Over the past couple of months, Google has been playing its own peculiar game of Moneyball. It may not make a ton of sense right now, but Google is setting itself up to leave its competitors in the lurch as it moves into the next generation of computing.
Google has been snatching up basically every intelligent system and robotics company it can find. It bought smart thermostat maker Nest earlier this month for $3.2 billion. In the robotics field, Google bought Boston Dynamics, Bot & Dolly, Holomni, Meka Robotics, Redwood Robotics and Schaft.inc to round out a robot portfolio group that is being led by Android founder Andy Rubin.
When it comes to automation and intelligent systems, Google started its acquisition spree in late 2012 with facial recognition company Viewdle and has continued picking up neural networks since then, including the University of Toronto’s DNNResearch Inc. in March 2013, language processing company Wavii in April and gesture recognition company Flutter in October. Google bought a computer vision company called Industrial Perception in December and continued its spree with a $400 million acquisition of artificial intelligence gurus DeepMind Technolgies earlier this week.
Dizzy yet? The rest of the technology industry surely is. It’s hard to compete with Google’s acquisition rampage when it seemed like Google had very little rhyme or reason to the purchases it was making as they were happening. Google beat out Facebook for DeepMind while Apple had interest in Nest. But, after more than a dozen large purchases, Google’s strategy is finally becoming clear.
Google is exploiting an inefficiency in the market to become the leader in the next generation of intelligent computing.
Google’s Moneyball Strategy
Moneyball is a term coined by author Michael Lewis in his 2003 book, Moneyball: The Art Of Winning An Unfair Game. The term is often falsely associated with the use of advanced statistical models used by Billy Beane, the general manager of the Oakland Athletics, to build a club that could thrive in major league baseball. But the principle of Moneyball is not actually about using stats and data to get ahead, it is about exploiting systems for maximum gain by acquiring talents that are undervalued by the rest of the industry.
This is exactly what Google is doing: exploiting market inefficiency to land undervalued talent. Google determined that intelligent systems and automation will eventually be served by robotics and has gone out of its way to acquire all of the pieces that will serve that transformation before any of its competitors could even identify it as a trend. By scooping up the cream of the crop in the emerging realm of robotics and intelligent systems, Google is cornering the market on talented engineers ready to create the next generation of human-computer interaction.
Technology Review points out that Google’s research director Peter Norvig said the company employs “less than 50 percent but certainly more than 5 percent” of the world’s leading experts on machine learning. That was before Google bought DeepMind Technologies.
To put Google’s talent hoarding into context, remember that many companies are struggling just to find enough talent to write mobile apps for Android and iOS. When it comes to talented researchers focused on robotics and AI components like neural networks, computer vision and speech recognition, the talent pool is much smaller, more exclusive and far more elite. Google has targeted this group with a furious barrage of aggressive purchases, leaving the rest of the industry to wonder where the available talent will be when other companies make their own plays for building next-generation intelligent systems.
“Think Manhattan project of AI,” one DeepMind investor told technology publication Re/code this week. “If anyone builds something remotely resembling artificial general intelligence, this will be the team.”
Of course, Google is not the only company working on intelligent systems. Microsoft Research has long been involved in the creation of neural networks and computer vision, while Amazon has automated many of its factories and built cloud systems that act as the brains of the Internet. The Department of Defense research arm, Defense Advanced Research Projects Agency (DARPA), has long worked on aspects of artificial intelligence, and a variety of smaller startups are also working on their own smaller-scale intelligent platforms. Qualcomm, IBM and Intel are also working on entire systems—from chipsets to neural mesh networks—that could advance the field of efficient, independent AI.
What Is Google Trying To Accomplish?
To understand what Google’s next “phase” will look like, it is important to understand the core concepts that comprise Google.
Google’s core objective—which has never really changed despite branching out into other areas of computing—is to accumulate and make accessible all of the knowledge in the world. Google makes money by charging advertisers access to this knowledge base through keywords in its search engine. If there is a brilliance to Google’s business model, it is that the company has basically monetized the alphabet.
The work is nowhere near done, but Google has already done an impressive job over the last 16 years making the world’s knowledge available to anyone with Internet access. Thanks to Google, the answer to just about any question you could think of asking is at your fingertips, and with smartphones and ubiquitous mobile computing, that information is now available wherever you go.
The next step for Google is to deliver that information to users with automated, intellectual context. The nascent Google Now personal assistant product that Google has been driving is the first step in this, but it has a lot of room to grow.
If we take the concept of Google monetizing the alphabet and apply it to everyday objects, we can see where artificial intelligence come into play for how Google plans on changing the fundamental nature of computing.
What if you could use a device—like a smartphone, Google Glass or a smartwatch—to automatically identify all relevant information in your area and deliver it to you contextually before you even realize you want it?
If we mix the notion of ubiquitous sensor data laden within the Internet of Things with neural networks capable of “computer vision” and comprehending patterns on their own, then we have all the components of a personalized AI system that can be optimized to every individual on the planet.
Academic researchers call these computing concepts “deep learning.” Deep learning is the idea that machines could eventually learn by themselves to perform functions like speech recognition in real-time. Google’s prospect is to apply deep learning to the everyday machines like smartphones, tablets and wearable computers.
But what about all the robots Google just purchased? This is a little trickier to discern, but if Google does eventually figure out the intricacies of artificial intelligence, it can then apply these principles to an army of automated machines that function without human interference. For instance, with computer vision, machine learning and neural networks, Google could deploy robots to its maintain its data centers. If a server breaks or is having problems, a robot could pay it a visit, tap into its internal diagnostics or see that it is having a hardware issue. Google’s driverless car could benefit from all of these technologies as well, including speech and pattern recognition.
Google’s research into robotics and deep learning doesn’t have to mean this new technology will be restricted to beefing up its current products. Advances in cognitive computing can be applied to many, many different types of industries, from manufacturing to analyzing financial data at large banks. Like the lessons learned with smartphones, the applications of machine learning technology can be applied almost anywhere—as long as patents don’t apply.
Google Has All The Ingredients To Make AI Work
Google has brains. Lots of different kinds of brains.
From a people perspective, Google’s head of engineering, Ray Kurzweil, is the world’s foremost expert on artificial intelligence. Andy Rubin joined Google in 2005 when the company purchased his Android platform to create a smartphone operating system. But currently, Rubin is taking his job of building Android much more literally as he now heads up Google’s fledgling robotics department. Jeff Dean is part of a senior fellow within Google’s research division working in the company’s “Knowledge” (search-focused) group, which will most likely be the team to incorporate DeepMind.
Those names are just a few examples of Google’s best brains at work. Google also has plenty of machine/computer brains that perform the bulk of the heavy lifting at the company.
The search product has been tweaked and torqued over the years to be one of the smartest tools ever created. In its own way, Google search has components of what researchers call “weak” artificial intelligence. Google’s server infrastructure that helps run search (as well as its Drive personal cloud, Gmail, Maps and other Google apps) is one of the biggest in the world, behind Amazon but on par with companies like Microsoft and Apple.
Google has the devices necessary to put all that it creates into the hands of people around the world. Through the massively popular Android mobile operating system, it fledgling computer operating system in Chrome OS and accessory devices like Google Glass or its long-rumored smartwatch, Google can push cognitive, contextual computing to the world.
All Google needs to do is make artificial intelligence a reality and plug its components into its large, seething network and see what happens. That is both very exciting and mildly terrifying at the same time.
The Risks For Google, The Internet & The World
“Behold, the fool saith, ‘Put not all thine eggs in the one basket.’ Which is but a matter of saying, “Scatter your money and your attention.” But the wise man saith, ‘Pull all your eggs in the one basket and … WATCH THAT BASKET.'” ~ Mark Twain, Puddn’head Wilson
In 1991, DARPA pulled much of its funding and support for research into neural networks. What followed was a period researchers called an “AI Winter,” where the field of artificial intelligence research became stagnant and did not progress forward in any meaningful way. The AI Winter from the 1990s was not the first and might not be the last.
Google is accumulating many individual brains in the field of AI into one big basket. If Google fails (or loses interest) to create the next generation of artificial intelligence, another AI Winter could definitely be a possibility.
Google also betting a lot of money on the fact that it can take the components of artificial intelligence and robotics and apply it to everything it touches. Between DeepMind and Nest, Google has spent $3.6 billion in the automation industry this year and those were just two companies. Google has lots of eggs in this basket, and if it fails, it could cost Google years, employees and the bleeding edge of next-generation computing.
Academics and pundits are also worried about the implications of privacy with Google’s chase of the contextual Internet centered around the individual. With Nest, Google could know just about everything that you do in your home (when you leave, when you get home, how many people are in the house and so forth). Part of aggregating the world’s knowledge is parsing information about the individual’s that inhabit that world. The more Google knows about us, the better it thinks it can enhance our lives. But the privacy implications are immense and well-justified.
There is also a large ethics question surrounding the use of artificial intelligence. Some of it centers around science fiction-like scenarios of enabling robots to take over the world (like Skynet in Terminator or the robots in The Matrix). But, in addition to privacy concerns, there are many different ways AI could be abused to severely augment the service economy. As part of the DeepMind acquisition, Google agreed to create an ethics board for how it will use the technology in any future applications.