Every Company, Big or Small, Whatever It Does, Is Talking About Artificial Intelligence — As If There Were No Other Choice | Why Is Silicon Valley in Such a Rush ❹
This is not the first artificial-intelligence boom of the past 50 years.
This is not the first artificial-intelligence boom of the past 50 years.
"I would have been glad to answer these questions. But I recently left my teaching post at Carnegie Mellon University to join DeepMind, so it isn't convenient for me to respond right now."
That was how Carnegie Mellon professor Chris Dyer turned down our interview request about artificial intelligence not long ago. His fields are machine-learning algorithms and natural language processing; his earlier papers were about getting computers to learn on their own, without human supervision or guidance.
Which is exactly what DeepMind does. The Go algorithm built by this Google subsidiary improved its play by playing against itself, and in the end beat world champion Lee Sedol.
Dyer is only one example of tech companies poaching from Carnegie Mellon. Last year, 40 AI researchers from the university's driverless-car lab — starting with the lab's director — were hired away by its partner Uber. The whole lab had only just over 100 people to begin with.
Companies and university labs used to be collaborators more than rivals: the labs did long-term basic research, and the companies turned the results into commercial products. But on artificial intelligence, tech companies can no longer wait.
Artificial intelligence is a boom that has spread across all of Silicon Valley
It isn't only the big companies, the ones able to poach an entire lab, that are betting on artificial intelligence.
According to data from the research firm CB Insights, from the second half of 2015 on, the number of venture-capital investments in startups kept falling, down to its lowest level since 2011 — but investment in artificial intelligence only grew.
In 2015 there were 397 investments in AI companies, raising $2.38 billion in all — nine times the 2011 figure.
A good many Silicon Valley founders are doing the same one thing: training artificial intelligence on large amounts of data, letting the computer learn from it and improve its own algorithms, then using the better algorithms to run one particular business — playing chess, diagnosing disease, retouching photos, or speeding up the checkout line.
"Ten items used to take a cashier a minute to scan. Now you just put your shopping tray on the machine, and the camera uses image recognition to tell you the total in under a second. Then you swipe your card and you're done." Mukul Dhankhar explained this while demonstrating, on the spot, the AI product made by his company Mashgin — an automatic checkout machine.
QDaily saw a prototype of the device at Mashgin's office in Palo Alto. We acted out a convenience-store purchase: picked up a can of soda and a bar of chocolate, set them on the counter, lifted our hand out of the basket. In under a second the machine had the total. The customer swipes a card, pays and walks out.
At an ordinary supermarket, a cashier would have to scan each barcode one at a time.
Mukul said the prototype has six cameras pointed in different directions to photograph the goods. It doesn't read barcodes; it reads shape and color.
Legal Robot, a startup founded a year ago, is also using artificial intelligence to solve a problem. It wants to translate obscure legal documents into plain language, so that ordinary people can see what traps a contract contains. That takes natural language processing.
Founder Dan Rubins wants to lower the barrier to getting legal help. Signing an employment contract, for instance, is a serious matter, but paying a lawyer to walk you through it clause by clause costs several hundred to a thousand dollars at least.
Rubins thinks his company may find a business faster than consumer AI products like Siri do. "In ordinary conversation there's far too much variability in what words mean, and the rigor and obscurity of legal writing exists precisely to keep meanings from multiplying. For natural language processing that's actually better, because there are fewer variables."
So far Legal Robot has obtained 100,000 sample legal contracts from its partners Stanford University and General Motors to train its algorithm.
Another hot company trying to replace basic human labor with artificial intelligence is X.AI. It has built a bot named Amy that reads and answers email, picking out meeting and appointment requests in English and settling a time and place — and the emails she sends back give no sign at all that a machine wrote them.
X.AI wants people who can't afford a personal secretary to have an artificial-intelligence one.
The finished product hasn't launched yet, but simply getting a slot in Amy's current closed beta means waiting several months.
The three companies work in different fields, but what they do adds up to much the same thing: use artificial intelligence to cut human labor and raise efficiency.
Two or three years ago, artificial intelligence was not something most companies took seriously
"When I started the company in 2013, there were hardly any AI companies. It's different now — everybody is doing artificial intelligence," Mashgin founder and CTO Mukul Dhankhar told QDaily.
In 2013 what startups did most was mine the potential of the smartphone.
That year smartphone shipments had just passed 1 billion units for the first time, up 39% from the same period in 2012. One in every seven people on Earth owned a smartphone.
Delivering services over the mobile internet on those phones was at its peak of popularity at the same moment: Uber and Airbnb, today the two most valuable startups in Silicon Valley, both began attracting investors' money in 2013.
But Google and Apple, the companies driving the smartphone forward, could see that the market was moving closer to saturation, and started looking for the next hardware platform.
By 2013 Google Glass was already on sale to developers. Google was also in talks to buy the smart-home company Nest; the deal was announced the following year.
Also in 2013, Apple CEO Tim Cook hinted in an interview that the company would release a smartwatch.
After failing to find a new hardware platform, every tech giant started talking about artificial intelligence
But over the next few years, none of these new hardware platforms took hold.
Google Glass failed outright, and Google executives stopped wearing the glasses in public. Nest's head left Google too.
And the Apple Watch, taken to be the next generation of computing device, saw sales fall year on year by its fifth quarter on the market. Never mind becoming the next iPhone — its growth didn't last as long as the iPad's.
Having failed to build the next smartphone-scale hardware platform, tech giants badly in need of their next source of growth all began talking about artificial intelligence.
Google has gone furthest in this wave: driverless cars, AlphaGo the world-champion-beating Go algorithm, a smarter search engine and a smarter email app. The company has thrown itself headlong into artificial intelligence.
At its annual developer conference this May, Google put its future strategy squarely on artificial intelligence. "Machine learning" was the phrase it used most, and three of the five new products announced at the conference were directly tied to AI.
"Right now every department in the company is trying to fold machine learning into its own products, whatever those products are," Google Translate engineer Otavio Good told QDaily. His own team recently used machine learning to let the translation app read the text in a photo.
Apple, a hardware company at heart, is talking about artificial intelligence as well.
"We will apply artificial intelligence in a focused way, to improving the user experience." On Apple's earnings call last week, CEO Tim Cook mentioned artificial intelligence and machine learning seven times.
Turning to Siri, Apple's AI assistant, Cook said, "Apple will put machine learning to work in other places too — Photos, Messages, Maps and the music service Apple Music."
It already has. At its June event this year, Apple showed that it could use artificial intelligence to process photos and suggest replies automatically in Messages.
Microsoft, which has put money into basic research for the long haul, has gathered artificial intelligence under one experience: conversation. CEO Satya Nadella announced a "conversation as a platform" strategy, embedding the image-, voice- and video-recognition work Microsoft has researched for years into conversational products like the chatbot Xiaoice.
The e-commerce giant Amazon is doing artificial intelligence too, by way of a line of speakers that understand spoken language. At a conference this year, Amazon CEO Jeff Bezos said more than 1,000 people at the company were already working on the Echo speaker and the Alexa voice assistant.
Facebook, the youngest of these five companies, is copying WeChat's early playbook and building official accounts with AI features. It is still crude, but Facebook has done more serious work on the underlying artificial intelligence than WeChat has.
Competing with each other, the big companies have built new platforms for AI startups
The "Google Brain" project, begun in 2012, is taken as the technical starting point of the past few years' AI boom.
A simulated neural network running on 1,000 servers could, within a week, accurately pick out "cats" in video, along with 3,000 other objects that appeared in the footage. Accuracy was only 60%, but at the time that was a major technical breakthrough.
Last November, members of the Google Brain team released this machine-learning project as free open source. That is TensorFlow, whose natural-language modules and AI functions let developers train their own AI products quickly. TensorFlow has since become one of the most popular open-source projects on the code-hosting platform GitHub.
For an individual developer, making a speaker or a television understand everyday speech, or making a homemade robot tell one image from another, is no longer hard to imagine.
The barrier to building AI products has come down a great deal — at hackathons for American college students, every project now uses artificial intelligence to some degree.
At the end of 2015, hacker George Hotz converted a car into a self-driving one on his own, on top of open-source AI projects. This used to be something only the best university labs, or teams like Tesla's and Google's driverless-car groups, could produce, and it took them ten years.
Google isn't the only one. Since the second half of last year, the major tech companies have opened up their AI projects one after another, to get more developers to help improve their algorithms:
- Amazon open-sourced Alexa in June 2015, the intelligent voice-control system that lives inside Amazon's Echo speaker and Tab;
- Facebook open-sourced the server hardware design behind its intelligent assistant Facebook M in December 2015;
- Microsoft open-sourced the CNTK framework in January 2016, the AI speech-recognition tool behind Microsoft's personal assistant Cortana and Skype Translator.
While the big companies make an AI startup easier to begin, they are also competing with the founders of those startups.
Artificial intelligence is a contest of data volume and algorithms. Big companies have piled up enormous amounts of data already, and they have the money to hire good engineers to refine the algorithms.
"I think it's going to be very hard for startups," Mukul, CTO of the AI checkout-machine company Mashgin, told us.
Selling out has become the main exit for an AI startup. In the past five years Google has bought nine AI startups. Apple, Intel and Salesforce have joined the buying spree, and so has Twitter, whose own business isn't doing well; together they have acquired more than 30 AI startups.
Artificial intelligence: Silicon Valley has changed direction to look for the future
Over the past 40 years, technology companies have moved ahead on one wave of new technology after another: personal computers in the 1980s, the internet in the 1990s, and after 2007 the spread of smartphones and the rise of the mobile internet.
With each change in the underlying infrastructure, founders used the new technology to make traditional businesses more efficient.
Now artificial intelligence is speeding that up as well, turning work that once needed human hands into work that needs fewer people, or none at all.
Every field we can think of seems to be getting an AI boost — e-commerce's "recommended for you," the carpool routes in a ride-hailing app, the AI assistants.
But although today's artificial intelligence has beaten Go masters, researchers still don't understand how the human brain works. AI still gets better by training on data.
For artificial intelligence to be the technology that carries the industry forward after the smartphone, it has to become good enough, and useful enough, within a few years.
And expectations of artificial intelligence have often run too high.
The earliest AI research began in the 1940s and 1950s. Right after the computer was invented, scientists at MIT and the Stanford Research Institute were already trying to give computers human intelligence — teaching them, for instance, to recognize human language.
Computer scientists then optimistically believed that ten years would be enough to build an artificial-intelligence computer that served humanity. We now know how unrealistic that was.
Artificial intelligence has run hot several times in the past 40 years. Neural-network computing in the 1980s never became the opening through which AI entered the commercial world, and the medical and banking projects taken on by IBM Watson, which beat humans a few years ago, have all stalled.
Jerry Kaplan, who founded two artificial-intelligence companies in the 1980s (including the one that later turned into the computer-security company Symantec), spoke recently about the AI boom and told the New York Times: "Sitting around with Silicon Valley's believers in artificial intelligence sometimes makes me feel like an atheist at a revival meeting."
Editor's note: the PC, the internet, the smartphone — the spread of every foundational technology has come with years of fast growth. For the technology industry, that growth has now run for forty years.
It may be about to pause: the growth the smartphone brought has already slowed, and since ride-hailing and food delivery we have not seen another breakthrough on the order of social networking or e-commerce, the kind that sets a hundred billion in wealth in motion.
What new technology will pave the way for the whole industry's next stretch of growth? And will it arrive in time?
These questions will affect everyone, from giants like Apple and Google, to venture-capital firms and startups, to every user of a technology product — which is to say, more or less everyone.
Over the past forty years, most of the technologies that laid the foundation for the industry's lasting prosperity came out of Silicon Valley. We are running a series of reports from Silicon Valley, looking at this shift from different angles:
In Silicon Valley, we found out what actually became of the WeChat official account, four years on
Translated from the original QDaily article. Original publication date retained.