Even the Most Sophisticated Computers Can’t Tell a Dog From a Cat
The Independent (London) (01/06/09) Bishop, Chris

Chris Bishop, chief research scientist for Microsoft Research in Cambridge, England, says there is still much that needs to be explored about digital intelligence. Bishop says that it is still very difficult for computers to recognize structures or objects in data.

Scientists tried to solve the pattern recognition problem by using artificial intelligence in the 1970s, but the fact that there are always exceptions to handcrafted rules ultimately led them to abandon the expert systems idea. Researchers are now focusing on having computers learn from experience similar to humans, by programming them to learn from data and then training them to solve the problem.

A supercomputer is no match for even a toddler when it comes to recognizing variations in size, color, shapes, lighting, and background. Still, researchers have made some advances, and the resulting practical applications now enable robots in factories to see what they are assembling, and allow tumors to be detected in medical images. Bishop says he is looking forward to developments in the years to come.

———-

I love that after all this time, a toddler can still outdo a computer with his/her recognition skills.