For years, activists and scientists have feared that face analysis software that claims to identify a person’s age, gender and emotional state may be biased, unreliable or invasive – and should not be sold.
Recognizing some of these criticisms, Microsoft said Tuesday it plans to remove those features from its artificial intelligence service for face detection, analysis and recognition. They will cease to be available to new users this week and will be phased out for existing users within the year.
The changes are part of Microsoft’s insistence on tighter controls on its artificial intelligence products. After a two-year review, a Microsoft team developed the Responsible AI Standard, a 27-page document that sets out the requirements for artificial intelligence systems to ensure that they do not have a detrimental effect on society.
Requirements include ensuring that the systems provide “valid solutions to the problems they are designed to solve” and “similar quality of service for identified demographic groups, including marginalized groups”.
Prior to launch, technologies that would be used to make important decisions about a person’s access to employment, education, health, financial services or life opportunities are subject to review by a team led by Natasha Crampton, AI’s Chief Executive Officer. Microsoft. .
There were growing concerns at Microsoft about the emotion recognition tool, which referred to someone’s expression as anger, contempt, disgust, fear, happiness, neutrality, sadness, or surprise.
“There is a huge amount of cultural, geographical and individual variation in the way we express ourselves,” Ms. Crampton said. This has raised concerns about reliability, along with larger questions about whether “facial expressions are a reliable indicator of your inner emotional state,” she said.
Read more about artificial intelligence
The age and gender analysis tools that are being eliminated – along with other tools to detect facial attributes such as hair and smiles – could be useful in interpreting visual images for the blind or visually impaired, for example, but the company decided it was problematic to make profiling tools publicly available, Ms Crampton said.
In particular, she added, the so-called gender classifier of the system is binary, “and this is not in line with our values.”
Microsoft will also add new controls to its face recognition feature, which can be used to verify identity or search for a specific person. Uber, for example, uses the software in its application to verify that the driver’s face matches the identification number in that driver’s account file. Software developers who want to use Microsoft’s face recognition tool will need to apply for access and explain how they plan to implement it.
Users will also need to apply and explain how they will use other potentially abusive AI systems, such as Custom Neural Voice. The service can generate an imprint of a human voice based on an excerpt from someone’s speech, so that authors can, for example, create synthetic versions of their voice to read their audiobooks in non-speaking languages.
Because of the possible misuse of the instrument – to give the impression that people have said things they did not – the speakers have to go through a series of steps to confirm that their voice is allowed and the recordings include watermarks found. from Microsoft.
“We are taking concrete steps to meet our AI principles,” said Ms. Crampton, who has worked as an attorney at Microsoft for 11 years and joined the Ethics AI group in 2018. It will be a huge journey. ”
Microsoft, like other technology companies, had obstacles with its artificially intelligent products. In 2016, he launched a chatbot on Twitter called Tay, which is designed to teach “conversational understanding” from the users he interacted with. The bot quickly started publishing racist and offensive tweets, and Microsoft had to remove it.
In 2020, researchers found that text-to-speech tools developed by Microsoft, Apple, Google, IBM and Amazon worked less for blacks. Microsoft’s system was the best in the group, but misidentified 15 percent of words for whites, compared to 27 percent for blacks.
The company had gathered a variety of speech data to train its AI system, but did not realize how diverse the language could be. So he hired an expert in sociolinguistics from the University of Washington to explain the language variations that Microsoft needs to know. It went beyond demographics and regional diversity in the way people speak in formal and informal settings.
“Thinking of race as a determining factor in how someone speaks is actually a bit misleading,” Ms. Crampton said. “What we learned after consulting the expert is that, in fact, a huge range of factors affect linguistic diversity.”
Ms. Crampton said the trip to correct this discrepancy between speech and text has helped to inform the guidelines set out in the company’s new standards.
“This is a critical period for setting AI standards,” she said, citing European-proposed regulations that set rules and restrictions on the use of artificial intelligence. “We hope to be able to use our standard to try to contribute to the bright, necessary discussion that needs to be held about the standards that technology companies need to adhere to.”
A lively debate about the potential harms of AI has been going on for years in the technology community, fueled by mistakes and errors that have real consequences for people’s lives, such as algorithms that determine whether people receive social benefits or not. The Dutch tax authorities mistakenly took away childcare allowances from needy families when the wrong algorithm punished people with dual citizenship.
Automated face recognition and analysis software is particularly controversial. Last year, Facebook shut down its decade-long system for identifying people in photos. The vice president of the artificial intelligence company cited “many concerns about the place of face recognition technology in society.”
Several black men were wrongfully arrested after improper face recognition matches. And in 2020, at the same time as the Black Lives Matter protests following the police assassination of George Floyd in Minneapolis, Amazon and Microsoft issued moratoriums on the use of their face recognition products by police in the United States, saying clearer laws were needed. use.
Since then, Washington and Massachusetts have passed a regulation requiring, among other things, judicial oversight of police use of facial recognition tools.
Ms Crampton said Microsoft was considering giving its software to state police with book laws, but has decided not to do so for now. She said that could change with the change in the legal environment.
Arvind Narayanan, a professor of computer science at Princeton and a renowned AI expert, said companies may shy away from technologies that analyze the face because they are “more visceral than various other types of AI that may be questionable.” But we don’t necessarily feel it in our bones. ”
Companies may also realize that at least for now, some of these systems are not as commercially valuable, he said. Microsoft could not say how many users it had for the facial analysis features it was getting rid of. Mr Narayanan predicts that companies are less likely to abandon other invasive technologies, such as targeted advertising that profiles people to choose the best ads to show them because they are “dairy cows”.
Add Comment