The technology and science community has been swift to embrace the opportunities presented by International Women’s Day (IWD). This comes from the belief (based on solid evidence) that diversity is essential to improving STEM, that women working in the field should be well supported and that more women should be encouraged to take up careers in related fields.
I am completely behind these efforts and am delighted to see IWD being used as a regular reminder to all of us working in technology that we must not lose sight of the importance of improving diversity in our subject areas. As a woman working in the field of data science and artificial intelligence (AI), I feel this particularly acutely, for I believe it is now urgent that we improve diversity in AI.
In the U.K. the government has acknowledged the importance of AI and data science by making it one of the four pillars of its Industrial Strategy. It is not only a critical driver of economic growth, but the algorithms that underpin AI are becoming transformative and far-reaching. Algorithmic systems are increasingly used in ways that can directly impact our lives, called upon to make decisions about job offers, loans and even criminal sentencing. This could be a golden opportunity to improve fairness in decision-making, particularly where human beings don’t have the best track record because of conscious and unconscious bias. However, if we do not ensure those working in AI are more representative of the society in which they operate, this is unlikely to happen and we are in grave danger of further entrenching unfairness and, worse still, giving this unfairness a rubber stamp from the technology industry.
Evidence of Bias
A classic question often used to illustrate the moral dilemmas at work in AI is that of a hypothetical autonomous vehicle that is about to crash and cannot navigate a path that would save everyone. Should it swerve onto one teenager to spare its three elderly passengers? It was questions such as this which were posed by the Moral Machine experiment recently. Millions of users from 233 countries and territories took an online quiz, making a total of 40 million ethical decisions. Nine separate factors were tested, including individuals’ preferences for crashing into men versus women, sparing more lives or fewer, killing the young or the old, and even choosing between individuals perceived to be high or low status. One outcome became apparent very quickly – that although there was a universality to some decisions, many responses to moral dilemmas are dependent on culture. For example, participants from collectivist cultures like those found in East Asia are less likely to spare the young over the old and participants from individualistic cultures, like the U.K. and U.S., placed a stronger emphasis on sparing the maximum number of lives than their collectivist counterparts.
This raises important questions. How can we make data more human, if as a human race we cannot even agree on what it is to be human? How do we ensure algorithms work fairly when we cannot agree on what fairness looks like?
I, like many others, struggle to answer this but what I do know is that giving up is not an option and that we need to make a concerted effort as an industry to make our technology work as well as it can for as many people as possible.
There is an urgent need to ensure that AI systems do not discriminate inappropriately against any individual or group. If only certain groups of people build the technology then it is highly likely that discrimination will happen, even if the discrimination is not intentional.
The number of “white men of a certain age” building technology has often been cited as the reason why bias has reared its ugly head in AI algorithms. This is of particular concern when machine learning methods are used to train systems on past human decisions as they will often reflect historic prejudice and unconscious bias.
Industry diversity today is poor – only around 15 per cent of women work in data science in Europe and this really does matter. In her new book, Invisible Women: Exposing Data Bias in a World Designed for Men, the journalist and activist Caroline Criado Perez makes a compelling case that data is actually far from objective and is often highly male biased. As a result of this bias, decisions around public spending, health, education, the workplace and society in general suffer.
How can we bring about change?
I wish there was a magic bullet that would fix all this but I know that change only happens as a result of a concerted, joined up programme of activity focused on making this change happen. There are, however, a number of common sense initiatives that we should be putting into practice as a matter of urgency to begin to solve these problems
Making sure female data scientists are much more visible is one of the first things we can do to start to fix this. Role models are vitally important in showing girls and women that they too can work in data science and that this career choice is an option for them. This is part of the wider work we need to do in getting young girls excited about coding and technology at an early age and encouraging them to study STEM subjects at A-level and at University.
The technology industry can certainly play its part in supporting female STEM graduates to pursue careers in technology and in doing more to ensure women do not leave these careers prematurely with initiatives focused on child care, work-life balance and more funding available for female entrepreneurs.
At Pivigo we are passionate about diversity and today, on IWD, we pledge to include a minimum of 40% female participants on our data science training programme (with a target of 50%). We can and should all be doing what we can to make positive change if we’re to improve the situation. If everyone reading this were to encourage a young girl they know to study STEM subjects; perhaps by buying them a science toy, encouraging them to read a science book, or by telling them about the trailblazing women who have worked and continue to work in our industry, it would open up our world to more girls in future.
The problem is huge and the stakes are high but the rewards we will reap if we fix this are considerable. On International Women’s Day, let’s all commit to making change happen and spreading the word – one woman at a time.
This article originally appeared in Forbes