Earlier this week, reports emerged that prominent AI ethicist and co-lead of Google’s Ethical Artificial Intelligence Team, Timnit Gebru was fired from her role. This was allegedly the culmination of about a week of disputes over the company’s request that Gebru retract a research paper she had co-written. A research paper contending that companies could do more to ensure that AI systems aimed at mimicking human writing and speech do not exacerbate historical gender biases and the use of offensive language.
Although Google maintains that the firing was in fact a resignation, in any case this story is yet another prominent example of the continued issues that we face in regard to ethics within the AI sector. And given the increasing use of these technologies in our society, it has sparked a difficult and timely discussion.
The truth is that examples of discriminatory AI are rarely down to technical errors or mistakes made in the creation of the technology. Rather, AI mirrors and replicates systems of inequality in society at large. This can be heightened by a lack of diversity amongst the creators, developers and researchers within the industry.
While the technical aspects of design should not be overlooked, it is important to remember that AI is not implicitly biased. As an industry, we must continue to identify and mitigate bias in given datasets, which will work to prevent AI programs from propagating existing inequalities. But we must also look much further afield if we are to tackle the underlying roots of the problem.
Progress in the industry so far
Although much progress has been made since the dialogue around ethics and diversity began, there is still much work to be done. In just over half a decade, under heavy pressure from activists, companies such as Google, Facebook and Apple to name but a few have published diversity reports, hired heads of diversity and inclusion, and overhauled their hiring practices.
As expected, the percentage of technical employees from underrepresented backgrounds at these corporations were poor, but this transparency was seen to be a step in the right direction.
Conversations of the past, much like those prompted by Gebru, have also made way for a host of non-profit organizations such as Project Includeand Diversity.ai. These organizations have urged companies into action, advocating improved hiring practices and diversity and inclusion solutions.
So why does the number of employees from underrepresented backgrounds remain low?
Where are we now?
Ultimately, although many organizations might now place an inclusive ethos at the heart of their company culture, most efforts so far have only made a difference on a surface level. In fact, a recent study from the AI Now Institute of New York has revealed that despite recent efforts, more than 80% of AI professors are men. Where race is concerned, the narrative is equally troubling, as the report uncovered that just 2.5% of Google’s workforce is black, meanwhile the percentage of black employees at Facebook and Microsoft is just 4% respectively.
In an era of heighted activism, where movements like Black Lives Matter and #MeToo have forced businesses across all sectors to take criticism on board, it seems like the AI industry has some catching up to do.
Closing the gap
To change the state of affairs and ensure that everybody in society is able to benefit from AI, as a society we must take practical steps in the right direction.
Although there will be no quick fix, work must begin with Governments and educational institutes that have the power and funds to perform better STEM outreach. Likewise, companies must reward hard-working members of staff from underrepresented backgrounds with promotions into leadership roles and technical positions. By doing so, companies will ensure that the pioneers of the future have strong role models, and can see themselves in the leaders of today.
Business leaders should also consider ways to foster an environment of inclusivity and trust. Not only should the company culture encourage employees to speak up if they don’t believe that products and practices reflect appropriate diversity standards, but it should also ensure that more marginalized voices are genuinely heard.
Businesses would do well to remember that real change goes beyond just audits and reports – transparency means little unless we act. Once these underlying factors are addressed, we will all be able to benefit democratically from the exciting advances that are constantly emerging in the AI space.