Researchers at Google are out to wow us again—this time by developing a machine-learning system that can analyze images and automatically write captions for them.
The technology, using artificial neural network software based on actual neural networks in the human brain, is fascinating. A computer that can teach itself to accurately interpret images has an untold number of applications—it can assist visually impaired people and make searching the web for pictures and videos a whole lot easier for everybody. Other uses are potentially alarming. Similar technology could be used by surveillance cameras to identify people via facial recognition.
Whatever its potential applications, the technology is far from perfect. Google revealed in a research study that it made a number of errors that threw off some identifications—some minor, others more obvious. Still, even the more egregious errors are charming, and they might even make you laugh:
Researchers at Stanford University also developed image-captioning technology, concurrent with Google’s. The technology is bound to improve as tweaks are made and more tests are performed, but for now it thinks a road sign is a refrigerator, and we’re okay with that.