This is the web version of Eye on A.I., Fortune’s weekly newsletter covering artificial intelligence and business. To get it delivered weekly to your in-box, sign up here.
Companies that are struggling to make their artificial intelligence useful should follow this important piece of advice: Get key departments involved, not just data science staff.
Take Nasdaq, which recently debuted a new system that uses A.I. to monitor for a kind of stock fraud known as spoofing. With spoofing, criminals covertly manipulate a company’s share price by submitting buy or sell orders and then quickly cancelling them before they are executed.
It took nearly a year for Nasdaq to build and test the monitoring system, which is based on the powerful deep-learning technique that’s created breakthroughs in computer vision and text analysis. Earlier this month, the stock exchange debuted the system more broadly.
Despite the fancy technology, one key element in making it useful was basic: Making sure non-techies have a voice in how the the technology is built.
Tony Sio, the Nasdaq executive who oversees the exchange’s regulatory technology, explained how workers had to first identify what spoofing looks like, using the data Nasdaq collects. In that way, the computer would know what to look for.
The process required making sure junior staff working on the technology and senior staff focused on stock fraud agreed on what spoofing is and what it looks like.
“It’s really important to make sure people are consistent in terms with how they interpret [data] and how they, I suppose, feed back into the machine-learning algorithm,” Sio said.
The next step will be for Nasdaq to prepare the new monitoring system to be used by other marketplaces. Dozens of outside exchanges and regulators use Nasdaq’s tools for policing trades, Sio said.
The process will entail using something called transfer learning, a technique that lets A.I. systems learn more quickly from new data without having to be trained from scratch. Nasdaq is hoping to sell variants of its new A.I. system to other markets, underscoring how the exchange’s sales staff, not merely the data crunchers, are also involved in A.I. projects.
“If we can use these transfer techniques to quickly transfer these models to those exchanges, I think we can get some real power from that,” he said.
Jonathan Vanian
@JonathanVanian
jonathan.vanian@fortune.com
A.I. IN THE NEWS
Microsoft trumps Amazon in JEDI cloud contract. The U.S. Department of Defense awarded Microsoft the Joint Enterprise Defense Infrastructure, or JEDI, cloud computing contract that is worth up to $10 billion over 10 years. Amazon and Microsoft were vying for the contract, which will help provide the Defense Department with cloud and A.I. services that are likely to be used in warfare.
Hey, big spender. The Stanford Institute for Human-Centered Artificial Intelligence released guidelines for a national A.I. policy that called for the U.S. to invest at least $120 billion over the next 10 years on education, research, and entrepreneurship initiatives intended to strengthen the U.S. in artificial intelligence. The hefty investment that the center of the proposal contrasts with the White House’s far smaller plan to spend $1 billion in A.I.-related research and development in 2020.
Recognizing faces from an MRI. Researchers from the Mayo Clinic released a study showing that Microsoft’s facial-recognition software identified patient faces in photos that were created using MRI imagery, The Wall Street Journal reported. The article stated that the study’s findings underscore “privacy threat that will increase with technology improvements and the growth of health-care data, experts in medical imaging and facial recognition said.”
McDonald’s has an appetite for machine learning. McDonald’s has spent “hundreds of millions of dollars” over the last seven months gobbling up companies specializing in machine learning and predictive analytics, The New York Times reported. According to the article, “In recent months, McDonald’s has tested voice recognition at some of its restaurants, seeking to replace the human workers who take orders with a faster system.”
GERMAN GOVERNING
A German government-backed task force released recommendations for rules governing A.I., and some pro-business lobbyists are concerned, Fortune’s David Meyer reported. Eline Chivot, a senior policy analyst at the Brussels office of the Center for Data Innovation, said, “The German Data Ethics Commission’s recommendations on A.I. send a worrying signal to businesses that they risk adopting A.I. at their own peril.”
EYE ON A.I. TALENT
Google hired Javier Soltero as vice president of its G Suite workplace software business. Soltero was previously a Microsoft vice president overseeing Cortana digital assistant and Office workplace software.
Data management startup DataStax picked Chet Kapoor as CEO, replacing Billy Bosworth. Kapoor was previously a vice president at Google and the CEO of enterprise startup Apigee.
EYE ON A.I. RESEARCH
Bias in healthcare. Researchers from Massachusetts General Hospital, the University of Chicago, and others published a paper in Science detailing how a popular algorithm used by hospitals to determine which patients need extra help from healthcare workers was biased against African-American patients. The paper explained that the algorithm “falsely concludes that Black patients are healthier than equally sick White patients,” because the algorithm uses health-care spending as a proxy for health. The paper underscores the concerns of some A.I. researchers studying bias who worry that cutting-edge machine learning systems could be prone to some of the same bias issues plaguing more rudimentary algorithms.
Smelly neural networks. Researchers from Google, Arizona State University, the University of Toronto, and others published a paper about using neural networks, software that learns, to predict odors based on the structure of certain molecules. Although the research shows that deep learning could be valuable for the science of smell, Wired noted, “It’s not clear if we can learn anything about human olfaction from a machine-learning model, since the design of the neural network isn’t the same as a human olfactory system.”
FORTUNE ON A.I.
A.I. Has a Bias Problem, and Only Humans Can Make It a Thing of the Past– By Gwen Moran
Google Says Its Latest Tech Tweak Provides Better Search Results. Here’s How– By Danielle Abril
Twitter Says A.I. Is Now Removing Over Half of Its Abusive Tweets Before They’re Flagged– By Alyssa Newcomb
What’s Next for Google After Claiming ‘Quantum Supremacy’?– By Robert Hackett
BRAIN FOOD
Have you scowled during a job interviews? The Washington Post profiled the company HireVue, which has gained attention for its software that analyzes people’s facial expressions and mannerisms in order to help companies automate aspects of the job-recruiting process. The article discusses some of the concerns lawmakers like Illinois State Rep. Jaime Andrade Jr. have with the software, who asked: “What are the data points being used? There has to be some explanation, and there has to be consent.” HireVue has previously argued that its software prevents the potential bias that can occur during job interviews, and claims that its software can actually be less biased than human interviewers. But that hasn’t satisfied critics, including a neuroscientist the Post talked to who discussed how a “scowl” can mean many things besides indicating a person is angry—in fact, people scowl “when they’re concentrating really hard, when they’re confused, when they have gas."
