CEO DailyCFO DailyBroadsheetData SheetTerm Sheet

Why corporate use of A.I. flatlined in 2020

April 20, 2021, 4:10 PM UTC

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.

Corporate enthusiasm for artificial intelligence appears to have been muted by the pandemic.

Instead of focusing more on A.I. in 2020, as had been expected before the virus hit, companies ended up spending about the same amount of time and money on the technology, according to a survey released this week by O’Reilly Media.

Twenty-six percent of survey respondents—from data scientists to executives—said they operated A.I. products that generated revenue. That roughly mirrored the responses respondents gave in 2019.

Additionally, 35% of businesses said they had evaluated or tested, A.I. projects last year—a number that was, again, mostly unchanged.

O’Reilly Media vice president Rachel Roumeliotis attributed the A.I. flat line to the fact that this year’s survey had three times (3,754 respondents) more respondents than the previous one. Perhaps the majority of the new survey respondants were only beginning to test or operate production-ready A.I. applications, offsetting the A.I. progress of last year’s respondents.

But based on anecdotal observation, Roumeliotis acknowledged that some companies have delayed A.I. projects because of reduced business spending during the pandemic.

“Some people were stepping back,” Roumeliotis said. “The nice-to-haves were maybe slowed down.”

Indeed, as Fortune has previously reported, many companies have paused more expensive technology projects, such as A.I., to cut costs. Companies were more likely to spend money on pressing technology projects, like video-conferencing software for remote workers, rather than A.I. that may not pay off for years.

But Roumeliotis expects that companies will invest more in A.I. as the economy recovers. Part of that will involve businesses shifting to tools from cloud computing giants like Amazon, Microsoft, and Google because of the possibility that those tools could be easier and cheaper to use than more conventional A.I. technologies.

Currently, many companies and researchers use open source, or free, A.I. developer tools in combination with powerful computers that they operate in their own data centers to create deep learning models that spot credit card fraud or cybersecurity problems. However, these projects require skilled A.I. experts, of whom there is a limited number.

If cloud A.I. tools are cheaper and easier to use than more conventional methods, businesses could be more willing to experiment with A.I., Roumeliotis believes.

“I think there will be a second wave of A.I. adoption,” Roumeliotis said, referring to a boom created by companies switching to cloud A.I. software.

If there’s a positive to the pandemic’s impact on A.I., Roumeliotis said it’s that some A.I. applications that have helped, such as the various chat tools used by healthcare firms for medical triaging, have made A.I. more attractive. 

“I’ve never seen that much in the news before,” Roumeliotis said about media reports of companies using machine learning to predict COVID-19 hotspots, among other uses. “I do think people have gotten more familiar with it.”

Whether that increased familiarity with A.I.’s potential leads to more investment remains to be seen.

Jonathan Vanian 


Self-driving tragedy? Local authorities in Texas toldThe Wall Street Journal that two men died as a result of a Tesla vehicle crashing into a tree without anyone behind the wheel, implying that the car’s semi-autonomous driving features were activated. Tesla CEO Elon Musk, however, claimed that the so-called Autopilot feature “was not enabled,” and that the vehicle’s owners did not buy Tesla’s “full self-driving” upgrade, CNBC reported. The National Highway Traffic Safety Administration and the National Transportation Safety Board are investigating the accident, the report said.

Google’s A.I. doesn’t speak legalese. Google’s language translation tool stumbles when translating legal terms in different languages, Reuters reported, citing a new research study. In an example of one egregious mistranslation, the study’s authors found that “Google's translation software turns an English sentence about a court enjoining violence, or banning it, into one in the Indian language of Kannada that implies the court ordered violence.” The finding underscores the difficulty A.I. researchers have creating language systems that understand how context can alter the meaning of certain words. It also shows that companies that operate in specific industries, like LexisNexis, for instance, could be better suited to create language systems tailored to specific verticals because they have the appropriate propriety data not found in the open Internet.

Putting the brakes on A.I. The European Union is considering banning certain uses of A.I. that could be considered detrimental to society, the BBC reported. Some of the A.I. applications that could be forbidden include tools that provide “indiscriminate surveillance applied in a generalised manner,” software that “manipulates human behaviour,” and any uses involving “social scoring,” the report said. Still, the EU considers military use of A.I. as OK because the technology can “safeguard public security.”

Volvo steps on the A.I. gas pedal. Volvo will provide “hundreds” of its vehicles, outfitted with autonomous driving software and hardware, to Didi Chuxing so the Chinese transportation giant can test robot taxis, Bloomberg News reported. The report noted that Didi plans to raise as much as $500 million for its self-driving unit as it’s preparing for a public listing that could happen as early as this quarter.


Salesforce picked Silvio Savarese to be the chief scientist and executive vice president of the enterprise software giant’s research arm. Savarese was previously an associate professor at Stanford University and was also the director of The SAIL-Toyota Center for AI Research at Stanford. He was also the co-founder and chief scientist of Beijing-based A.I. startup AiBee.

Nvidia hired Sarah Tariq to be the chip-maker’s vice president of autonomous driving. Tariq was previously the senior director of perception for the self-driving car startup Zoox. Before joining Zoox, Tariq spent nearly a decade at Nvidia and was once the technical lead on an autonomous driving project related to self-parking.

AI.Reverie chose Aayush Prakash to be the startup’s head of machine learning. Prakash was previously a deep learning researcher at Nvidia’s A.I. research lab in Toronto. He will help AI.Reverie develop its tools that help businesses create synthetic data to power deep learning systems.


The future of A.I. could be GLOM. Deep learning pioneer Geoffrey Hinton published a paper about a theoretical approach to neural networks, dubbed GLOM, that he believes could lead to more advanced A.I. systems. He notes that if GLOM (taken from the slang term “glom together”) “can be made to work,” it may be easier for A.I. researchers to interpretate how certain deep learning systems reach their conclusions, among other feats. Although this is merely theoretical at this point, Hinton’s contributions to modern-day A.I. are so important that this paper has received some special attention.

In an interview with MIT Technology Review, Hinton discussed how GLOM could potentially lead to computers that can mimic human intuition.

From the article:

Hinton is the first to admit that at present GLOM is little more than philosophical musing (he spent a year as a philosophy undergrad before switching to experimental psychology). “If an idea sounds good in philosophy, it is good,” he says. “How would you ever have a philosophical idea that just sounds like rubbish, but actually turns out to be true? That wouldn't pass as a philosophical idea.” Science, by comparison, is “full of things that sound like complete rubbish” but turn out to work remarkably well—for example, neural nets, he says.

GLOM is designed to sound philosophically plausible. But will it work?


Data-labeling company Scale AI valued at $7.3 billion with new funding—By Jeremy Kahn

Robots are now scanning shelves at Save Mart and Lucky Supermarkets in California—By Jonathan Vanian

Google Earth’s new tool shows the Antarctica-melting ruins of climate change—By  Danielle Abril

The idea of COVID-19 vaccine passports raises privacy concernsFortune Editors

Why Facebook and LinkedIn’s data scraping fiascos are a huge security problem for their users—By  Jonathan Vanian


The road to Aurora.The autonomous vehicle startup Aurora is the subject of a Bloomberg Businessweek profile that probes the nascent but competitive world of self-driving cars. Aurora recently acquired Uber’s Advanced Technologies Group as part of a $400 million deal in which Uber would take a major stake in the company.

Aurora CEO and co-founder Chris Urmson is no stranger to self-driving vehicles, having spent 8 years working at Google’s self-driving car unit, which now operates as the Alphabet subsidiary Waymo. Now the pressure is on for Urmson to help usher autonomous driving to the mainstream, an effort that will require an incredible amount of money. Like competitor TuSimple, which went public last week, Aurora aims to revolutionize the trucking industry.

From the article:

Urmson says Aurora’s driving system can become better than the average trucker in a matter of years, not decades. He has a target in mind for when the first trucking product will be ready, though he’s not yet willing to share it. He knows as well as anybody how the work of building autonomous vehicles can expand endlessly.

Our mission to make business better is fueled by readers like you. To enjoy unlimited access to our journalism, subscribe today.