Tackling gender prejudice in Artificial Intelligence: Are we on right track?

UNESCO’s 2019 publication “I’d Blush if I Could” revealed how much gender bias and stereotypes were engineered into Artificial Intelligence-powered voice-assistant applications. Beyond highlighting the overall gender imbalance of teams creating these new tools, it also showed evidence of alarming gender gaps in technology industries, even in countries that are close to achieving gender equality.

In this Q&A, Mark West, Project Officer at UNESCO and lead-author of the publication, shares insights on where we stand regarding gender prejudice in AI since the report came out.

Two years since the publication came out, where do we stand on the fight against gender prejudice in AI assistants?

On the positive side, awareness is much higher than it was when we researched and wrote our report. When AI voice assistants were first released, we were so enthralled with the novelty of speaking to computers that we forgot to ask critical questions. This honeymoon period is over. We now know-often from personal experience-that these systems are imperfect. We are on the lookout for ‘racism’, ‘classism’, ‘sexism’, ‘ageism’ and other ‘isms’ that get cooked into AI. This recognition and alertness means that problems are more likely to be flagged and corrected. On the negative side, transparency remains a problem. Even when people have a hunch that an AI system might be disadvantaging them for some unfair reason, it can be hard to prove. The recommendations we put forward about the importance of algorithmic audits remain relevant. Systems and AI engines need to be pulled out of black boxes so that people can study and test them. If a company claims to have an unbiased system, that’s great: prove it, show us what is under the hood, explain how it works and how it learns.

So, you are saying we need to be vigilant?

It’s important to remember that rooting out bias in AI will be a perpetual undertaking. So long as humans are at the helm of AI (as we very much are today), the decisions, opinions and recommendations offered by technology will be reflections of our own assumptions and worldviews or, more precisely, the worldviews of the relatively small groups of people-overwhelming men-that actually build AI systems. In this sense, there is nothing ‘artificial’ about AI. It will take conscious efforts and determination to keep rooting out prejudice. Getting more women in the field of AI is a key part of the solution. There are some encouraging signs that the field is slightly less male-dominated than it was when we wrote our report. However, there is still a long way to go. And backsliding remains a risk. As an example, in a number of countries a lower percentage of women are getting bachelor-level degrees in computer science today than in the late 1990s.

Are voice-assistant technologies today more gender-neutral than they were a few years ago?

Well, many of the most clichéd gendered responses have been dialed back. Siri no longer says, “I’d blush if I could” when users lob gender insults at ‘her’, so this is progress, even if the starting point was really low. Mainstream voice assistants are much more likely to shut down abusive speech than they were previously. After our report came out, a lot of technology teams looked over the scripts they had written for voice assistants and said “yeah, a lot of this plays to problematic gender stereotypes” and made changes. We’re glad they did-this was an aim of our work.

To what extent have the report’s recommendations been implemented?

Encouragingly, a there has been traction on several of our recommendations. For example, in the report we called on technology companies to “end the practice of making digital assistants female by default.” Recently, Apple announced it would do exactly this. Going forward, iOS users will select what voice they want Siri to use. Apple has, in effect, abandoned its former practice of assuming that people prefer a female voice to set the kitchen timer, make calendar appointments and read email messages. It is estimated that iOS is used on over 1 billion devices globally, so it’s a change that will be felt by a lot of people. We’ve seen progress in other areas as well. Governments are thinking more deeply about the voices, accents and gender projections of everything from chatbots that help with tax returns to systems that help people navigate public transportation. In the past most of the voices giving commands (things like “Exit the bus”) were male, while most of the voices offering assistance (“What can I do for you?”) were female. Since our report has come out, I’ve noticed a lot more of a mix: male and female voices being used both for commands and offers of help.

What still needs to change?

As communicated in our report, we continue to believe that projecting a human voice, gender and personality on non-human technology presents challenges. A way out of this conundrum is to project voice assistants and other AI applications as non-humans-a sort of ‘let’s keep AI artificial’ ethic. In our research we encountered lots of examples where AI helpers assumed a voice that is clear, distinct and pleasant, but still immediately recognizable as non-human and not overtly male or female. We saw examples of AI assistants that were projected as cartoons-talking animals, for example. I don’t mean to say there will never be a place or reason for projecting technology as a human person-there almost certainly will be. But if companies want to avoid tricky questions about gender, there is no ‘rule’ that AI assistants have to be ‘cast’ as young women or young men. Makers of AI assistants would do well to lean into the non-human identify of their creations, rather than trying to give a human veneer.

How about voice assistant Alexa, which still only has a female voice?

That’s correct. For certain subtasks it is possible to change Alexa’s voice but there is still no male option for general purpose Alexa functions. Recently, Amazon released “Communication Guidelines for Alexa” which gave rare insight into the level of thought that goes into these things-nothing accidental about it. The guidelines contained a section titled ‘Alexa and Gender’ which seemed like a direct response to our report. It’s full of contradictions. For example, the guidelines say Alexa ‘does not have a gender’ and should not be labelled with words like ‘she’ and ‘her’. But later in the same document Amazon refers to Alexa as ‘she’ and ‘her’ and says the technology has ‘female persona’. The guidelines reveal the problems of trying to give a machine a gendered human identity. If an AI technology speaks like a woman and has a ‘female persona’ people will understandably make associations between this technology and actual women. The first thing many voice assistants say when you call on them is “How can I help you?” Why should this question and subservient obedience always have a female voice? Amazon should let users choose whether or not they want a voice assistant with a male or a female voice or maybe even a non-gendered voice like C-3P0 from Star Wars.

What’s UNESCO doing?

Gender bias in AI and in other aspects of technology are, as we say in the report, about ‘bias in, bias out’. The teams working at the frontiers of technology are heavily male. If technology is to help communities and countries to become more gender-equal, we need women as well as men to steer the development of this technology that is, day-by-day, changing our world in profound ways. And for this, we need more girls and women to take up studies in computer science and other technology fields and join the teams shaping the technologies that will enter our homes, schools and workplaces in the future. In our report, we included lots of ideas and recommendations to help education systems accomplish this. We are working with countries to implement the actions we suggested.

UNESCO is currently developing a Recommendation on the Ethics of Artificial Intelligence (AI). This comprehensive global standard-setting instrument will be submitted to Member States for the adoption by the General Conference of UNESCO at its 41st session in November 2021. If adopted, the Recommendation will be the first global normative instrument in this critically important field.

/Public Release. This material comes from the originating organization and may be of a point-in-time nature, edited for clarity, style and length. View in full here.