognition tool can provide edge computing capabilities and can be executed on low-power devices to meet Apple’s expectations. The need for data privacy protection. Xnor.ai has also developed a standalone AI chip that can run on a coin-sized battery using solar energy.
These acquisitions reflect one of Apple’s unique aspects in the AI field-de-clouding.
Many companies’ AI functions are cloud-based, but Apple has long expressed its tendency to be “cloudless”, that is, to focus on local computing on the device.To protect user data privacy.
Apple’s other 7 acquisitions in the field of machine vision are aimed at improving user experience.
In January 2016, Apple acquired Emotient, an American startup that focuses on emotion recognition. The company uses AI technology to analyze facial expressions, judge people’s emotions, and does not store any identity information about this person. Nowadays, there are functions of understanding facial emotions in Apple’s photos, Animoji and other applications.
In the second year, Apple successively acquired 4 machine vision-related companies, including RealFace, SensoMotoric Instruments, Regaind and Spektral.
RealFace is a face recognition company based in Tel Aviv, Israel, with about 10 members. The main technology is to unlock devices and applications by recognizing faces.
The German company SensoMotoric Instruments has little to do with privacy and security. It was originally an eye-tracking system and glasses products, but it was acquired by Apple and used in the augmented reality platform ARKit.
Regaind, a French AI startup, is mainly engaged in image classification. By using AI technology to analyze the aesthetics, sharpness, exposure, color and other attributes of the photo, it detects the face of the photo and infers the person’s gender, age, and emotions. Information, used by Apple to optimize the memory function of the photo application.
Spektral is a Danish machine learning startup that does real-time green screen technology. It is good at using spectrogram theory and deep learning technology to cleanly replace video backgrounds. This technology debuted in the video background removal tool of Apple’s iOS 13 beta. .
In 2019, Apple acquired Fashwell, a Swiss visual search startup, and Spectral Edge, a British image fusion company.
Fashwell was mainly engaged in shopping search before the acquisition. Its AI visual search engine can identify products in pictures and automatically tag them to improve the user’s shopping experience. It is currently unclear whether Apple uses its technology for photos or other apps.
Spectral Edge’s AI technology is good at enhancing image details in blurred or overexposed situations. It can enhance mobile phone photos in low-light environments, improve white balance accuracy and improve color accuracy, and it is likely to be used to improve the future Night mode and other photography functions of iPhone camera.
Data optimization, music analysis, autonomous driving, seven types of subdivision applications acquired by Apple AI
In addition to optimizing for virtual voice assistants and machine vision applications, Apple has eight AI acquisitions for applications in 7 different segments.
These acquisitions include not only open source machine learning frameworks and data structuring that reduce the difficulty of AI development, but also optimizations focused on applications such as podcasting, music, marketing, automobiles, and weather forecasting.
Open source makes machine learning easier
In August 2016, Apple acquired Seattle A for approximately US$200 million.I founded Turi. Turi was founded by a professor at the University of Washington and provides AI development tools and frameworks for developers.
After acquiring Turi, Apple established a machine learning department, and based on this, launched the open source Turi Create framework in 2017 to simplify the development of machine learning models and enable ordinary developers to play with AI.
Turn unstructured “dark data” into structured data
In May 2017, Apple purchased Lattice Data, a two-year-old startup, for approximately US$200 million.
Lattice Data owns the business core technology of the Stanford DeepDive system “Dark Data(Dark Data)” project, and is good at using machine learning technology to transform unstructured Converted dark data (dark data) into structured and useful information.
The technology is somewhat similar to Google’s knowledge graph, that is, the connection between people, places, and things can be mined through data.
Quickly retrieve audio by converting speech to text
In December 2017, Apple acquired Pop Up Archive, a start-up company in Oakland, USA.
This company has developed tools for transcribing, organizing, and searching audio files. Its podcast search engine can use machine learning to convert speech into text, and crowdsource the task of correcting transcription errors to help users recover from lengthy speech Find what you want quickly in. This is very beneficial to the experience optimization of the Apple Podcast.