AnyVision open letter calls on facial recognition providers to eliminate bias

Companies developing biometrics and artificial intelligence algorithms must rid their systems of demographic bias and make their methodology transparent, according to an open letter published by AnyVision in response to the U.S. National Institute of Standards and Technology’s (NIST’s) call for public comment on its proposed method for evaluating user trust in AI systems.

NIST developed a document listing factors in the trust of AI systems and called for comments as part of an effort to stimulate debate on the topic.

AnyVision’s letter, signed by CEO Avi Golan and titled ‘Purging Demographic Bias while Increasing Transparency in Facial Recognition,’ emphasizes the importance of understanding the use case to evaluating how much trust users should place in systems. Golan describes the EU’s recent moves towards AI regulation, which categorize remote biometric identification as ‘high-risk,’ as an attempt to provide clarity around use cases.

“These are steps in the right direction in understanding and categorizing AI because it’s understood that AI is providing significant benefits including improved speed, accuracy, cost savings, fraud detection, medical diagnoses and customer experience,” Golan explains. “At the same time, it’s vital to address its historical weaknesses. Consequently, AI companies must continue to purge demographic bias from their algorithms and be transparent about their methodology and the training data used to develop their models. Unfortunately, this level of nuance is missing from most discussions today related to facial recognition.”

“That’s why we’re calling upon NIST to help define and shape the discussion around the responsible use of facial recognition and video surveillance by drafting similar guidelines – best practices that can help organizations, of all stripes, deploy the technology – safely, securely, and ethically,” he adds.

AnyVision conducted the Fair Facial Recognition Challenge as part of the European Conference on Computer Vision 2020, and says the results of the top 10 teams demonstrated that racial bias can be reduced or even eliminated with an appropriately wide range of training data, while still providing a high degree of biometric accuracy.

Golan also offer’s AnyVision’s insights gained from its experience working with global players and agencies from around the world, and expressed the company’s desire to work with NIST and leading academics and NGOs to draft guidelines and best practices for bias reduction.

AnyVision ranked among the leaders across all five category groups in NIST’s biometric verification accuracy benchmarking in March.

Article Topics

accuracy  |  AI  |  algorithms  |  Anyvision  |  biometric bias  |  biometric dataset  |  biometric identification  |  biometrics  |  biometrics research  |  facial recognition  |  NIST  |  research and development  |  training