Talat Yaqoob: Prejudiced AI isn’t up to the job

Significant advancements have been made in development of artificial intelligence over the past decade, which has seen speech recognition technology become a standard part of households.

Voice recognition tends to favour men. Picture: Getty/iStockphoto

However, an area in which this technology has yet to deliver the progress we need as a society, is in equality and diversity.

In her recent book, Invisible Women, Caroline Criado-Perez provides hundreds of examples of where science, engineering and technology have forgotten about women when innovating. Speech recognition is just one example – AI technology has been found to understand male voices more accurately.

Sign up to our Opinion newsletter

Sign up to our Opinion newsletter

Accurate and bias-free analysis of human linguistics, emotion and cultural context are currently beyond the parameters of artificial intelligence. Instead, the research tells us that some AI tools are further embedding inequality, particularly in the area of recruitment.

For example, Amazon’s internal AI recruitment tool had to be shelved as it replicated gender inequality patterns, shortlisting a disproportionate number of male candidates and overlooking qualified women.

The inequalities that exist in society have been shown to be replicated in AI because the algorithms behind them are created by people (the almost 80 per cent male tech sector) with their own biases, which are “passed on” into the coding DNA of some of the technology around us – all the more reason to have a diverse workforce developing tomorrow’s technology.

Most employers are aware (or should be) of biases that can, all too easily, exist within recruitment processes. Which CVs are shortlisted for interview, what we define as “relevant experience” and how we assess performance at interview, are all susceptible to biases, the same biases that are the cause of and consequence of institutionalised inequalities we experience across society. For example NatCen research found that candidates with “ethnic minority sounding names” were half as likely to be invited to interview, and research by HP found that women were less likely to apply for jobs unless they met at least 90 per cent of the essential criteria.

Equate Scotland has been leading work in this area since our first inclusive recruitment training in 2013, which focused on tackling the inequality that can exist in interview processes.

Crucially we deliver this alongside wider and deeper training for employers on identifying bias and creating inclusive workplaces. We do this through an intersectional lens, ensuring that we have challenging conversations around sexism, racism, homophobia, disablism and classism.

Language is complex; how it is used comes with wider societal context and linked prejudices. Some online tools already exist which attempt to identify or eliminate language bias; the Gender Decoder was launched in 2011 which allows individuals to upload documents to alert them to gendered word use; similarly Eploy launched a review of language that goes beyond gender bias and works across a number of inequalities.

However, these types of online tools provide only a surface-level analysis and fail to provide assessment beyond basic recommendations. Until technology can master the nuance of language, culture and prejudice, the recruitment process will continue to require a “human touch”. It is perhaps an acknowledgement of this reality that has seen Equate Scotland’s job review services garner so much interest from employers, engaging five times the number of employers in the first six months than was predicted for the first year of this new initiative.

To support our work and the pursuit of equality and diversity, Equate Scotland is working with developers to consider ways in which these online assessment tools can become more accurate and be coded to use a wider, more nuanced approach.

To genuinely assess deeply embedded prejudices in workplace culture and exclusionary recruitment processes, it makes clear business sense for employers to consult real experts and engage equalities organisations. The recruitment process is just one area where exclusionary practices can exist. If we are to make any real and lasting impact on inequality within employment then the focus must be on policy change, training and changing cultures within workplaces. If we succeed here, fairness will become the status quo in today’s workplaces and in tomorrow’s innovation.

Talat Yaqoob is Director of Equate Scotland