Another woman has moved into my home. Or at least, my partner thinks so. For him, Alexa is “her.” But I dispute that. For me, Alexa is most definitely “it.”
Studies have repeatedly shown that both women and men prefer digital assistants to have female voices because they believe them to be more welcoming and understanding – which is why Alexa and friends are made with dulcet feminine tones. But isn’t that just a historic, old-fashioned stereotype? Do service bots really need to be ‘female’ to be relatable? If AI is going to run our lives, perhaps we need to challenge some of these preconceptions and come up with fresher perspectives.
AI and machine learning infiltrates almost everything we do: it helps organise our diaries, gives us the latest headlines, sorts out pizza for tea and soundtracks our evening with our favourite tunes. More than that, it determines our behaviours, thought processes, buying patterns, and even our worldviews – it’s like Facebook’s filter bubble but on a much grander scale. Increasingly in the future, AI will shape our perceptions of our world, and not just influence the choice of what we eat, buy or what we listen to.
Which means the hands and minds making the technology have a direct impact on us as humans, and on the world around us. Therefore it is vital that we include more female perspectives in this new world, and we need more women of diverse backgrounds to have a hand in creating it – and not just give virtual assistants female voices.
AI is like a child. How it grows is down to how we nurture it, and unless we design these systems from the start with inclusion in mind we will create systems that reflect the multiple biases and stereotypes that have damaged and limited our world today. AI becomes biased through the data that is used to train it and it’s hard for any of us, men and women, to be aware of our biases.
Bias creeps in when your data sets aren’t inclusive enough and AI then learns from our own prejudices. We’ve seen Facebook algorithms influence how we view world events, by skewing what’s on our newsfeed. We’ve seen how Twitter taught Microsoft’s AI chatbot Tay to become a misogynistic, anti-Semitic racist in less than a day. If we don’t challenge bias now, like a supertanker, it’ll be very hard to turn around.
Inclusion has to be the aim for the future of AI, but there’s a potential paradox in building diversity while being tolerant of all points of view. Who decides which points of view we want to take forward into this brave new world and which to leave behind? How do we create AI which is in tune with human minds, but ignores the worst elements of human prejudices? Where do we find neutral data sets – do they even exist? Should AI to be programmed around an aspirational image of the world we want or a realistic version of the world today? Data can take time to catch up with culture. How do we create AI that can be more sensitive to this?
It’s a fine line to tread but ultimately, we need to build trust with AI and the people creating it and trust they are creating unbiased technology.