CEO DailyCFO DailyBroadsheetData SheetTerm Sheet

Why Most Companies Are Failing at Artificial Intelligence: Eye on A.I.

October 15, 2019, 3:55 PM UTC

Most companies that say they’re using artificial intelligence have yet to gain any value from their A.I. investments. 

A survey from MIT Sloan Management Review and Boston Consulting Group released Tuesday found that companies that view A.I. as merely a “technology thing,” akin to a product rather than a business overhaul, fail to gain financial results. The survey’s authors defined the “value” of an A.I. project as lifting sales, reducing costs, or creating a new product.

The survey, based on responses from nearly 2,500 executives, found that seven out of ten companies report little to no impact from their A.I. projects so far. Overall, 40% of the surveyed companies that have made “significant investments” in A.I. have yet to report any business gains.

There is a clear difference in the A.I. strategies between the “winners” and “losers,” according to Boston Consulting Group managing director Shervin Khodabandeh. For instance, companies that are getting some value from their investments view A.I. as a way to upend and change current business practices likes sales, rather than simply buying an A.I. tool from a vendor, he said.

Also, at the most successful companies, business leaders oversee A.I. initiatives. These executives, who control budgeting and resources, then build a group of data scientists and key personnel from departments like sales or marketing to oversee the A.I. project to completion.

This process is markedly different than the traditional technology approach at most businesses, in which CIOs decide which data-crunching projects to pursue. The downside to this CIO-driven tactic, Khodabandeh said, is that the A.I. projects become isolated and neglected by the overall executive team. 

The report confirms the findings of other recent surveys about A.I. and business that show companies struggle with their data-crunching initiatives. A KPMG survey earlier this year found that most executives believe it will take many years before their A.I. projects create a “significant return on investment.” 

Beyond the latest survey, Khodabandeh said companies that are successful in using A.I. often create their own mini-IT departments, built specifically for A.I. projects. Doing so allows the companies to brainstorm a specific business process they want to improve, like forecasting which products to sell, and then letting their data scientists pick and choose the A.I. technologies to do the job.

“He or she starts with something like, ‘I want my marketer to do their business differently,’” Khodabandeh said about how business-side executives should approach A.I.. “They don’t say, ‘I need reinforcement learning.’”

Jonathan Vanian 
@JonathanVanian
jonathan.vanian@fortune.com

Subscribe to Eye on A.I. here.

A.I. IN THE NEWS

Rush to market. The U.K. government is using facial-recognition technology to check people’s passports, even though officials know that the technology doesn’t work well on people with darker-color skin, the New Scientist reported. The article said that “documents released by the Home Office this week show it was aware of problems with its website’s passport photo checking service, but decided to use it regardless.”

Human eyes are watching you. Amazon is using human reviewers to annotate and label some video recordings captured from the company’s Cloud Cam security product, Bloomberg News reported. The report said that those “video snippets are then used to train the AI algorithms to do a better job distinguishing between a real threat (a home invader) and a false alarm (the cat jumping on the sofa).”

Bye-bye bankers. Nearly 200,000 bankers could lose their jobs by 2030 because of A.I., the Winston-Salem Journal reported, citing a Wells Fargo analyst report that was not released in its entirety to the media. The article cited a senior Wells Fargo banking analyst who wrote, “There is recognition that tech can enable the biggest capital-for-labor swap in the history of banking, thereby allowing tech spend to reduce non-tech spend.”

California says no. California governor Gavin Newsom passed a bill that blocks police departments from using facial-recognition technology in their body cameras, The San Francisco Chronicle reported. The law, which takes effect in January, is considered a victory for advocates of civil liberties, the newspaper said.

REMEMBER THE SPACE RACE?

Pascale Fung, the director of the Centre for A.I. Research at the Hong Kong University of Science and Technology, explains how the Trump Administration’s blacklisting of Chinese A.I. startups could result in Chinese companies developing their own computer chips to wean themselves from using U.S. technology. "It's a very familiar old story," Fung says. "It's like the space race in the old days, and now this is an A.I. race between two different superpowers."

EYE ON A.I. TALENT

Elemental Cognition, an enterprise software startup, added Breyer Capital CEO Jim Breyer, Riverwood Capital managing partner Chris Varelas, and former IBM CEO Sam Palmisano to its advisory board. The startup was founded by David Ferrucci, who led the original IBM team that built Watson.   

Enterprise startup Data Republic hired Geoff Sweeney to be its chief technology officer. Sweeney was previously the global engineering head of Australian software company Nuix and the vice president of software engineering and operations at Mastercard.

EYE ON A.I. RESEARCH

Can you solve a Rubik’s Cube with one hand? A team from A.I. research group OpenAI published a paper about using reinforcement learning—in which computers learn through trials—to teach a robotic hand to solve a Rubik’s Cube. Peter Welinder, an OpenAI research scientist, told Fortune that OpenAI used a variant of the reinforcement learning technology that powered its A.I. system that beat professional video game players in the strategy game Dota 2.

The OpenAI researchers said their paper is important because it shows how reinforcement learning can be used to master increasingly difficult challenges, such as teaching a robotic hand to manipulate an object like a human. The researchers conducted “tens of thousands of years” of virtual attempts via simulation software to ensure that the A.I. system could handle any complications that could occur in the physical world, said OpenAI researcher Matthias Plappert.

Plappert estimated that the data training process required 64 graphics processing units (GPUs) and a whopping 30,000 computer processors, which underscores the tremendous amount of computing required for advanced reinforcement learning projects.

“Just demonstrating that this is feasible, that this approach unlocks the potential in robotics is pretty significant,” said Plappert. “It wasn’t clear before when we started working on this two years ago.”

FORTUNE ON A.I.

10 Best Large Workplaces in Manufacturing and Production By Fortune staff

Is A.I. a Trillion-Dollar Growth Engine or a Jobs-Killer? There’s Reason for Optimism – By Bernhard Warner

Pinterest Says It’s Using A.I. to Dramatically Reduce the Amount of Self-Harm Posts Users Are Seeing– By Lisa Marie Segarra

Amazon Says Its Delivery Drones Won’t Crash Into You or Your Clotheslines. Here’s Why– By Bernhard Warner

A.I. Remains a Disruptive Force in Finance—Even for Fintechs– By Bernhard Warner

How A.I. Is Changing Virtually Every Business, from Commodities to Cosmetics– By Bernhard Warner

BRAIN FOOD

Surveying the A.I. frameworks. Tech publication The Gradient analyzed the state of A.I. frameworks to learn which of the toolkits are getting traction with businesses and researchers. Data scientists use these open-source frameworks to build neural networks, the software created decades ago to loosely mimic how the human brain learns. Although many companies like Amazon and Microsoft have created A.I. frameworks, the most popular are Facebook’s PyToch and Google’s TensorFlow. The article describes how PyTorch has become a favorite tool for researchers while companies are predominantly using TensorFlow. And while PyTorch is growing faster than TensorFlow, the article said, “It will certainly take a long time before PyTorch can make a meaningful impact in industry—TensorFlow is too entrenched and industry moves slowly.”