We need to address the coded bias in Machine Learning algorithms
We try to write a lot about the positive ways in which Artificial Intelligence and other technologies are impacting our world. It’s a subject close to our hearts as a company. Quite frankly, we think it’s something worth celebrating. Given all the doom and gloom we’re so often bombarded within today’s media!
From healthcare to sustainable cities, climate change and industry, investment in AI is making an impact in many areas. Applications of machine learning and deep learning help shape the trajectories of our daily lives. So much so, we are barely even aware of it.
However, all this do-gooding aside, one of the biggest obstacles in AI programming is the inherent bias that exists within it. This progressive technology at times leads us right back to the uneducated, discriminatory ideologies that are better left in the past. A place where race and criminality go hand in hand, where we associate women with kitchens and cooking, and men with the boardroom.
There are some cringe-worthy examples out there, such as the time when Nikon’s Coolpix kept asking an Asian family ‘Did someone blink?’ These messages stopped once they took a picture with a relative holding their eyes open wide… Yes, that happened.
How about racial profiling in risk assessment modelling for criminal sentencing? ProPublica unearthed some very concerning details about America’s justice system. Their racially biased AI predicts a greater likelihood of reoffending for ethnic minorities.
And yet, it’s not just limited to race. In the UK, a doctor was locked out of the women’s changing rooms at a Pure Gym. This was due to the gym’s computer system recognising her ‘doctor’ job title as male-only. Or how about Googling CEO to find only pictures of white men – because, minorities or women cannot be CEO’s, right?
For a long time, algorithms have been imbued with racial, gender, and other stereotypical biases. What is extremely troubling is that this technology is present in so many aspects of our daily lives.
The brilliant Joy Buolamwini dubs algorithmic bias as ‘the coded gaze’ in her 2016 TedTalk – if you haven’t seen this, you need to! Joy details how her journey towards working on tackling racial discrimination in AI began. Unsurprisingly, a computer system only recognised her face when wearing a white mask.
Joy’s experience highlights the fundamental problems associated with training limited data sets. When your data is limited it does not create a diverse representation of society. In turn, you omit certain societal groups from this technology.
“Success in creating AI, could be the biggest event in the history of our civilization… But it could also be the last unless we learn how to avoid the risks. Alongside the benefits, AI will also bring dangers, like powerful autonomous weapons, or new ways for the few to oppress the many.”
Stephen Hawking for Wired magazine
Oppression through algorithmic bias is a very real threat, and we need to combat this. Providing platforms for the voices of society with less power than those of the predominantly white, male Silicon Valley circles.
There is light at the end of the tunnel with many bright lights in the field of AI, like Joy, working to tackle these problems.
The non-profit American Civil Liberties Union (ACLU) has partnered with research institute AI Now, to tackle AI bias. They consider, how do we program AI? Answering this question is crucial to their work in a bid to accomplish algorithms free from bias. The ACLU protects the rights of individuals in areas most affected by AI discrimination. Including but not limited to, housing, credit and lending, prosecution and the criminal justice system.
ProPublica’s shocking revelations of how unfair machine learning bias can be in criminal sentencing highlights just how desperately we need regulation and investigations. Hopefully, the partnership between ACLU and AI Now will go some way in tackling these issues.
Similarly, the movement ‘Data For Black Lives’ is working to address other issues of inequality and discrimination. Areas such as financial and credit services, predictive policing and risk assessment sentencing. They aim to use “data science to create concrete and measurable change in the lives of Black people.”
Microsoft’s research group FATE (Fairness, Accountability, Transparency, and Ethics in AI), are working on the social implications of AI to produce ethical algorithms. Part of this team is researcher Timnit Gebru, who co-founded the ‘Black in AI’ event. Popular culture is also addressing these issues, using its platform to inspire a new generation of brilliant minds. The recent box office hit Black Panther hailing from the Marvel Universe shows Princess Shuri, a young black woman as central to the development of high technology for the most advanced nation on earth, Wakanda. This portrayal (by the very talented Letitia Wright, may I add), is inspiring notably for women of colour, to enter into STEM fields.
To Summarise
The advancement of AI and its ability to do good in the world goes hand in hand with its perception and profiling of people from different walks of life. We must avoid building discriminatory AI systems. If AI continues to oppress certain communities within our society, we will never achieve social good.
