Here Nikolas Kairinos, Founder and CEO of Soffos, delves into the difficult subject of diversity standards in the AI space
Back in June this year, a pixelated image of Barack Obama upsampled to the image of a white man had the internet up in arms. Created through artificial intelligence (AI), the emergence of the photograph on Twitter prompted a difficult discussion about racial bias in AI and machine learning (ML). And it wasn’t the first time.
From Uber’s failure to recognise trans drivers, sentencing algorithms that discriminate against black defendants, and chatbots that adopt racist and misogynistic language, there is an ongoing equality issue in the AI industry: one that has been unfolding for many years. The truth is that problems with AI are often not the fault of technical errors or mistakes in the creation of products. They reflect and replicate systems of inequality in society. Namely, a lack of diversity amongst the creators, developers and researchers within the sector.
Given the increasing use of AI in our society, this poses important questions which must be urgently answered. With this in mind, how can AI businesses ensure that they are doing all they can to remedy the ongoing diversity crisis?
Considering the journey so far
Firstly, businesses must consider how events have unfolded prior to now.
The dialogue about inclusion in the AI sector took off around 2014, when big names in the industry came under fire for their lack of transparency. Driven to quick action by industry professionals like software engineer Tracy Chou and investor Ellen Pao, organizations such as Apple, Google and Facebook to name but a few began publishing their diversity reports.
As anticipated, the reports did not inspire confidence and showed poor representation of female workers, as well as employees from minority and disadvantaged backgrounds. To remedy the imbalances, these companies started to hire diversity inclusion leads, as well as overhauling their hiring practices and launching inclusivity initiatives.
Progress indeed, but unfortunately these efforts did little to quell the trouble at hand. Even now, over half a decade on, the outlook remains uninspiring. Although great efforts have been made to foster inclusivity, a recent study from the AI Now Institute of New York University found that more than 80% of AI professors are men. The report also found that just 15% of AI researchers at Facebook and 10% at Google are women. Meanwhile, just 2.5% of Google’s workforce is black.
Looking beyond technical aspects of development
One consideration to keep in mind is that AI is not implicitly biased – it is only as good as its data. From a technical standpoint, then, as an industry we must continue to create and utilize detection tools for identifying and mitigating bias in given datasets. This will prevent AI programs from propagating any existing inequalities, and building on unrepresentative data.
But clearly, greater action is needed so that workforces are created with equality in mind, to guarantee that these biases can be quickly spotted and corrected. To ensure that progress with diversity matches the speed of advancement in AI technologies, businesses and Governments now must consider what practical changes they can make to change the state of affairs.
Unfortunately, there is no quick fix. True progress will entail better funding and outreach from Governments, in order to encourage those from underrepresented backgrounds into STEM education and training, as well as further improvements to hiring practices to maximize diversity.
AI businesses can nurture their teams by promoting their hard-working members of staff from minority backgrounds into more senior positions. After all, it is only by ensuring that the AI pioneers of the future see themselves represented by those in positions of power that real change can be made. People from all walks of life must have strong role models if we are all to benefit from AI innovations equally.
Above all else, transparency is vital, and business leaders in the AI sector should do all they can to promote an environment of trust amongst their workforce. Employees should feel that they are always at a liberty to speak freely if they believe a product or practice does not reflect appropriate diversity standards. If the practice is adopted correctly, this will ensure that products do not have any underlying biases, and that technology is built to serve the masses – not just a select few in positions of privilege.
Ultimately, it is up to business leaders to be honest with themselves about where their company currently stands when it comes to progress made with diversity and inclusion initiatives. To improve fairness and equality, and ensure that everybody benefits from the many advantages that AI has to offer in society, more work must be done to examine how technologies operate on a social level. After all, AI is much more than just data sets and code.
