The value is greatest when companies harness the differences between employees from multiple demographic backgrounds to understand and appeal to a broad customer base. But true diversity relies on social mobility and therein lies the problem: the rate of social mobility in the UK is the worst in the developed world.
The root cause of the UK’s lack of social mobility can be found in the very place that it can bring the most value – the workplace. Employers’ recruiting processes often suffer from unconscious human bias that results in involuntary discrimination. As a result, the correlation between what an employee in the UK earns today and what his or her father earned is more apparent than in any other major economy.
This article explores the barriers to occupational mobility in the UK and the growing use of predictive analytics or algorithmic hiring to neutralise unintentional prejudice against age, academic background, class, ethnicity, colour, gender, disability, sexual orientation and religion.
The UK government has highlighted the fact that ‘patterns of inequality are imprinted from one generation to the next’ and has pledged to make their vision of a socially mobile country a reality. At the recent Conservative party conference in Manchester, David Cameron condemned the country’s lack of social mobility as unacceptable for ‘the party of aspiration’. Some of the eye-opening statistics quoted by Cameron include:
The OECD claims that income inequality cost the UK 9% in GDP growth between 1990 and 2010. Fewer educational opportunities for disadvantaged individuals had the effect of lowering social mobility and hampering skills development. Those from poor socio economic backgrounds may be just as talented as their privately educated contemporaries and perhaps the missing link in bridging the skills gap in the UK. Various industry sectors have hit out at the government’s immigration policy, claiming this widens the country’s skills gap still further.
Besides immigration, there are other barriers to social mobility within the UK that need to be lifted. Research by Deloitte has shown that 35% of jobs over the next 20 years will be automated. These are mainly unskilled roles that will impact people from low incomes. Rather than relying too heavily on skilled immigrants, the country needs to invest in training and development to upskill young people and provide home-grown talent to meet the future needs of the UK economy. Countries that promote equal opportunity for everyone from an early age are those that will grow and prosper.
The UK government’s proposal to tackle the issue of social mobility, both in education and in the workplace, has to be greatly welcomed. Cameron cited evidence that people with white-sounding names are more likely to get job interviews than equally qualified people with ethnic names, a trend that he described as ‘disgraceful’. He also referred to employers discriminating against gay people and the need to close the pay gap between men and women. Some major employers – including Deloitte, HSBC, the BBC and the NHS – are combatting this issue by introducing blind-name CVs, where the candidate’s name is blocked out on the CV and the initial screening process. UCAS has also adopted this approach in light of the fact that 36% of ethnic minority applicants from 2010-2012 received places at Russell Group universities, compared with 55% of white applicants.
Although blind-name CVs avoid initial discriminatory biases in an attempt to improve diversity in the workforce, recruiters may still be subject to similar or other biases later in the hiring process. Some law firms, for example, still insist on recruiting Oxbridge graduates, when in fact their skillset may not correlate positively with the job or company culture. While conscious human bias can only be changed through education, lobbying and a shift in attitude, a great deal can be done to eliminate unconscious human bias through predictive analytics or algorithmic hiring.
Bias in the hiring process not only thwarts social mobility but is detrimental to productivity, profitability and brand value. The best way to remove such bias is to shift reliance from humans to data science and algorithms. Human subjectivity relies on gut feel and is liable to passive bias or, at worst, active discrimination. If an employer genuinely wants to ignore a candidate’s schooling, racial background or social class, these variables can be hidden. Algorithms can have a non-discriminatory output as long as the data used to build them is also of a non-discriminatory nature.
Predictive analytics is an objective way of analysing relevant variables – such as biodata, pre-hire attitudes and personality traits – to determine which candidates are likely to perform best in their roles. By blocking out social background data, informed hiring decisions can be made that have a positive impact on company performance. The primary aim of predictive analytics is to improve organisational profitability, while a positive impact on social mobility is a healthy by-product.
A recent study in the USA revealed that the dropout rate at university will lead to a shortage of qualified graduates in the market (3 million deficit in the short term, rising to 16 million by 2025). Predictive analytics was trialled to anticipate early signs of struggle among students and to reach out with additional coaching and support. As a result, within the state of Georgia student retention rates increased by 5% and the time needed to earn a degree decreased by almost half a semester. The programme ascertained that students from high-income families were ten times more likely to complete their course than those from low-income households, enabling preventative measures to be put in place to help students from socially deprived backgrounds to succeed.
Bias and stereotyping are in-built physiological behaviours that help humans identify kinship and avoid dangerous circumstances. Such behaviours, however, cloud our judgement when it comes to recruitment decisions. More companies are shifting from a subjective recruitment process to a more objective process, which leads to decision making based on factual evidence. According to the CIPD, on average one-third of companies use assessment centres as a method to select an employee from their candidate pool. This no doubt helps to reduce subjectivity but does not eradicate it completely, as peer group bias can still be brought to bear on the outcome.
Two of the main biases which may be detrimental to hiring decisions are ‘Affinity bias’ and ‘Status Quo bias’. ‘Affinity bias’ leads to people recruiting those who are similar to themselves, while ‘Status Quo bias’ leads to recruitment decisions based on the likeness candidates have with previous hires. Recruiting on this basis may fail to match the selected person’s attributes with the requirements of the job.
Undoubtedly it is important to get along with those who will be joining the company. The key is to use data-driven modelling to narrow down the search in an objective manner before selecting based on compatibility. Predictive analytics can project how a person will fare by comparing candidate data with that of existing employees deemed to be h3 performers and relying on metrics that are devoid of the type of questioning that could lead to the discriminatory biases that inhibit social mobility.
“When it comes to making final decisions, the more data-driven recruiting managers can be, the better.”
‘Heuristic bias’ is another example of normal human behaviour that influences hiring decisions. Also known as ‘Confirmation bias’, it allows us to quickly make sense of a complex environment by drawing upon relevant known information to substantiate our reasoning. Since it is anchored on personal experience, it is by default arbitrary and can give rise to an incorrect assessment.
Other forms of bias include ‘Contrast bias’, when a candidate is compared with the previous one instead of comparing his or her individual skills and attributes to those required for the job. ‘Halo bias’ is when a recruiter sees one great thing about a candidate and allows that to sway opinion on everything else about that candidate. The opposite is ‘Horns bias’, where the recruiter sees one bad thing about a candidate and lets it cloud opinion on all their other attributes. Again, predictive analytics precludes all these forms of bias by sticking to the facts.
Age is firmly on the agenda in the world of recruitment, yet it has been reported that over 50% of recruiters who record age in the hiring process do not employ people older than themselves. Disabled candidates are often discriminated against because recruiters cannot see past the disability. Even these fundamental stereotypes and biases can be avoided through data-driven analytics that cut to the core in matching attitudes, skills and personality to job requirements.
Once objective decisions have been made, companies need to have the confidence not to overturn them and revert to reliance on one-to-one interviews, which have low predictive power. The CIPD cautions against this and advocates a pure, data-driven approach: ‘When it comes to making final decisions, the more data-driven recruiting managers can be, the better’.
The government’s strategy for social mobility states that ‘tackling the opportunity deficit – creating an open, socially mobile society – is our guiding purpose’ but that ‘by definition, this is a long-term undertaking. There is no magic wand we can wave to see immediate effects.’ Being aware of bias is just the first step in minimising its negative effect in the hiring process. Algorithmic hiring is not the only solution but, if supported by the government and key trade bodies, it can go a long way towards remedying the inherent weakness in current recruitment practice. Once the UK’s leading businesses begin to witness the benefits of a genuinely diverse workforce in terms of increased productivity and profitability, predictive hiring will become a self-fulfilling prophecy.
Being able to access interview automation just got so much easier inside Tribepad, with Sapia. To explore the use cases for Sapia, let’s chat.
Here’s a quick rundown:
And now that we are integrated into Tribepad, you get all of these smarts inside your existing Tribepad application. At Sapia, we interview every applicant in-depth and at scale for you. Overall, this is by using a text chat that helps you find the best people fast. Our underlying data science has been accepted and published in international journals.
Firstly, no one’s time is served well by screening thousands of CVs. With every additional applicant costs your business an extra $20 in screening if you are doing it the old way, automating the screening process is the commercial decision companies are now making.
Once your vacancy is created in Tribepad, a corresponding interview link will also be created.
Candidates click this link to enter their text-based interview. This is known as the ChatInterview.
As soon as candidates complete their interview their results are displayed inside Tribepad. You also get to see the candidate’s personality assessment. With the pre-assessment already done for you, it makes shortlisting much faster. Thus, by sending out one simple interview link, you nail speed, quality and candidate experience.
The SmartInterview experience is most commonly used for high-volume recruiting. Our customers typically use it in frontline customer-facing roles (like contact centres, customer service) and/or for low-skill roles.
We help manage the disconnect between attraction and retention. This is all done by allowing Recruitment Teams to work more efficiently to hire the best talent. All is done whilst ensuring the applicants feel good about applying for a job role.
Sapia solves the time problem of managing a large applicant pool. It also tackles the quality problem of pin-pointing the best people from that pool. Additionally it solves the candidate experience problem by offering every applicant a fair chance at the opportunity (everyone gets an interview) on platforms they love to use. Simultaneously every candidate gets something of immense value in return for their application.
We are glad you are asked! The first thing to note is Sapia is a paid app and sold separately. Next, to explore the pricing that suits your organisation, let’s chat. Lastly, our team can take you through the integration process and describe how the interview automation experience works.
Also, to keep up to date on all things “Hiring with Ai” subscribe to our blog!
Artificial Intelligence-based interview scoring learns from past interview answers, which makes it hard for it to determine if a candidate is legitimately answering the question if their response includes context or an example rarely seen in training data.
Moreover, AI interviewers may be susceptible to adversarial inputs where an irrelevant answer may receive a high score. Both scenarios raise fairness concerns and can erode trust in AI interviewers (Madaio et al, 2020).
This is why identifying outliers that differ significantly from the majority of answers and flagging them for manual review become crucial steps toward responsible and fair use of AI interviewers. While simple rule-based methods (Reiz and Pongor, 2011) could help filter out some irrelevant answers based on answer length and regular expressions, these methods do not take into account the context and content of the answer and question. Someone may describe a very unique, yet relevant, situation in response to an interview question, which you wouldn’t want to disregard.
In this study, we introduce an unsupervised, question-aware, multi-context outlier detection model that can help detect anomalous answers contextually and semantically. The unsupervised approach is deemed to be more practical compared to a supervised model that requires a large labeled dataset of outlier answers. It helps bootstrap an outlier detector that can then be enhanced through human feedback.
We tested the outlier model to ascertain how well it is able to correctly identify 177,691 actual hired candidate interview answers from outliers, (e.g., movie reviews, news articles, nonsensical text, and sentences generated using BERT (Vaswani et al, 2017) with random starting words).
Our model outperformed the baseline One-class SVM outlier detector (Li et al, 2003), in detecting outliers from actual interview answers. The performance of our model over the baseline unsupervised model can be explained by both question-aware learning and multi-context learning, which help yield better contextual representations for detecting outlier answers from typical interview answers.
We also conducted a human evaluation on 10,689 interview answers of candidates who were not hired and might have provided outlier answers. Our model predicted 0.16% of the answers as outliers with only 5.9% of them being false positives. All of these false predictions describe contexts related to family and personal life in their answers but are relevant to the question. It is reasonable that these answers are labeled as an outlier by our model since they are contextually and semantically different from most interview answers.
While a data-driven AI interviewer can help counter flaws in human interviewers, answers that are significantly different to training data can lead to spurious predictive outcomes. In this study, we show how a
Dai, Y., Qi, J., & Zhang, R. (2020). Joint recognition of names and publications in academic homepages. In Proceedings of the 13th International Conference on Web Search and Data Mining (pp. 133-141).
Li, K. L., Huang, H. K., Tian, S. F., & Xu, W. (2003). Improving one-class SVM for anomaly detection. In Proceedings of the 2003 international conference on machine learning and cybernetics (IEEE Cat. No. 03EX693) (Vol. 5, pp. 3077-3081). IEEE.
Madaio, M. A., Stark, L., Wortman Vaughan, J., & Wallach, H. (2020). Co-designing checklists to understand organizational challenges and opportunities around fairness in AI. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (pp. 1-14).
Reiz, B., & Pongor, S. (2011). Psychologically Inspired, Rule-Based Outlier Detection in Noisy Data. In SYNASC (pp. 131-136)
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., … & Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.
What can AI help us discover? How can we make better people and business decisions by looking at the data?
By using SOM maps https://en.wikipedia.org/wiki/Self-organizing_map to map personality for more than 85,000 applicants using their HEXACO scores, 47/53% male and female, candidates spread across 2 regions – the UK and Australia, we identified 400 unique personality profiles.
It turns out that personality is somewhat more complex than the 16 types long promoted by Myers Briggs.
Following SOM’s show the percentage density of male, female and sales candidates across the 400 different HEXACO profile groups. The size of each bubble represents the total count of individuals mapped to each profile. Darker shades represent higher % of each category.
Personality is widely accepted as an indicator of job performance. Until now, the only way to accurately measure personality was through long and repetitive 100+ item personality tests, where the candidate experience is proven to be weak. The Sapia team breaks new ground disrupting decades of assessment practice. They do this by showing that answers to standard interview questions related to past behaviour and situational judgement can be used to reliably infer personality traits. Thus by leveraging NLP, machine learning and personality theory, we validate that text is a reliable indicator of hidden personality traits. Additionally, this approach to candidate interviews is blind to gender, race and any characteristics that are not directly relevant in job selection. Instead, every applicant is given a fair opportunity to express themselves and be evaluated equally.
To keep up to date on all things “Hiring with Ai” subscribe to our blog!