Siri on iPhone screen

‘Siri, why are you a woman?’ EqualAI project challenges sexism in AI

Image credit: Dreamstime

A group of industry leaders in artificial intelligence (AI) have agreed to work together for a set of standards to prevent the perpetuation of gender inequality and stereotypes in AI.

According to Robert LoCascio, founder and CEO of LivePerson - a company that provides a chat-bot service and is partnered with Google, Facebook, Microsoft, IBM and other tech giants – AI brings a new danger of “unconscious human biases being programmed into the heart of the technologies that will shape our future”.

For instance, commercial facial recognition technology has been demonstrated to perform poorly when profiling dark-skinned women, while neural networks trained on online text tend to parrot existing gender stereotypes (e.g. men are doctors and women are nurses).

LoCascio and his colleagues argue that it is necessary to take proactive steps to reach equality in the AI sector. Already, French President Emmanuel Macron has committed to challenging bias in AI, although no formal commitments have been made by the US, UK or German governments. The EqualAI project aims to fill these gaps by laying out guidelines to prevent the perpetuation of sexism and other biases in AI. Today, a group of leaders in AI gathered at the Cannes Lion Festival of Creativity to discuss the creation of these guidelines, which are expected to be revealed later this year.

“The biggest tech companies tend to be very male-dominated, particularly for engineering groups,” said Rurik Bradbury, global head of conversational strategy at LivePerson.  “We know that what tends to happen is that people make software and algorithms that reflect them and their viewpoints and biases: conscious or unconscious. So what you get is shipped to everyone; it’s basically the reflection of a relatively narrow slice of the population, the white male techy types.”

For instance, the decision for Amazon’s voice assistant Alexa to be female was reportedly made by CEO Jeff Bezos due to his fondness for Star Trek and the female-voiced computer featured in the film and TV series. Bradbury describes this as an example of the “quick decision making” valued by many tech companies. This is not malicious, Bradbury says, but can lead to the perpetuation of unconscious biases.

The domination of the voice-assistant market by four female-voiced bots – Google’s Assistant, Microsoft’s Cortana, Amazon’s Alexa and Apple’s Siri – is one of several issues that could be targeted by the EqualAI project. It is argued that the across-the-board feminisation of robotic voice assistants perpetuates a stereotype of women as submissive - something LoCascio realised when he saw his young daughter treating Alexa as a real woman in a box to be ordered around, rather than as an object.

Professor Justine Cassell, a renowned expert in human-computer interaction, has suggested that it could be helpful to de-gender voice assistants and has been developing her own gender ambiguous synthesised voice which could remove gender stereotypes from this field entirely.

Meanwhile, LivePerson is aiming to make its clients think harder about the messages they are sending with their bots.

“We’ll see this bot explosion where every brand will have its own bot,” said Bradbury. “If we can make the companies aware of the issue, we can affect quite a lot of change because they’re all rolled out.”

“We had a client that was launching a bot for selling grills and they wanted to give it a man’s name like Tom the Grillbot. Then we talked to them and said: well, women grill as well, why perpetuate a trope when you could have something neutral like Grillmaster Bot? We’re trying to get a number of big brands - and a large number of brands after that - to adhere to these standards.”

Companies which adhere to these gender equality standards by giving sincere consideration to the messages they send when they choose the gender of their bots could earn an ‘EqualAI’ seal of approval.

While the EqualAI project will help push back against gender stereotyping in AI with a series of guidelines, the colossal issue of sexism in engineering and technology remains. Just 11 per cent of working engineers are women in the UK, women tend to be paid less in engineering and tech jobs – a pattern that persists across most sectors – and many women in the field have reported dismissive attitudes and harassment by their male colleagues.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles