By Rick Weiss
Washington Post Staff Writer
Monday, March 10, 2008
The expert taster sat silently in the brightly lighted room, surrounded by 53 samples of ruby-red wine.
Fifty-three sniffs and 53 sips later, the judgment was in: a hint of black cherry . . . some acid . . . a floral nose. Every one of the wines, the taster reported, was an Italian Barbera, and all were made from exactly the same variety of grape.
But there was more. The grapes used for 23 of the bottles were grown in one region of northern Italy, the expert asserted, while those in the other 30 bottles came from a different region -- a region, it turns out, just 60 miles from the first and featuring only minor differences in soil and sunlight.
That degree of discrimination would be impressive in any tasting club. But it was especially notable because, in this case, the expert was a bundle of high-tech chemical probes.
The successful test of that electronic tongue and nose was one of several in recent years hinting that automated food and beverage sensors may someday match, or even outperform, their human counterparts.
Human sommeliers are not at risk of being replaced by machines just yet, scientists say, although a Japanese consortium recently released a Health and Food Advice Robot that can distinguish among 30 kinds of wine, as well as various cheeses and breads, and has the irritating capacity to warn its owner against poor eating habits.
But that day may come. Recent improvements in sensors, and in computer programs able to interpret their highly complex inputs, give credence to the once-discounted idea that machines may someday become the ultimate arbiters of taste.
Last month, for example, the Agriculture Department launched a program that uses machines to grade livestock carcasses as USDA Prime, Choice or Select, the agency's official ratings of tenderness and flavor that for 80 years have been based solely on the judgment of journeymen meat graders.
The system is still being perfected, said William Sessions, associate deputy administrator of the department's livestock and seed program, which oversees meat grading. "But a high percentage of the time we can predict with a large degree of accuracy the eating experience you will have."
The robotic graders, being tested at four Nebraska slaughterhouses, capture photographic images of sides of beef as they cruise by at rates of up to 400 head per hour. The graders focus on the rib-eye muscle, between the 12th and 13th ribs, and measure the redness of the meat, the degree to which it is marbled with tasty fat, and the thickness of the outer fatty layer.
Human graders are being kept on hand to confirm the machines' ratings and override the robots when necessary. But the degree of agreement is very high, officials said. And one handy thing about a hard-wired judge is that it is not susceptible to pressure from plant owners, some of whom have been known to lean on agency graders.
At the same time, placing trust in computers to make such economically important decisions raises risks, including "cyber-security issues," Sessions said, "like somebody being able to hack into the system. It could be a huge advantage to a plant if the equation [in a machine's software] were altered."
For that and other reasons, Sessions said, live employees will be needed "well into the future."
Electronic tongues and noses face even tougher technical challenges than meat-grading cameras. They must properly identify single molecules of interest among the billions or trillions of background molecules in foods, beverages and the air spaces around them. Most of these sensors are wands with metallic or polymer surfaces that undergo changes in electrical conductivity in the presence of particular molecules that contribute to taste or odor.
Complex computer programs are crucial, too, to calculate the myriad ways those molecules will interact -- something the brain does automatically with ease.
Alan Gelperin, a neuroscientist at the Monell Chemical Senses Center in Philadelphia, worked for years at Bell Labs in New Jersey on an "e-nose," developed for use by grocery checkout clerks. The initial goal was to make a sniffer that could differentiate between navel and juice oranges in two seconds, the average time a checker spends ringing up an item, instead of the 30 seconds or so that earlier e-noses needed to get a clear reading.
"Looking back on it, it was truly a miracle that we succeeded," Gelperin said. "And the patent made it sound like every supermarket would have one of these things within a few months."
But the device never got commercialized, in part because of another innovation: Shoppers started putting the produce in plastic bags.
"How are we going to sniff?" Gelperin recalled, asking with obvious exasperation. "Are we going to stick a needle in there?"
Many university laboratories and companies around the world have joined the race to develop e-noses and e-tongues, some with such evocative names as FreshSense and LibraNose.
The journal Nature last month described a Japanese-made hand-held fiber-optic infrared sensor that measures oleic acid levels in raw meat, reputed by some to be a better indicator of tastiness than conventional marbling measures.
In Australia, supermarkets already are using a tastiness rating system in which cuts of meat earn labels with as many as five stars, based on a number of automated inputs including fat depth, extent of marbling and post-slaughter aging.
And in Russia, Andrey Legin and colleagues at St. Petersburg University have made an electronic tongue that can distinguish among various blends of coffee or soft drinks just as accurately as people can-- a potential boon for beverage makers wishing to ensure batch-to-batch consistency while avoiding the expense of human tasting panels.
Of all the potential applications for robotic tasters and sniffers, wine is perhaps the most attractive -- and also the most controversial.
As the Italian Barbera experiment showed, e-tongues and e-noses can be very good at differentiating among closely related wines, which means they can help authorities identify counterfeit and mislabeled wines. "The origin of wine for us is very important," said Saverio Mannino, chief of food science and microbiology at the University of Milan, who led that test.
Such systems are also very good at detecting contaminants in wine, such as the mold or must that can come with bad corks.
But wine enthusiasts bristle at the idea that a machine might someday get the final word on whether a wine is truly great.
"I guess it would be nice if I could sit with my laptop and a sampling machine and have it spit out a tasting note," said Anthony Dias Blue, editor of The Tasting Panel Magazine, a Los Angeles-based industry monthly. "But no machine can muster the level of creative smoke-blowing that wine writers can come up with to describe what they are tasting. It takes that special human ability to come up with the hyperbole to really describe a wine."