Good morning. Sensing technology—like cameras and microphones—is good at capturing the sights and sounds around us and feeding them into AI models for analysis. But what about the things we smell? 

Sensing these invisible molecules is much less straightforward, yet it holds a lot of business promise—especially for a company like Moët Hennessy.

The luxury wine unit at LVMH said a pesky defect known as Fresh Mushroom Aroma has started to appear across its grape crop and the whole Champagne region in recent years. The defect is undetectable to humans until after the grapes are fermented into wine, when it then spoils the yield. It doesn’t occur every year (most recently manifesting in the 2023 crop). But when it does, the results are disastrous and can cost the company millions of euros in wastage, said Manuel Reman, member of the Moët Hennessy Executive Committee, overseeing R&D.

Finding a way to detect the defect in the grape juice before the lengthy fermentation process was the focus of a recent research collaboration between the winemaker, the semiconductor and software firm Analog Devices and the University of California, Davis.

Analog Devices built a machine that used a new type of chemical sensor developed in-house to capture olfactory data from the grape juice. Then together with Moët Hennessy and UC Davis, it trained an AI model to determine the likelihood that a it was infected with Fresh Mushroom Aroma. Early testing showed 99% accuracy.

“I’m not telling you everybody was crying in the room, but nearly. I still have goosebumps. It’s so huge,” Reman said. 

So how did they do it? 

Today’s sensing technology struggles with smell partly because it tries to identify and measure individual chemicals, a challenging task, said Max Shulaker, chief of Health Solutions at Analog Devices. Shulaker said he took a different approach, building miniaturized sensors that capture the overall aroma fingerprint of a sample. The AI learns to recognize patterns associated with outcomes of interest. 

“Rather than asking ‘What chemicals are present?,’ we ask ‘Does this smell like a good sample or a bad sample?,’ For many real-world applications, that’s the more relevant question,” Shulaker said. 

Moët Hennessy provided data from previous spoiled crops to help train the AI algorithm on what to recognize. 

Moët isn’t ready to start using the machine in production yet. The amount of time it takes to read a sample has come down significantly, but it’s still around two hours—too high given that a small Champagne house like Moët Hennessy’s Krug would need to run around 300 samples (and a bigger one like Moët & Chandon might need thousands), Reman said. 

Analog Devices’s Shulaker said he’s confident about bringing that time down as the AI model continues learning from the samples it takes in, getting smarter about exactly what it’s looking for. 

Reman said he hopes to put some of the devices in limited production next year. 

But beyond detecting Fresh Mushroom Aroma, the possibilities are vast for this technology, according to Ben Montpetit, professor and chair, Department of Viticulture and Enology at UC Davis. For example, smoke from fires is another problem impacting wine production that is often undetectable early in the production processes. 

“In 2020 this cost the wine industry close to $4 billion in losses here in California,” Montpetit said “So we’re also exploring that use case as well as plant viruses that are emerging.”

Matt Calkins, the chief executive of AI automation software company Appian, is taking a firm stance on AI safety: The U.S. needs to institute a strict model “alignment” test, which would ultimately slow the pace of AI development, he told a small group of reporters on Tuesday night.

Model alignment is a term commonly used by researchers to describe AI that acts in ways that match human intentions. And model misalignment refers to AI that acts in ways that ignore or conflict with human intentions.

The idea of model alignment has become more well-known recently as debate rages over whether AI could one day wipe out humans. When taken to the extreme, misaligned AI models could see killing humans simply as a necessary step toward accomplishing their goals.

Avoiding cataclysm. Calkins, who said he doesn’t like to use “cataclysmic language,” still sees misaligned AI as a serious threat: “You come up with a technology as powerful as AI, with the capabilities that it has, and it’s just inevitable that 10 years from now, it’s either massively empowering us or substantially oppressing us,” he said.

His proposed solution is a model alignment test designed by top AI researchers and thinkers, such as Nobel laureate Geoffrey Hinton and Alphabet chief scientist Demis Hassabis. “They would be willing to set the precedent for an alignment test that would be applied to the U.S. and would therefore be copied elsewhere,” he said.

Calkins’s thinking isn’t too different from that of Anthropic and OpenAI leaders, who’ve said they’re researching how they will make sure superintelligent AI models remain aligned. However, both say they don’t yet have a reliable way to do so.

The problem is, Calkins doesn’t think the AI labs will go far enough on their own. “What they’re doing right now is testing alignment gently, realizing that their models are not aligned, and letting them loose anyway,” he said.

Appian itself isn’t strictly an AI company—the firm builds automation software that it sells to governments and enterprises. But to Calkins, talking about AI safety isn’t a matter of selling more software. “I was on an investor tour, and my CFO said, ‘Stop talking about alignment. This isn’t getting you any more investors,’” he said. “But I feel a duty to say this.”

—Belle Lin

• Neocloud company Lambda is raising up to $4 billion in a final round of fundraising before the company’s planned IPO. The new funding will give the company a valuation of $14.5 billion, excluding the amount of the money being raised, WSJ reports.

• France’s Mistral said it would soon release Mistral Large 4, dubbed Le Chonk, a new open-weight AI model that it claimed would rank among the best globally. It plans to release the weights, or the trained parameters of the model, on Oct. 27, WSJ reports.

• A month after its Millennium Prize solution, OpenAI released findings on more than 300 problems—and tried to win back the world of math, WSJ reports.

• Google and Constellation Energy agreed to a 20-year nuclear-power deal that would boost output at 11 existing reactors, the latest tie-up between the tech and energy industries to power new data centers. The WSJ reports that the upgrades will provide additional power to the grid that is roughly equivalent to building a new large reactor.

• JPMorgan Chase CEO Jamie Dimon in an interview with Bloomberg said the release of Anthropic’s Mythos model raised the stakes of cybersecurity risks across the globe. Risks “went up 10-fold after Mythos,” Dimon said. “AI created vulnerabilities that we didn’t know about.”

The WSJ Tech Council brings together CIOs, CTOs and CISOs advancing innovation and shaping the future. Join this trusted community where tech executives connect with peers to explore emerging trends and gain the perspective they need to stay ahead of disruption.

Request Information