This week IBM, Microsoft and Amazon announced that they would suspend the sale of their facial recognition technology to law enforcement agencies. It’s yet another sign of the dramatic impact of the protests for social justice.
But the moves from the tech giants also illustrate the inherent risks of AI, especially when it comes to bias and the potential for invasion of privacy. Note that there are already indications that Congress will take action to regulate the technology. In the meantime, many cities have already instituted bans, such San Francisco.
Because of the advances of deep learning and faster systems for processing enormous amounts of data, facial recognition has certainly seen major strides over the past decade. Yet there is still much to be done.
“AI face recognition technology is damn good, but it is not very robust,” said Ken Bodnar, who is an AI researcher. “This means that the neural network is well trained and capable of amazing feats of identification, but if one little parameter is off, it mis-identifies you. The way that it works, is that everything is a probability with AI. So when it looks at a face, it has a range of proprietary algorithms and parameters it measures. The most accurate AI tools are Deep Belief Networks that winnow out features like double chins, eye distance, hair type, bushy eyebrows, fat lips, age parameters etc. But the ‘not-very-robust’ categorization means that it is easy to fool because of the intrinsic nature of the way that neural networks work.”
True, when it comes to certain applications of facial recognition, the accuracy issues may not necessarily be a big deal (such as with a cool social media app). But of course, this is a different matter when its about whether someone should be arrested.
Facial recognition has also been shown to be less effective when analyzing videos and images of minorities. “As for the issues with this technology, a study out of MIT last year found that all of the facial recognition tools had major issues when identifying people of color,” said Michal Strahilevitz, who is a professor of marketing at St. Mary’s College of California. “Another study out of the US National Institute of Standards and Technology suggested facial recognition software had far more errors in attempting to recognize black and Asian faces than it had in recognizing Caucasian ones. This means that black and brown people are more likely to be inaccurately identified, and thus unfairly targeted. This may not be intentional, but it ends up having a racial bias that is dangerous and unethical.”
Yet the debate over facial recognition can certainly get complicated and may even lead to unintended consequences.
“The moves reflect a lack of popular understanding of the technology–the public is conflating facial recognition with body recognition and tracking, facial analysis, facial detection, gender/age/ethnicity recognition, biometric validation, etc. as well as misunderstanding the difference between the use case and the technology,” said Kjell Carlsson, who is an analyst at Forrester. “It is very unclear exactly what is being renounced and for what use cases. The result is almost certainly the worst of both worlds: Ineffectual policy to prevent misuse together with applying the breaks on valuable use cases. For example, wouldn’t we want folks to use facial recognition to help identify kidnapping victims in Nigeria? Or to help diagnose rare genetic disorders in children with facial analysis for genetic phenotyping? Or to help catch the Boston Marathon bombers? Or to reduce the false positives in the techniques that the police use to match faces to mugshots that they have been using for decades?”
It’s interesting to note that facial recognition technology will likely continue to see more innovation and development. regardless of the bans and regulatory actions. Just take a look at GitHub and you’ll find sophisticated systems…for free.
So if anything, when it comes to companies like IBM, Microsoft and Amazon, it’s important for them to do more than ban the technology. “I do hope that they come back with a better-designed system that respects the user’s privacy and allows for usage in times of need and with conscious protection for users and their data,” said Lovelesh Chhabra, who is the Vice President of Membership Platforms at Verizon Media.
This should then be backed up with a strong set of principles and ethical standards that the industry can follow. The result will likely be a stronger foundation for AI.
“The good news is that bias-free identification at all levels of a device is possible, but every player involved–for apps, hardware and OS providers—has to be willing to commit to it,” said André Ferraz, who is the founder and CEO of Incognia.
Tom (@ttaulli) is an advisor to startups and the author of Artificial Intelligence Basics: A Non-Technical Introduction and The Robotic Process Automation Handbook: A Guide to Implementing RPA Systems. He also has developed various online courses, such as for the Python programming language.