Microsoft is updating its Responsible AI Standard and revealed that it’s retiring Azure Face’s emotional and facial recognition abilities (for the most part).
The Responsible AI Standard (opens in new tab) is Microsoft’s internal ruleset when it comes to building AI systems. The company wants AI to be a positive force in the world and to never be misused by bad actors. It’s a standard that’s never been shared with the public before. However, with this new change, Microsoft decided now would be the time.
Emotional and facial recognition software has been controversial, to say the least. There are multiple organizations calling for this technology to be banned. Fight for the Future, for example, wrote an open letter back in May asking Zoom to stop its own development of emotional tracking software and called it “invasive” and “ a violation of privacy and human rights.”
As it’s laid out, Microsoft will rework its Azure Face service to meet the requirements of its new Responsible AI Standard. First, the company is removing public access to the AI’s emotion scanning capability. Second, Azure Face will no longer be able to identify a person’s facial characteristics, including “gender, age, [a] smile, facial hair, hair, and makeup.”
The reason for the retirement is because the global science community still doesn’t have a clear “consensus on the definition of ‘emotions’”. Natasha Cramption, Chief Responsible AI Officer at Microsoft, said that experts from inside and outside the company have voiced their concerns. The problem is “the challenges in how inferences generalize across use cases, regions, and demographics, and the heightened privacy concerns…”
In addition to Azure Face, Microsoft’s Custom Neural Voice will be seeing similar restrictions. Custom Neural Voice (opens in new tab) is a text-to-speech app that is shockingly lifelike. Now the service will be limited to a select few “managed customers and partners,” which are people who work directly with Microsoft’s account teams. The company states that while the technology has great potential, it may be used to impersonate. In order to keep having access to Neural Voice, all existing customers must submit an intake form and get approved by Microsoft. They have to be approved by June 30, 2023, and if they aren’t selected, these customers will no longer have access to Neural Voice.
Still in the works
Despite everything that’s been said, Microsoft isn’t totally abandoning its facial recognition tech. The announcement only pertains to public access. Sarah Bird, who is the Principal Group Project Manager at Azure AI, wrote about responsible facial recognition (opens in new tab). And in that post, she states “Microsoft recognizes these capabilities can be valuable when used for a set of controlled accessibility scenarios.” One of these scenarios, according to a representative, is Seeing AI (opens in new tab)which is an iOS app that helps the visually impaired with identifying people and objects around them.
It’s good to see another tech giant recognizing the problems with facial recognition and the potential for abuse. IBM did something similar back in 2020, although its approach was more absolute.
Back in 2020, IBM announced it was abandoning work on facial recognition because the company was afraid it could be misused for mass surveillance. Seeing these two titans of the industry get rid of this tech is a win for anti-facial recognition critics. If you’re interested in learning more about AI, TechRadar recently published a piece on what it can do for cybersecurity.