Executives who are afraid of long-term commitments should avoid artificial intelligence.
Like with romantic relationships, using the technology requires an appetite for hard work, planning, and patience. Even then, failure is a real possibility.
“If you are not going to make this investment for the long term, it is not a good idea to do machine learning,” LinkedIn’s vice president of artificial intelligence Deepak Agarwal told Fortune in a recent interview.
Agarwal leads the social network’s many machine-learning projects that power tasks like recommending job openings to users or determining which posts are the most relevant to them and that they are most likely to click on. He’s been involved with machine learning and statistics for years, after previously working at Yahoo and AT&T in senior technology research roles.
One thing Agarwal has learned is that it can take at least a year to see a financial return from machine learning. During that time, companies must navigate the mundane work of cleaning and properly labeling data, and figure out the correct machine-learning algorithms and data infrastructure to use.
Agarwal recommends that small companies, which often have little data, make an effort to collect it before implementing machine learning. Without a large archive of information, the technology is nearly useless.
For some companies, machine learning makes no sense because it requires a lot of computing power to crunch the data. And renting cloud computing infrastructure can cost more than any additional profit that machine learning can create, Agarwal explained.
But when the technology works, the payoff can be huge. For large companies like Google that have embedded machine learning throughout their businesses and apps, a modest 2% improvement on a particularly important metric could result in financial gains in the hundreds of millions of dollars.
Jonathan Vanian
@JonathanVanian
jonathan.vanian@fortune.com
EYE ON A.I. NEWS
Here come the autonomous delivery vans. Chinese startup Neolix told Bloomberg News that it plans to deliver a thousand self-driving vans to customers like JD.com and Huawei. The autonomous vehicles, which cost $30,000, are intended to be used for deliveries and reducing the need for human drivers.
Who’s really funding Chinese A.I. startups? Buzzfeed News has linked U.S.-based organizations like the Alaska Retirement Management Board, Rockefeller Foundation, and Princeton University to private equity and venture capital firms funding some of China’s high-profile facial recognition startups SenseTime and Megvii. The ties between the U.S. firms and Chinese startups comes amid a trade war between the two countries as well as China’s controversial use of facial recognition technology to detain ethnic minorities, the report noted.
A.I. needs global counseling. The World Economic Forum has created The Global Artificial Intelligence Council, which is intended to develop artificial intelligence policy guidelines for countries and corporations. Microsoft president Brad Smith and Sinovation Ventures CEO Kai-Fu Lee are the A.I. council’s co-chairs.
Russia’s big A.I. push. Foreign investors plowed $2 billion into the Russian Direct Investment Fund, Russia’s sovereign wealth fund created in 2011, to help grow Russian companies specializing in A.I., according to The Moscow Times, which cited a report by the Vedomosti business daily. Spokespeople from the sovereign wealth firm told Vedomosti that the foreign investors included "partners from large sovereign funds and global corporations from the Middle East and Asia."
CAN RESEARCH BE PRACTICAL?
Sheldon Fernandez, the CEO of Canadian startup DarwinAI, talks to tech publication ITPro Today about the challenges businesses face with machine learning: “While A.I. has accomplished some dazzling things in an academic or research capacity, businesses are still figuring out how to translate its predictive capabilities into more mundane but practical use cases for the enterprise.”
EYE ON A.I. HIRES
Walmart hired Suresh Kumar to become its chief technology and chief development officer. Kumar was previously Google’s vice president and general manager of display and video ads, a corporate vice president for Microsoft’s cloud infrastructure and operations, and a vice president of worldwide retail systems for Amazon.
David Cameron, the former prime minister of the United Kingdom, will lead the advisory board of Afiniti, a startup specializing in applying machine learning to customer relationship management software.
Cybersecurity company Cyren hired Dr. Richard Ford to be chief technology officer. Ford was previously chief scientist of cybersecurity firm Forcepoint.
EYE ON A.I. RESEARCH
A.I. goes phishing. Researchers from Texas Tech University published a paper about using reinforcement learning—in which computers learn through repetition—to automatically detect malicious websites used for so-called phishing attacks. The researchers acknowledge the while the work is “not optimized for real world implementation,” some of the techniques could be used to detect spam and improve security for wearable devices.
A.I.-powered receipt readers. Researchers from Tokyo’s Center of Open Data in Humanities published a paper about using deep learning to scan and understand information in receipts. The researchers, some of whom are based in Vietnam, claim that their A.I. system can recognize and understand hand-scribbled numbers in receipts as well as handwritten Vietnamese.
FORTUNE ON A.I.
Trump's Smart Move on Artificial Intelligence – By Adam Lashinksy
Keep Your A.I. Buzzwords Straight – By Jonathan Vanian
Chip Wars 2019: What Nvidia, AMD, Intel, and Dell Announced at Computex – By Aaron Pressman
BRAIN FOOD
A.I. Teamwork. The New York Times explores the burgeoning use of video games as a training ground for A.I. by some leading research labs from Google’s DeepMind unit and the Reid Hoffman-backed OpenAI. Some of DeepMind’s latest work involves creating A.I. software that learns to master a virtual replica of Capture the Flag, the schoolyard game that requires teamwork and collaboration. The report said “Through thousands of hours of game play, the agents learned very particular skills, like racing toward the opponent’s home base when a teammate was on the verge of capturing a flag.” Despite the feat, however, the report cites an A.I. professor who questioned if the A.I. software “agents” were collaborating with each other, and instead “merely responding to what is happening in the game, rather than trading messages with one another, as human players do.”
