IBM whistles at Flickr image rights for AI tests
IT giant IBM used nearly one million photos from the photo-sharing community Flickr to train its facial recognition software, but overlooked one important detail: the consent of the people appearing in the photos.
According to US media, the technology group has even passed on the data material in question to other companies and research institutions to help them develop their own programs. This has drawn criticism from data protectionists and photographers.
"That's the dirty little secret of datasets like this that are used to train artificial intelligence. A lot of scientists just grab all the photos they can get their hands on for that," NBC News quotes legal expert Jason Schultz of New York University as saying. Hundreds of thousands of images would be needed to develop a well-functioning facial recognition system. "Those photos all come from the Internet and are used without the consent of the people who put them there and sorted by categories like age, gender, skin color and a dozen other factors, and then shared with university and corporate researchers," Schultz explains.
The lack of information for those affected is particularly incomprehensible to photographers. They have to be very careful when taking and using their pictures and clarify the relevant rights in writing with the people photographed. "None of the people I photographed had any idea that their pictures were being used in this way," criticizes Greg Peverill-Conti, a PR consultant from Boston who contributed over 700 photos to the IBM dataset. The fact that these can simply be "misused" in this way, he says, came as a great surprise to him.
"We take the privacy of individuals very seriously and make great efforts to always act in accordance with the applicable data protection regulations," says an initial response from IBM. This also applies to the specific data set with Flickr photos that is the subject of the current discussion. According to the IT giant, this can only be viewed by verified scientists. "Furthermore, it only contains material that is freely accessible to the public. If someone does not want to be included in this data set, they also have the option of having themselves removed from it," the company emphasizes. (pte)
