Each month we will be sharing four, carefully-chosen articles from the Latest Issue of our flagship publication ‘theHRDIRECTOR’ which exemplify the high standards we strive to achieve. We hope you find this in-depth article of interest and decide to become one of our valued Subscribers.
Artificial intelligence is revolutionising recruitment, taking evidence-based decision-making to a new level and factoring in an unprecedented amount of data from a wide array of sources. But when an algorithm being tested as a recruitment tool by Amazon was discovered to be sexist, it had to be scrapped. Whilst unconfirmed, this raises a number of questions about unconscious bias and how AI or machine learning should be applied to talent acquisition.
Article by Charles Hipps, CEO – Oleeo
There are, of course, big benefits to AI in recruiting because, in theory, it affords employers an even greater ability to quickly flag candidates that have certain key indicators of success, thus streamlining the selection process and affording more time to nurture top talent ahead of competitors. Used well, practitioners can test theories, proactively solve problems and conduct more complex predictive analytics related to sourcing and hiring strategies. This isn’t something that’s in the future, it’s already here, we’re already using it in our smartphones – where there is a blurring of lines that needs consideration, as they are equally used for work, social and leisure lives – and that is becoming one of the major ways in which AI works, including in the world of recruitment. So, in consideration of the exposure this presents, there has to be caution and safeguards in place, so that AI is used appropriately, both morally and with new legislation governing personal data clearly in mind. So, it is important that when we are using aggregated data sets to make decisions, we need to make sure that we are correcting for existing biases.
Obviously, AI technology has advanced a long way in thirty years, but it would be foolish to assume that bias is no longer a problem. For example, the journal Science, reported on a study of an advanced AI language tool, which exhibited racial and gender bias, it associates the word woman with a home, arts and humanities and men with maths and science. It also associated European sounding names with words like gift and happy, and African sounding names with negative terms. Clearly relying on those sorts of tools to make decisions is going to create discrimination law risks if an employer cannot prove the logic to the methodology it has applied to an AI algorithm. It is important though that a tool never does this based on historical data alone. Efforts must be made to constantly improve the robustness of any tool to help employers benefit from the best possible early evaluation of applicants, based on responses given within online application forms. This can be achieved by collecting enormous amounts of structured and unstructured data, processes that data using the best from thousands of machine-learning algorithms to most accurately predict outcomes, and refines that process as it learns. Indeed, Constant machine learning will work to reduce unconscious biases and enhance diversity by uncovering strong candidates, who may have gone unnoticed in a non-intelligent or manual process. In turn, recruiters gain insight and reasoning into which characteristics score the strongest.
“The journal Science, reported on a study of an advanced AI language tool, which exhibited racial and gender bias, it associated the word woman with a home, arts and humanities and men with maths and science”
We commissioned the Department of Computer Science at University College London, to look into how algorithms can ensure that they do not inadvertently fall into gender bias, as Amazon appears to have done. It revealed that removing any wording or phrases that could unconsciously predict the gender of a candidate would enable algorithms to make any gender prediction to be no better than random, with no direct impact from the loss of information in the transformation and de-biasing steps. In fact, more consistent disparate impact scores of close to 1.0 – i.e. no disparate impact observed – are recorded in hiring predictions undertaken in this way providing better hired prediction performance. It is also shown to have consistent negligible disparate impact across a range of hiring values, providing room for adjustment in recruitment screening thresholds without increasing disparate impact. Working in this way allows employers to foster diversity and accelerate candidate selection, promising no adverse selection, in compliance with established selection rate guidelines around the four-fifths guidelines. Customised, algorithms can elegantly handle high-volume automation and deliver at-a-glance qualified, quality candidate recommendations critical to recruiting success in large-scale hiring events. It’s important though to put this into focus. An AI future is not about people versus machines, it is about people and machines collaborating in harmony, using intelligent organisational design. After all, technology was created by people to enhance their lives. So, AI should be considered more as a leveller, helping any recruiter to highlight the diamonds in the rough, that no one else knows about.
Increasingly, companies want to do the right thing when it comes to fostering diversity, right from the start of the recruiting process. In terms of compliance, however, many companies don’t have standard processes in place to ensure they are meeting set standards. Correctly tuned, algorithms can help companies shift from being reactive to proactive in balancing the need to accurately and quickly identify high-quality candidates, while simultaneously ensuring compliance. This can lead to a greater democratisation of recruitment by recommending candidates who unequivocally perform better and stay longer. It also leads to better record keeping and motivate reproducible decision making. It also removing the economic bias to exclude, enabling employers to better understand what drives performance, whilst moving away from the familiar “tried & tested”. The automated cycle of recruitment means you should have a better talent pool of candidates coming through that reflect the future leaders you want joining your organisation. Clever data techniques will recommend candidates who unequivocally perform better and thereby deliver more revenue, profit, or stay longer in the business. It means that a business can go on to use algorithms, based on how employees perform in the business rather than what line managers decide at interview. In so doing, it is feasible that technology could effectively and significantly free up recruiter resources each year – time which could be spent on adapting better engagement techniques, to ensure a leading candidate, with many offers at their disposal, is more likely to buy into the; culture, mission and vision, ahead of market competitors with equally tempting offers on the table. In the recruitment game, closing down top talent ahead of competition is a big challenge and this technology is helping to offer a solution, and reduce decline rates to suit corporate objectives.
So, as AI plays an increasing role in helping firms reduce reliance on gut instinct of recruiters and hiring managers, by enabling them to effectively utilise the plethora of recruiting data they already have, such as; data on high, medium and low performing employees; candidate demographics, sources of hire and background data; assessment and psychometric data; structured interview data etc. Knowing what worked well in the past can help to fine-tune the types of candidates that carry high favour within a firm. Harnessing the potential of AI means you’re not just dismissing elitism theories but you’re also identifying and quantifying any historic bias reducing bias in future decision making. The algorithms eliminate the chance of disparate treatment by not accepting protected category data and replicating collective decision making to reduce the influence of bias by individuals or process. It means you can mitigate the influence of disparate impact and focus on just winning great hires. Reporting wise, it helps with ensuring that you are providing stronger evidence and recordkeeping to support hiring decisions and can accept more applications with lower resource implications. Clever algorithms replicate your collective decision-making, reducing the influence of bias by individuals or process, and recruiters shouldn’t fear it as a threat to their careers – humans still have to understand how it’s being used. Data, in itself, will not land you a decision – you need to have data, you need to be able to generate insight and you also need to be able to link that to action. Indeed, humans should never be taken out of the equation, as the actual technology itself is just an enabler. When leveraged efficiently, predictive analytics allows staffing teams to create economic value from their talent data, helping them become more competitive and successful – we all know the huge cost and disruption a wrong hire causes – it leads to lower productivity, reduced levels of employee morale and engagement and ultimately leads to more attrition. A vicious circle that is best avoided, if at all possible.
