Can AI reduce cognitive bias in decision making?

Humans make biased judgments

The degree of bias is dependent on a huge number of factors which affect a human’s perspective of an issue. Too many to mention here but to focus on just one cognitive bias, confirmation bias is frequently featured in leadership development as the single greatest hindrance to good decision making.

We found the best overall representation of the Cognitive Biases in Terry Heick’s Cognitive Bias Codex: a visual of 180 cognitive biases.

The Codex is found at the heart of the world’s most recognised leadership qualifications

Mr Heick focuses on confirmation bias. His view is that confirmation bias kills critical thinking. We agree.

In “what is confirmation bias” we looked at this very common thinking mistake: the tendency to overvalue data and observation that fits with our existing beliefs.

The pattern is to form a theory (often based on emotion) supported with insufficient data, and then to restrict critical thinking and ongoing analysis, which is, of course, irrational. Instead, you look for data that fits your theory.

While it seems obvious enough to avoid, confirmation bias is particularly sinister cognitive bias, affecting not just intellectual debates, but relationships, personal finances, and even your physical and mental health. Racism and sexism, for example, can both be deepened by confirmation bias. If you have an opinion on gender roles, it can be tempting to look for ‘data’ from your daily life that reinforce your opinion on those roles.

This is, of course, all much more complex than the above thumbnail [sic]. The larger point, however, is that a failure of rational and critical thinking is not just ‘wrong’ but erosive and even toxic not just in academia, but every level of society.

Extract Terry Heick

Who ensures that the appropriate checks and balances are applied in order to reduce human bias?

It may seem obvious. Of course there is legislation. Regulated professions such as law, science, medicine, teaching, governance and accountancy, set out their policies and codes of conduct and ensure compliance, to act ethically and to reduce bias.

At the root of any remedy to bias, must lie a subjective and unmeasurable level of self-awareness held by the decision maker in order to make the best unbiased decisions.

We refer to Terry Heick again whereby he eloquently sets out the steps which lead to confirmation bias, together with recommended remedies.

Commonly, there are five steps to confirmation bias:

1. Form a theory (or ‘have an opinion’)

2. Find ‘data’ that supports that opinion

3. Work hard to collect more and more data that really ‘confirms’ your theory (i.e., what you believe)

4. Identify the kinds of data that are most compelling to the people you most frequently want to convince of your theory (because that’s what people like to do), then collect and memorize and repackage and refine that data to more neatly fit your theory

5. Become more emotional in your theory-holding (i.e., your opinion) because you’re now surer than ever that ‘you’re right’

6. Continue to discount and discredit new or better data because then you’d have to reconstruct your belief system, apologize to people, admit you were wrong, etc.

How To Resist Confirmation Bias

How can you keep from committing confirmation bias? Constantly re-evaluate what you believe that you know, insist on the highest quality data, and embrace the possibility that we’re all wrong more frequently than we’re right.The pattern, crudely put, would look something like this:

1. Consider all data equally and know how to separate good data from bad, fact from opinion, misrepresented facts from properly-contextualized facts, etc.

2. Form a theory based on said data/data sources

3. Be open to new data, ideas, constructs, and perspectives and revise your theories in response as necessary

4. Moving forward, use data to inform your theory forming instead of using your theories to inform your data seeking.

Summary

Confirmation bias is a killer of critical thinking.

And the opposite approach is exhausting.

To constantly consider a broad set of evidence and data (historical patterns, existing trends, widespread indicators, alternative explanations, etc.) and then narrow it down to identify higher-quality data in order to form a ‘fluid’ conclusion that you then consistently revisit in light of ‘new’ data as it becomes available–data that wasn’t handpicked to support a theory but rather is fresh, valid, credible, and relevant–takes a lot of cognitive energy and thinking strategies and human determination, not to mention humility, which is why it’s not as common as it could be.

In the HBR article The Key to Inclusive Leadership by Julia Bourke and Andrea Titus, the impact of cognitive bias on leadership is summarised

Inclusive leadership is emerging as a unique and critical capability helping organisations adapt to diverse customers, markets, ideas and talent. For those working around a leader, such as a manager, direct report or peer, the single most important trait generating a sense of inclusiveness is a leader’s visible awareness of bias. But to fully capitalize on their cognizance of bias, leaders also must express both humility and empathy. This article describes organizational practices that can help leaders become more inclusive and enhance the performance of their teams.

The Key to Inclusive Leadership by Julia Bourke and Andrew Titus

Shifting the debate into the technology arena, if humans were judged in the way AI based systems are judged, there would be no leaders

Humans flawed by cognitive biases are still relied upon in the most important critical decision making processes. What makes humans so well qualified to make unbiased decisions when they are so inherently biased?

Terry Heick leaves us a big clue in his summary:

To constantly consider a broad set of evidence and data (historical patterns, existing trends, widespread indicators, alternative explanations, etc.) and then narrow it down to identify higher-quality data in order to form a ‘fluid’ conclusion that you then consistently revisit in light of ‘new’ data as it becomes available–data that wasn’t handpicked to support a theory but rather is fresh, valid, credible, and relevant–takes a lot of cognitive energy and thinking strategies and human determination, not to mention humility, which is why it’s not as common as it could be.

Humans are frequently unable to effectively process the amount of information required to make the best decisions

Therefore it could be argued that AI based systems which assist human decision making may result in the correction and gradual removal of confirmation bias. The human must hold the final check and balance, however with access to a large amount of the right analysed data – could the human decisions become more balanced?

AI systems such as automated facial recognition have been judged, banned and feared across the world. In the state of Oregon in the USA, the City of Portland passed ordinances which ban the use of the technology in order to protect the privacy of individuals in public places, especially those from targeted groups. The theory behind the ban was to prevent law enforcement agencies and any other authority from using “biased” technology which may amplify human bias rather than reduce it. This is a compelling argument for a ban, to which the technology developers have responded, looking inwardly at their development and test processes and specifically the training data sets, (the images used to teach the AI to recognise faces), and fixing the skews and biases built in by biased humans.

In 2018, NIST compared human forensic examiners with technology and concluded that humans performed better if there were multiple opinions of a subject and that facial recognition software performed as well as the human forensic examiners.

NIST further concluded that a human/technology partnership was the most accurate over and above a human/human partnership

Tests have since been carried out on some of the world’s leading high end facial recognition technology which revealed that newer developments exhibit less that 1% bias in their readings of a diverse range of facial images. This is a step in the right direction and with the advent of synthetically generated training data sets, it is conceivable that facial recognition technology could soon be proven to be 100% accurate. That’s more than we can say about humans.

Is there a single human being on this planet who is 100% unbiased?

The AI is revealing human bias. This is an opportunity to spotlight the bias which may appear in these technologies – they are after all a reflection of human flaws, and systematically eliminate the source of the bias through education and inclusion.

With unbiased AI technology assisting the decision making process, could we see a gradual correction of systemic bias?

The British Security Industry Association have released a Guide to the legal and ethical use of automated facial recognition technology, and while regulation is still around the corner, the guidance leads the user through the best ways of using the technology legally and ethically. Anekanta Consulting’s leader was one of the creators of the guide. To download a copy, head over to the BSIA web site here.

For further information about how AI can further assist human decision making, contact us at ask@anekanta.co.uk.

We'd love to hear your views.....

This site uses Akismet to reduce spam. Learn how your comment data is processed.