Unconscious bias: what is it and can it be eliminated?

In the ranking of taboos, racism and sexism come close to the top of the list. So it is perhaps unsurprising that the concept of unconscious or implicit bias has gripped the popular imagination to a greater degree than any other idea in psychology in recent decades.

Spearheaded by a team of social psychologists at the University of Washington and Yale, the Implicit Association Test (IAT) promised to lift the veil on people’s subconscious attitudes towards others. Upon publishing their landmark paper in 1998, the team described “a new tool that measures the unconscious roots of prejudice” that they said affected 90-95% of people.

Unconscious bias – the subject of the Guardian’s Bias in Britain series – offered a new explanation for why, despite equalities apparently being enshrined in law, society still looked so unfair. And by framing prejudice as something that could be involuntarily soaked up from the world around us, the IAT provided people and businesses with an acceptable way to talk about the problem.

Since then, countless studies have confirmed the power of racial biases to shape everyday decisions in almost every aspect of life. White job applicants were found to be 74% more likely to have success than applicants from ethnic minorities with identical CVs. University professors were found to be far more likely to respond to emails from students with white-sounding names. US doctors have been found to recommend less pain medication for black or Latino patients than white patients with the same injury. White participants in a study were found to perceive black faces as more threatening than white faces with the same expression.

Neuroscientists have uncovered brain regions involved in racial and gender stereotyping and shown that such stereotypes begin to form early in childhood. Recent work found that the brain responds more strongly to information about ethnic groups who are portrayed unfavourably, suggesting that the negative depiction of minorities in the media can fuel bias.

Scientists believe that stereotypes in general serve a purpose because clustering people into groups with expected traits help us navigate the world without being overwhelmed by information. The downside is that the potential for prejudice is hard-wired into human cognition.

The evidence is overwhelming that unconscious bias seeps into decisions that affect recruitment, access to healthcare and outcomes in criminal justice in ways that can disadvantage people from ethnic minorities.

However, at the individual level, the extent to which such biases are internalised and acted on varies widely and in complex ways. Life experience, such as dating outside your racial group or having a boss from a minority group, can strongly protect against holding negative stereotypes. And there is dispute about the extent to which such biases are truly unconscious.

Lasana Harris, a neuroscientist who studies prejudice and social learning at University College London, said the concept of unconscious bias should not absolve people of discriminatory behaviour. “If you’re aware of these associations then you can bring to bear all of your critical skills and intelligence to see it’s wrong to think like that,” he said. “We all have the ability to control that.”

In the years since its incarnation, the concept of implicit association testing has been extensively promoted by Project Implicit at Harvard University, a nonprofit to “educate the public about hidden biases”. More than 30m tests have been taken on the project’s website. During the first 2016 US presidential debate, Hillary Clinton argued that “implicit bias is a problem for everyone”.

However, there have been rumblings of discontent in the academic world, with disagreement about what IAT scores actually reveal about a person. Multiple studies have found IAT results are not strongly reproducible in an individual: one day you might have a “moderate” bias and the next day you might come out in the clear.

Brian Nosek, of the University of Virginia, who was part of the team that developed the IAT, acknowledged that the extent to which the result measures a meaningful trait had been misconceived and said the test should not be taken in this spirit.

“That’s an incorrect interpretation. There is some consistency but not high consistency,” he said. “Our mind isn’t that stable.”

A series of four meta-analyses during the past decade, each taking into account dozens of studies, found the IAT to be at best a weak predictor of behaviour. Nosek estimates that about 4% of the variation in someone’s behaviour might be accounted for by an IAT score – a big enough margin to make a difference over thousands of hiring decisions, for example. Others dispute this interpretation: several experimental studies, for instance, have counterintuitively shown people with more biased IAT scores are, in some contexts at least, less discriminatory in their behaviour.

There are also questions about unconscious bias training, with some raising the troubling possibility that if it is not rigorously run it could backfire.

A Canadian neuroscientist, Jacquie Vorauer, looked at the impact on interactions between white and aboriginal Canadians. Before meeting, some white participants took an IAT focused on attitudes towards aboriginal people, and a control group took a non-race IAT. Afterwards, the aboriginal participants in the race IAT group reported feeling less valued by their white partners, hinting that alerting people to their supposed biases could make them unusually cautious or inhibited.

“There’s definitely a risk that training can make things worse,” said Nosek. “There has been too much leaping to applications that the evidence doesn’t yet support. Diversity training is full of good intentions and weak evidence.”

Harris agrees there is a potential for people to become socially paralysed by a heightened awareness of their own supposed biases. “It’s like trying to have a conversation with kids present,” he said. “All this noise going on in your head and you can’t just chat.”

Rightwing commentators such as the psychologist Jordan Peterson have used the shortcomings of unconscious bias testing as a springboard to question whether women and minority ethnic groups are subject to any biases or disadvantages in the workplace at all. The sacked Google employee James Damore complained that diversity sessions were worthless because social and professional disparities had a biological basis.

In reality, many diversity programmes do more than simply sending employees to take an IAT, and some shun the approach altogether.

Erin Thomas, a partner at Paradigm, a US-based diversity consultancy that has worked with leading tech and media companies, said her company used IATs as a way to open up conversations before moving on to “more rigorous findings” on hiring procedures and structural barriers in the workplace.

For instance, Thomas said, there is good evidence that people shift their recruitment criteria (the weight given to academic versus practical qualification, say) depending on the profiles of competing candidates in a way that could disadvantage women and minority ethnic groups. But when recruiters were required to write down their priorities before selection began, this effect tended to disappear.

“Part of why we’re still having the same conversation is because there hasn’t been a ton of rigour in the practice of diversity and inclusion,” said Thomas.

Unconscious bias testing transformed people’s ability to discuss prejudice, and that prompted a wave of hope that inequalities could be tackled more effectively in future. A great irony, according to Nosek, is that the rush to apply it in business risks entrenching the problem rather than eliminating it.

“If you implement an intervention that doesn’t work, it can reinforce people’s beliefs that nothing will work,” he said. “It’s incumbent on us to use the best evidence available.”