Amazon's face recognition falsely identified 28 members of Congress as criminals. Perhaps AI isn't quite ready for business.
Amazon's face recognition AI incorrectly matched 28 members of Congress with criminal mugshots, according to an ACLU blog post on Thursday. Using Amazon's Rekognition software--the exact same facial recognition system offered to the public--the ACLU conducted the experiment with a database of 25,000 publicly-available mugshots and photos of every member of Congress, said the post.
Controversy surrounding the use of Amazon Rekognition isn't new, particularly since it's been tailored to be used by law enforcement. A report from Gizmodo displays previous faults in the facial recognition technology, citing a training presentation that matches a picture of O.J. Simpson to a white male with long hair and a mustache.
SEE: Artificial intelligence: Trends, obstacles, and potential wins (Tech Pro Research)
A disproportionate number of mismatched images were people of color, which further supports the ACLU's reasoning for pushing back on law enforcement's use of face recognition technology, the post said. If law enforcement is implementing Amazon Rekognition, false identification could provide false bias to an officer before a situation even starts, explained the post.
The ACLU emphasized how their test method wasn't a "hypothetical exercise," as Amazon is marketing their facial recognition software primarily to police. In fact, the ACLU noted that a police station in Oregon started using Amazon Rekognition to scan people's faces against a mugshot database.
The implementation of face surveillance AI is real, and so are the problems and risks associated with it. Amazon Rekognition's flaws reveal how AI may not be ready for its primetime in business, with still many kinks to work out.
Discussions surrounding face recognition technology in business have been increasingly popular. From talk of applications in healthcare, retail, security, marketing, banking, and more, facial recognition software could have a presence in almost any market. However, the ACLU's analysis proves that the AI implementation may not be quite ready for the enterprise.
The big takeaways for tech leaders:
- Amazon's face recognition software misidentified 28 members of Congress, matching them to publicly available criminal mugshots. -- ACLU, 2018
- The flaws in Amazon Rekognition's programming reveal how this type of AI technology may not be accurate enough for businesses to use. -- ACLU, 2018
- Artificial intelligence: A business leader's guide (free PDF) (TechRepublic)
- Now Amazon employees rebel: 'End police facial-recognition contracts, ICE support' (ZDNet)
- Amazon AI: Cheat sheet (TechRepublic)
- AWS demonstrates Amazon Rekognition Video in real time (ZDNet)
- Why business leaders should be concerned about AI-powered face swapping technology (TechRepublic)