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.
A lot of attention is being given to the role that artificial intelligence can play in helping to combat the coronavirus pandemic.
Much of it been focused on areas with a high degree of what, for lack of a better term, I’ll call sex appeal: A.I. that can spot and track emerging epidemics; A.I. that can help with contact tracing and social-distancing enforcement; A.I. that can possibly help diagnose COVID-19; and, perhaps most importantly, A.I. that might find possible treatments or vaccines.
Last week, Stanford University’s Institute for Human-Centered Artificial Intelligence held a virtual conference in which researchers discussed many of these uses of A.I. (If you didn’t catch the live broadcast, I highly recommend watching the recorded version here.)
This is all important and potentially game changing. And I don’t want to belittle the work of these researchers and companies in any way. But much of the technology is relatively immature and unproven. It will likely take time to validate and perfect. A lot of these machine learning techniques will probably be more helpful in ensuring we can beat the next emerging epidemic than in fighting this one.
If you want to know where A.I. might be able to have the biggest impact in the battle against coronavirus, it might be wiser to look at some uses of the technology that are more mundane and, frankly, boring.
To show you what I mean, I want to tell you just a little bit about my wife. After having spent the better part of two decades in the diplomatic service, she recently retrained as a medical doctor. She’s now a junior doctor here in London, and she’s working, like most of her colleagues, on the front lines of this pandemic. Last week, she worked three consecutive overnight shifts in her hospital’s intensive care unit.
What is one of the biggest challenges her hospital has faced? Scheduling.
Rostering medical staff—“doing the rota” in the parlance of the British National Health Service—is a big administrative hassle even in normal times. Most British hospitals employ a small cadre of administrators whose sole job is to perform this thankless task, usually using an array of Excel spreadsheets. (U.K. hospitals are notoriously behind-the-times when it comes to administrative I.T.)
During the current crisis, scheduling has become monumentally more complex: Staff are moved off their usual assignments so they can work in emergency departments and intensive care units, routine elective surgeries have been cancelled, leaves have been curtailed, shift times have been extended from 8 hours to 12 hours, final-year medical students, graduated early, have been drafted in to help fill staffing gaps, and a large number of workers are absent on any given day because they are ill themselves or are self-isolating because a member of their household is sick. Some British hospitals have reported absence rates as high as 30%.
On top of all this, here in the U.K., the government has been asking retired doctors and nurses to return to active service. So far, more than 11,000 of them have answered that call. The government has said it also wants to recruit 250,000 volunteers, from all walks of life, to help support the NHS and other vital services during the crisis.
Figuring how best to deploy these volunteers alongside regular staff is a colossal job. But, it turns out, A.I. is ideally suited to help—and the technology has already proved its mettle in the corporate world.
Last week, I spent some time chatting with Helge Bjorland and Jan Kristiansen, two co-founders of Globus.AI, a startup that has created software for exactly this purpose. The company, founded in Oslo in 2017 by four friends who had worked together in Norway’s oil sector, helps companies across industries find the right workers to fill shifts.
Among Globus’s customers is Scandinavian medical staffing company Dedicare. Globus helps the company match doctors and nurses with shifts in private hospitals. The software was handling about 4,500 shifts per week, Kristiansen, Globus’s chief operating officer, says.
When the coronavirus pandemic began spreading, Globus realized it might be able to help, says Bjorland, the company’s CEO. Globus tweaked its software and offered it for free to Norway’s public hospitals. The software allows a hospital to match healthcare workers’ competencies to its needs and to align doctors’ availability with open shifts.
The software, Kristiansen says, saves about 90% of the time it takes to fill each available slot, saving a rota manager between two and four hours every day. What’s more, because it can more efficiently match staff to slots, hospitals find their allocation capacity actually increasing between 30% and 40%. And it only takes an administrator about an hour, he says, to learn to use the software.
The technology is not that fancy—though Globus uses natural language processing to extract some information and deep learning to do some matching of candidates’ competencies to jobs, it also uses much simpler machine learning techniques, like logistical regression, to help fill available time slots. And it incorporates some good old-fashioned rules to take into account legal requirements, such as those that limit the number of hours doctors and nurses can work in one stretch, or hospital policies, like the need for at least one senior doctor to be rostered on to each shift to supervise more junior staff.
So far, the system has been deployed in Oslo and Sola, another Norwegian city. Ernst & Young, with which Globus has a partnership, is helping the company roll the system out to public hospitals elsewhere in the country.
But Globus wants to help hospitals and healthcare organizations around the world. “The main thing for us is having the word out to let other countries know that we have actually have something that can help them,” Kristiansen says.
If I had to bet, it will be simple uses of A.I. such as this that wind up being the real life-saver in this pandemic.
And now here’s the rest of this week’s news in A.I.
Jeremy Kahn
jeremy.kahn@fortune.com
@jeremyakahn
A.I. IN THE NEWS
Apple acquires A.I. startup Voysis to improve Siri. The Cupertino-based tech giant acquired Voysis, a Dublin, Ireland-based A.I. startup that specialized in voice applications, for an undisclosed sum, according to Bloomberg News. Apple may use Voysis's technology to bolster the capabilities of its Siri digital assistant, which is perceived to lag behind those of rivals Google and Amazon.
Investment firm linked to Ikea acquires A.I. home-decor startup. Ingka Investments of the Ingka Group, a company that owns and operates 380 Ikea stores in 30 countries, has acquired Geomagical Labs, a U.S. A.I. startup, for an undisclosed amount. The company, founded in 2016, creates software that lets people take photographs of a room and then create three-dimensional simulation of the space that lets them experiment with different interior design ideas.
Athena Security's coronavirus-fever detection is under fire. The Austin-based company is known for creating a camera surveillance system that it said could detect weapons in video feeds. Now it says it has trained a computer vision system to take video from a thermal imaging camera and detect whether someone has a fever. But an investigation by video surveillance trade publication IVPM found that Athena fabricated several elements of its marketing materials for the new product. Athena has denied the allegations.
British A.I. startups ask government adviser for financial support. A group of British A.I. startups has written to Dominic Cummings, a chief adviser to Prime Minister Boris Johnson, appealing for funds to help them weather the economic downturn caused by the coronavirus pandemic, The Daily Telegraph reported. In their letter, CEOs say that if they are forced to put their businesses into "six months of hibernation," the U.K. risks falling far behind other nations, particularly France and Germany which, the CEOs note, have previously announced billions in dedicated funding for A.I. companies.
U.K. health service calls in Big Tech to help tackle Covid-19, but privacy concerns haunt the project. The British National Health Service (NHS) has commissioned a number of tech companies, including Amazon, Microsoft, and Palantir, as well as the A.I. consultancy Faculty AI, to create a centralized data system that will show how the coronavirus outbreak is spreading across the country and the availability of medical resources to fight it, Business Insider says. But elements of the system are raising privacy concerns, The New Statesman and The Daily Telegraph report.
EYE ON A.I. TALENT
Lyft, the ride-hailing firm, has boosted its autonomous driving research with the hiring of Sacha Arnoud. Arnoud peviously led a team at self-driving car company Waymo that worked on vehicle perception, according to a story in The Information.
Kneron, a San Diego, Calif., A.I. company, has hired Davis Chen to be its vice president of engineering. Chen had previously worked as head of engineering in Taipei for chipmaker Qualcomm, where he spent 25 years. Kneron specializes in A.I. that works "on the edge," running on devices such as phones or smart speakers, without the need to send a constant stream of data back to a cloud-based server.
Microsoft has hired Ruben Cabellero as a corporate vice president, working on both mixed reality products and artificial intelligence, according to Bloomberg News. Cabellero had previously been vice president of engineering in charge of developing wireless technology at Apple.
U.S. President Donald Trump has nominated Sethuraman Panchanathan to be the next director of the National Science Foundation. Panchanathan, an expert in cognitive computing, is currently Arizona State University's executive vice president and chief research and innovation officer.
EYE ON A.I. RESEARCH
Helping medical professionals make sense of A.I. models. "Model cards"—sort of like baseball cards for algorithms—that can help software developers quickly grasp the advantages and limitations of various algorithms have recently gained traction. Now researchers at Duke University, writing in the journal Nature Digital Medicine, have proposed a similar system designed to help healthcare professionals quickly understand where various algorithms designed for medical settings might be helpful—and, critically, when they might be harmful to patients.
Researchers cite a model Microsoft created in 2015, based on historical data from hospitals, that helped assess which pneumonia patients were at the highest risk of death.
The model famously found that patients with asthma were at lower risk. But when doctors queried this counterintuitive assessment, they discovered this was only because, in the historical data, asthma patients were more likely to end up in intensive care, where they received a high degree of attention without which they most likely would have died.
"There has not been a systematic effort to ensure that front-line clinicians actually know how, when, how not, and when not to incorporate model output into clinical decisions," the researchers write. "Nor is there an expectation that those who develop and promote models are responsible for providing instruction of model use and for the consequences of inappropriate use."
An A.I. for air traffic control. Researchers from IBM's Singapore research lab and Amazon Web Services have taken a publicly available air traffic control simulator created by Eurocontrol, the agency which handles air traffic throughout Europe, and used it to train a reinforcement learning algorithm, in which a system is trained through trial-and-error.
The algorithm the researchers designed involved one part that is trained to "minimize the combined cumulative long-term effect of conflicts, congestion, delays and fuel costs" and another part that looks at each individual aircraft. The software must learn to arbitrate between these two components and then recommend adjustments to the air speed of each plane to avoid collisions while meeting its other goals.
According to their paper published on Arxiv.org, the non-peer reviewed research repository, the new algorithm "significantly outperforms state-of-the-art benchmarks" including previous A.I. work using the same dataset. But, the researchers caution, the system needs further work to incorporate other aspects of air traffic control, such as suggesting changes to aircraft headings and altitude. They also say the system is probably more applicable to controlling aircraft in transit, rather than during take off and landing.
FORTUNE ON A.I.
Teaching a machine to see: Italian doctors turn to Chinese A.I. to diagnose COVID-19—by Eric J. Lyman
Your next potato chip may be flavored by artificial intelligence—by Katie Sehl
LinkedIn tries to improve equality across its site—by Jeremy Kahn
Some of these stories require a subscription to access. Thank you for supporting our journalism.
BRAIN FOOD
OK, maybe this is more like "brain candy" than brain food. But if you're looking for amusement in these dark times, you might want to check out the website "AI Gahaku," created by a developer who goes by the name Sato and who says he is located in Japan. Gahaku is a Japanese word meaning master painter or artist, and the site lets you upload a headshot or selfie and transform it into the style of a classically painted portrait.
AI Gahaku uses an algorithm trained to do "style transfer," morphing one image into the style of another. There are different ways to do this—but most of the best results involve using a generative adversarial network (GAN). That's the A.I. technique invented by Ian Goodfellow in 2015 in which two different deep learning networks, one which generates images and another that tries to classify them, compete against one another. The generator is pushed by the classifier towards the original artists' style, forced to get closer and closer until it can reliably fool the classifier into thinking the new image is part of the original artists' oeuvre.
It is just one of a growing number of uses of A.I. to create art. But is the work of AI Gahaku art at all? Ah, here's the big, meaty, brain food question: What is art? (See, I tricked you into it with that sweet appetizer.) I would argue that art requires "intent" on the part of the artist. It doesn't matter what that intention is—to express something, to make the viewer feel or think, or simply to bring beauty into the world—but the intent must be there. An A.I. algorithm has an objective function, sure, but I would argue that this is not the same thing as a human artists' intent—and so this kind of automatically generated A.I. art is not art at all. Others might disagree. Or they might say that the real art here is Sato programming AI Gahaku. That way, AI Gahaku is itself art, even if its individual creations are not. What do you think?
