Top Artificial Intelligence Fails in Image and Facial Recognition
The field of AI is rapidly advancing, and pretty soon, we will get to the point where we no longer even have to search for something to find it. We will simply be able to point our smartphone cameras at it, and the AI algorithms will take care of the rest. Even though a lot of companies have been at the forefront of adopting this technology into their service offering, for the most part, it is still being used to extract information from a given image.
In order to train the AI algorithms to identify objects or people in an image, researchers input lots of annotated data into the system so it can learn to recognize whatever is needed. The image data can be virtually in any form, such as video, views from many cameras, multi-dimensional data, and many other types. However, this technology is not yet infallible and is prone to making mistakes. Let’s take a look at some image recognition fails.
Flickr is a top-rated image hosting service where users upload thousands of images every day. In order to sort through all of the images, the company decided to create an auto-recognition tool that soon went haywire. It started to identify some people as monkeys and other animals and soon had to be shut down.
2. Did Someone Blink?
We all have pictures where ended up blinking by accident, and technology companies tried to help us avoid this situation in the future by using facial recognition to warn us that we might have blinked. However, users in Asian countries have reported some interesting situations with this technology because it did not adjust to the contour of their faces and started asking if they blinked almost every time.
3. Apple Face ID
When Apple unveiled its Face ID technology along with the release of iPhone X, it was a very hyped up technological development. While it replaced Touch ID as the unlock method for the newest generation of devices, a lot of users started reporting an iPhone facial recognition to fail. In case you are wondering how Face ID works. It is supposed to account for the slight changes in people’s appearance when they put on makeup, grow a mustache, or even when they wake up in the morning. The very first time Face ID failed was during its own presentation when it didn’t recognize Craig Federighi. While there have been many strides made in the field of AI, the development of computer vision is something that will still require homing in the future.
There are thousands of various species of animals out there, and AI has been trained to assist researchers and ordinary people with identifying them. However, an interesting data set called Imagenet-A shows how much AI still has to learn. It consists of 7,500 images that were carefully selected, but not manipulated and put in front of image recognition cameras. Not only were the animals improperly labeled, but they were confused with objects such as washing machines and snowmobiles.
5. Law Enforcement
Police and other law enforcement agencies rely on facial recognition to identify suspects among a crowd of people. One such technology, called Rekognition and developed by Amazon, openly declared that their newest invention could be used for law enforcement purposes. However, the American Civil Liberties Union (ACLU) decided to put this technology to the test. They used Rekognition to identify Congress members, but instead, it matched a lot of them with publicly available mug shots of criminals. Amazon actively markets its product to law enforcement agencies, and such a Face ID fail shows that this technology requires further development.
6. Chihuahua or Muffin?
While most people would not have any problems distinguishing a chihuahua from a muffin, image recognition still does not see a difference. In fact, it confused a raisin muffin with a chihuahua because the raisin was placed in such a way that it resembled eyes on a dog.
While there are a lot of fun computer vision projects, there are a lot of serious ones as well. When computer vision is used for law enforcement purposes and other matters of great importance, it is absolutely necessary that they are right all the time. As we saw from some of the humorous and severe examples above, image and facial recognition technology still have a long way to go to be considered foolproof.
However, it is essential to point out that technology is continually evolving, all of the bugs and other issues are gradually starting to get fixed, and image recognition products are gradually starting to become more reliable. We at Skywell Software always rigorously test our projects during the development stages to make sure that it is functioning as intended. Artificial intelligence is a very complex field requiring a lot of industry and technological knowledge. Therefore you need to trust your product with the best specialists possible.