Written by: barbara-hyman
Does video hiring productise bias?
In recent years, we have all wisened up to the risk of using CVs to assess talent. A CV as a data source is well known to amplify the unconscious biases we have. A highly referenced study from 2003 called “Are Emily and Greg More Employable than Lakisha and Jamal?” found that white names receive 50 per cent more callbacks for interviews.
However, during COVID, we reverted to old ways in a different guise.
HR substituted CV as a data input with video interviews.
This isn’t a step forward.
Video hiring productises bias. It actually enables bias at scale.
It leads to mirror hiring – those who look and sound most like me. Instead of screening CVs in 30 seconds now, your team is watching 3-minute videos, so recruiting takes longer, and it’s exhausting.
Video platforms are being challenged in the US (EPIC Files Complaint with FTC about Employment Screening Firm HireVue) for concerns over invisible biases that may be affecting candidate fairness given the opaque nature of those algorithms. Facial recognition systems are worse at identifying the gender of women and people of colour than at classifying male, white faces. This year IBM openly pulled out of facial recognition, fearing racial profiling and discriminatory use, partly due to the questionable performance of the underlying AI.
How did we substitute one inferior and biased methodology with another that’s arguably even more biased?
We get that at some point you and the candidate need to meet, although no rule says you need to see someone to hire them. That’s just a bias (much like the bias pre-Covid) that you need to see someone at work to know that they are doing the work.
Blind hiring means you are interviewing a candidate without seeing them or knowing what school they went to, the jobs they have had. It’s a real meritocracy in that it’s fair for the candidate – and also smart for your organisation.
If you are hanging your hat on the fact you just finished bias training- research has shown consistently unconscious bias training does not work.
While we have all been dutifully attending it for years, the truth is the change factor is zero.
At a recent event attended by academics and data-loving professionals –whilst there was a welcome recognition that humans are more biased than Ai, and despite hearing that Wikipedia lists more than 150 biases we humans have – the majority of the audience still believe the impossible: that we can be trained out of our unconscious biases.
Algorithms are better at dealing with biases
The Nobel Prize winner Dr Daniel Kahneman prescribes an algorithmic approach as better at decision-making to remove unconscious biases. He claims “Algorithms are noise-free. People are not. When you put some data in front of an algorithm, you will always get the same response at the other end.” Also, see why machines are a great assistive tool in making hiring a fair process, here.
We know your inbox is flooded with Ai tools with each proclaiming to remove bias and give you amazing results and it’s tough to discriminate between what’s puffery, what’s real and what you can trust.
If your role requires you to know the difference between puffery and science, then read this. Buyers Guide: 8 Questions You Must Ask.
The 30-second due-diligence test that every HR professional should be asking when presented with one of these whizz-bang Ai tools is this:
- No data scientists in the team = not likely to be based on Ai
- No research available even under NDA to substantiate the method of assessment being used = pseudoscience or science that’s flawed if the company is not prepared to share it
- No regular bias testing to review = the Ai is likely to be biased in application
- Data used to training the models is 3rd party/ social media data = high risk of bias.
It’s critical, in fact, it’s a duty of care you have to your candidates and your organisation to be curious and investigate deeply the technology you are bringing into the organisation.
We have to be careful not to think that all AI is biased. AI is based on data, and that data can be tested for bias. ‘Data-driven’ also means transparent. Testing for bias, fairness and explainability of AI models is an active area of research and has advanced a lot. If built with best practices, AI can be used to challenge human decisions and interrupt potential biases. In the end, hiring is a human activity, and the final outcome should always be owned by a human.
If you want to know more about the research that defines the Sapia approach, look here.
If you want to know more about our bias testing, look here.
Have you seen the Inclusive e-Book?
Making inclusion an HR, not a PR priority.
It offers a pathway to fairer hiring in 2021. In this Inclusivity e-Book, you’ll learn:
- How to design an inclusive recruitment path. From discovery to offer and validation of the process.
- The hidden inclusion challenges that are holding your organisation back.
- How to tell if Ai technology is ethical.
Get diversity and inclusion right whilst hiring on time and on budget.