AI can already see, hear, read and speak. Now researchers are teaching it to smell.

Over the past few years, scientists have been developing and refining a technology called the e-nose—which is exactly what it sounds like. These systems detect and distinguish aromas, sometimes with about 1,000 times as much precision as humans can, and without the loss of sensitivity that comes when our noses get accustomed to an odor. AI can then scrutinize those smells, and the tech can figure out, for instance, exactly what volatile gases comprise a scent, in what combination, and what those combinations mean.

And they can mean quite a lot. Right now, researchers are exploring—or even commercializing—e-nose systems that can scan a person’s breath to detect deadly infections, sniff the air in a building to seek out signs of potential contaminants, or even develop perfumes more quickly and cheaply than before.

The path to the perfect e-nose, however, isn’t a smooth one. That’s because getting an AI to identify smells is very different from having it analyze images. Many environmental factors can affect the accuracy of e-noses—including the humidity, how widely a smell is dispersed, and even how it is “inhaled” by the device, says Haritosh Patel , a postdoctoral fellow studying biologically inspired engineering at Harvard’s School of Engineering and Applied Sciences.

And unlike images—which can be distilled into objective gradations of red, green and blue light—smell is highly subjective, making it tough to standardize and calibrate. Then there is the lack of data: Visual images, audio and text have been stored in astounding quantities on the internet for decades, giving AI lots of raw material to train on. But there is no equivalent store of scent data, and creating one will take a long time.

“It took about 30 years for computer vision technology to mature,” says Jack Liu , head of corporate development at Ainos , a Texas artificial-intelligence company focused on AI scent technology. “I think we are at the beginning of those evolutions in smell data.”

Here are some of the most intriguing e-nose applications in the works.

Sniffing out a diagnosis

Ancient Greek doctors used the odors in patients’ breath to judge their health. Now scientists want to use AI to revive this practice.

In testing, scientists have shown that e-nose technology reliably detects lung cancers , urinary-tract infections , and gastrointestinal disease . The e-noses can sniff out the unique signatures of volatile organic compounds not only in the breath, but in excretions like sweat or urine, and have AI quickly identify them. However, commercializing such technologies has been challenging, due to fiscal and regulatory hurdles in the healthcare industry.

The National Aeronautics and Space Administration, meanwhile, has developed e-nose technology to track astronaut health on long-distance or long-duration missions. And the agency says it has shown promise in detecting scent signatures linked to stomach ulcers, and infectious and inflammatory diseases , among other things. ​​It also tested well in detecting Covid-19 and certain forms of influenza.

Simpli-Fi Automation of St. Paul, Minn.—which has an exclusive license to use NASA’s scent technology for medical applications—is preparing to field-test a commercial device that analyzes breath on a larger scale. The device studies the air in hospitals to search for traces of common but debilitating infections that are time consuming and expensive to treat.

Boosting a patient’s senses

Harvard professor Joanna Aizenberg ’s lab just received seed funding, in collaboration with the Massachusetts Institute of Technology and Mass General Brigham healthcare system, to use AI to help patients with a diminished sense of smell. The initial goal is to create a diagnostic test to measure olfactory loss—a marker of Covid-19 and other conditions. Then the team hopes to treat the problem by placing sensor chips in the patient’s olfactory tissue to help preserve or restore scent perception. Patel, who is part of the lab, likens it to “a cochlear implant for the nose.”

Seeking out impurities

Olfactory detectors like radon monitors or smoke alarms have existed for decades, but they mainly search for one danger only. “A carbon-monoxide sensor only monitors carbon monoxide. Any other gas, there is no reaction,” says Ainos’s Liu.

The goal for AI-powered e-noses is to create what Patel calls “a universal smeller,” which can detect numerous hazards at the same time. These gadgets could hunt for hazardous compounds like benzene or mold in home interiors, he says, or potential threats to first responders in wildfires or burning buildings. They could even be used to track the skin or breath signatures of disaster survivors.

Some companies are testing out that kind of wide-ranging sensor. In February, Ainos began deploying e-nose technology to monitor testing and packaging facilities related to semiconductor manufacturing. It is also testing its sensor array in chip-making facilities.

These highly automated sites run 24/7 and can’t afford even tiny traces of extraneous gases or chemicals to contaminate their precision production, says Liu. The sensors can also help with proactive maintenance, by tracking early scent signatures of electronic components overheating.

Creating aromas

Some companies, meanwhile, are studying how to use e-noses and AI to create scents. Designing original, safe fragrances for perfumes or consumer products can be a lengthy and expensive process, says Alex Wiltschko , founder of Osmo, a New York City AI company. The company hopes to change that with an “olfactory intelligence model”—an AI system that uses the company’s scent database to predict the odor character of certain molecules and how they will smell once they are combined, among other things.

“Then we make that into an actual fragrance formula,” Wiltschko says.

Osmo has created fragrances for a variety of clients including Loucil, Alice Panikian and the Museum of Pop Culture. It also “vibe scented” a horror movie, creating a fragrance, distributed at screenings, that “captures the chilling, damp, metallic stillness of a nightmare that won’t let you escape,” according to the film’s press materials.

The technology could also be used to create signature scents emitted by businesses—retail, hospitality, aviation, entertainment—lodging them in consumers’ memories to enhance brand connection. For example, subtle scents that cue feelings of relaxation or trust, atomized into an establishment, might cue visitors to stay longer and shop or purchase more, or associate the brand with a sense of well-being.

Of course, the holy grail of olfactory tech is blocking undesirable odors, like noise-cancelling headphones for your nose, according to Paul Liang , director of the Multisensory Intelligence research group at the Massachusetts Institute of Technology’s Media Lab. Finding a solution that works for all smells may prove difficult, though, in part because of all the vagaries in how we uptake and interpret scent. “I don’t think there’s a catchall solution,” says Osmo’s Wiltschko.

Finding counterfeits

When products are made, their component materials, and even the factories they are made in, give them a unique scent signature, says Wiltschko. When counterfeiters make fake versions—phony name-brand sneakers, for instance—they don’t use the same materials, glues, dyes and paints that the legitimate manufacturer does. Different components mean different scents—and a way for AI to identify fakes.

The scents “show up as sense fingerprints that allow us to differentiate even where they came from,” says Wiltschko. The company has worked with a sneaker reseller to detect dupes with high accuracy, he says.

This technology could potentially be applied to authenticating wine, perfumes or other luxury goods, says Liu of Ainos, which trained its e-nose to differentiate varieties of coffee beans.

Keeping livestock healthy

Livestock farming contributes around 5% of America’s greenhouse-gas emissions. E-noses could be used to diminish that output. If farmers can figure out, for example, that a certain breed of cattle releases more methane when it consumes a particular kind of feed, they could change the feed to cut down on the methane.

New technologies could place e-nose sensors on the wall or ceiling of a barn or feedlot to measure emissions and allow people to respond in real time. Aizenberg’s lab is developing an e-nose device that livestock would wear as an earring, allowing unconstrained real-time data collection.

Agscent, an Australian company that has an exclusive license to use NASA’s scent technology in agriculture, is focusing on bovine pregnancy testing. Right now, such tests typically require very invasive physical exams. The company says its product, which works solely with bovine breath samples, has detected gestation with 94% accuracy in lab tests. It hopes to have the product available commercially by the end of the year.

The company is also targeting early detection for bovine respiratory disease, which costs the American cattle industry $1 billion annually, says Ashley Sweeting , Agscent’s vice president for the U.S.

Brett Berk is a writer in New York City. He can be reached at reports@wsj.com .