What the Nose Knows - Avery Gilbert [108]
Executives in the fragrance and flavor business dream of an e-nose that could stand in for a consumer test panelist. The device would be programmed with the exact preferences of urban preteens or suburban soccer moms in different zip codes. When presented with a test sample, it would respond “I like it” or “it’s too floral.” A roboconsumer has many advantages over human panelists: it’s always on time and you don’t have to pay it.
A surprising number of scientists are working on smell-capable robots; one of them published an entire book on the topic in 1999. Amy Loutfi, a researcher at Sweden’s University of Örebro, has attached an e-nose to an intelligent, mobile robotic system. Her prototype resembles a Roomba—it wanders around an apartment under its own control, locating and identifying smells in the air. Loutfi improved her nose-bot’s performance by adding psychological context to its decision-making process. The device identifies smells better when it knows it’s in the living room rather than the bathroom.
Will police departments deputize the e-nose for remote drug sniffing? The U.S. Supreme Court held that thermal imaging of a suspected marijuana grower’s home, because it relies on sense-enhancing technology that is not “in general public use,” is an unconstitutional invasion of privacy. Under this standard, waving an e-nose downwind of a suspected grow house would also violate the Fourth Amendment’s guarantee against unreasonable search and seizure. Until e-noses are available at Circuit City, police officers are going to have to rely on their own noses.
As with all technology, the law of unanticipated consequences will undoubtedly affect how the commercial e-nose market develops. For example, one near-term application is a pocket-size sniffer that tells from a woman’s breath whether she is ovulating. The Ovulatron 5000 certainly will be a boon to couples trying to conceive, but it might also become a must-have technology for single guys on the prowl.
YOU CAN’T EXPECT an e-nose to work as soon as you take it out of the box. Training is essential, even to achieve competence at a simple same/different task. If its job is to pick out rotten apples, you must fill its database with examples of good apples and bad apples, so that it can create a statistical profile for each, and a decision-rule for telling them apart. An untrained e-nose would probably group wine samples according to alcohol content. It must be trained to distinguish Pinot Noir from Zinfandel. An e-nose is only as impressive as the training it gets. You can’t follow your e-nose—you have to lead it.
An electronic sensing device appeals to hard-boiled process engineers because it is “objective.” It frees them from discussions with sensory experts, and from dealing with emotional consumer panelists, at least in theory. But wait until the e-nose in Manufacturing gives a different reading than the one in Quality Control. Who does the engineer believe then? Good luck finding an objective way to settle that argument.
One thing our brain does very well is separate signal from noise. We can, for example, follow a single conversation at a cocktail party full of chattering voices. Similarly a perfumer can work in an office reeking of background smells that change from day to day. But tracking a target against ever-changing background odors is hard for an e-nose. Even harder is following a moving target against such a background: a ripening peach in a farmer’s market, for example. Until it solves the cocktail-party problem, the e-nose will not be serious competition for the human nose.
AS TECHNOLOGY ADVANCES, the line between biology and hardware starts to blur. A group in Britain has developed what it calls “a truly biomimetic