Better Bacon through Technology
Because I've spent years working at the Geophysical Institute, many things remind me of people I've met here. Some are reasonably geophysical: for example, the sight of a bumper sticker reading "Stop Continental Drift" leads me to think of David Stone, whose work in paleomagnetism helped delineate where the continents drifted. Others aren't so obvious. Every time I see mention of global-positioning systems, I think of Paul Perreault. Perreault worked on auroras when he was a graduate student at the institute, but more recently he's been active in developing GPS technology. And when I see something about pigs, Henry Cole naturally comes to mind.
I'd better explain. Geophysicist Cole disengaged from academic research (though he does teach at the University of Alaska Fairbanks) to become a public servant. Former science advisor to Governor Steve Cowper, he is now working in the Division of Economic Development. There, among other things, he's coping with the problems and possibilities inherent in the Danish proposal to bring industrial-strength pig farming to Alaska.(And I bet you thought a Ph.D. in geophysics wasn't good for anything practical.) So it's really not such an insult to Cole that the picture of a Danish pig in a recent issue of the magazine New Scientist made me think of him.
The article accompanying the porker's portrait was a quick reminder that Denmark not only takes its pig production seriously, it's also a country swift to apply splendid Scandinavian---for that matter, international---advanced technology to matters of national interest. The Danish Meat Research Institute in Roskilde has just reported the results of a two-year test in which a very advanced kind of computer cooperated successfully with a specialized robot. The result: more efficient and economical meat grading.
Hog carcasses are graded according to the percentage of lean meat they contain. Accurate grading is important not only for purposes of establishing the price paid to the farmer and by the purchaser but also because the grade affects the most effective way to cut the meat. In most places, meat grading is the responsibility of trained human eyes and brains. The ability to grade meat accurately stems from much experience and at least some talent; there's an art to it. In Denmark, the process is less artistic. Since 1989, the job of grading Danish pork has belonged to robots.
The robots operated by inserting 17 fiber-optic probes into the carcass and measuring the amount of light reflected off the flesh. Fat reflects more light than lean meat does, so the data collected by the optical probes easily can be converted by a computer into a gauge of the meat percentage in the carcass.
It was an easy task for the computers, but not necessarily one that they performed accurately. The calculations were done by an expert system, one that made a series of decisions according to rules derived from the procedures of human experts in meat grading. But the more the system was updated, the more mistakes it made. The Danes turned to another kind of computer, a so-called neural network---a computer that learns. This high-tech meat grader was given data from 2000 carcasses and the correct grade for each. The computer deduced how to weigh the importance of each bit of data from the probes. It could then apply what it had learned to new data.
The Danes report that the neural-network computer learned so well that the automated meat-grading system has now achieved 98 percent accuracy. Furthermore, the computer taught its operators something: they only need 9 probes, not 17, to measure the reflectance properly. The Danish Meat Research Institute scientists estimate that the new smarter system has already provided considerable savings in maintenance costs, and could boost their industry's sales by the equivalent of several million dollars a year. I wonder if that'll be enough to support an Alaska pig-farming venture?
