If cognitive biases can be removed from the decision-making process , the decisions that are made are likely to be based purely on fact with no emotions
Column
20 The HEALTH | January-February . 2024
Humans v machines - who makes better decisions ?
If cognitive biases can be removed from the decision-making process , the decisions that are made are likely to be based purely on fact with no emotions
BRA�NNOTES
BY ��NY �ERE�RA
Tony Pereira is an Independent Consultant and Founder of SuperTrouper365
Type of bias What it is Steps to deal with the bias Similarity bias
Experience bias
Experience bias
Distance bias
Safety bias
We are motivated to look for ourselves in the other person . It mainly affects recruiting , and promotion decisions .
We rush to make a decision without considering all the facts .
We overemphasise experience when making new decisions . Experience is useful . The brain will always look for past experiences when making decisions . However , a balanced approach is necessary when dealing with each new situation . The phrase “ in my experience ” is an example of experience bias .
This is where we prioritise something close rather than something in the distance -whether physical , space or time . An example during this period of hybrid working conditions is giving more weight to the view of someone in person versus someone remote .
With safety bias , we aim to avoid loss rather than take risks . This is typical in sports when the underdog will seek to avoid losing rather than try to win .
Ensure that more than one person makes decisions and actively look for differences . Promote diversity .
Put in place a process where a decision needs to take into account multiple factors or needs group or majority agreement . I would also suggest a “ cooling off ” period where a decision is revisited after ( say ) 24 hours . The decision can be marked “ draft ” and then revisited . This cooling-off period will allow for the expedience bias to be reduced , if not eliminated . I see this mostly in marginal promotion / hire decisions when a preliminary decision is revisited after a short period of ����������
The moment we hear the phrase “ in my experience ,” it is a signal of experience bias . Just to be clear , we should not underestimate the experience curve . However , we should have a check and balance process in place and invite alternative views , reframe the discussion and perhaps even ask how much should be relied on past experience .
The way to deal with this is to continuously acknowledge individuals who may not be present in person . A good leader will be aware of the distance bias .
I remember a mentor of mine advising me on how to deal with the safety bias . He said the best way is to “ take yourself out of the equation .” What he meant by that is to assume the outcome does not affect you . If that is the situation , what decision would you make ?
“ Of the 5,500 that the system recommended should not be released , 48 per cent of them were in fact , given bail by the judges . The research by Mullainathan found that a high number of the defendants released on bail by the judges did in fact subsequently commit a crime whilst on bail .”
SOME years ago , an American economist from Havard University , Sendhil Mullainathan together with three elite computer scientists and a bail expert from the University of Chicago developed an Artificial Intelligence ( AI ) system to assess whether criminal defendants awaiting trial should be allowed bail .
They gathered the data of 554,689 defendants awaiting trial in New York . Using the AI system developed by Mullainathan ’ s team , the objective was to compare the decisions of the AI system with those of the judges who were hearing each case .
AI projected that 5,500 of those defendants would commit a crime if released on bail . The system of course , could not “ hear ” the arguments of defence counsel or that of the district attorney .
It simply made its decision based on hard data . The judges hearing the pleadings could , however see the defendant and hear the arguments put forward by the respective lawyers .
Of the 5,500 that the system recommended should not be released , 48 per cent of them were in fact , given bail by the judges . The research by Mullainathan found that a high number of the defendants released on bail by the judges did in fact subsequently commit a crime whilst on bail .
Had the courts followed the recommendations of AI , the number released would have been considerably lower - by up to 25 per cent and the number who committed crimes whilst on bail would have been reduced .
Does that mean that a machine is better at making decisions than humans ? When humans make decisions , we know that they use the limbic system ( which incorporates emotions ) and the prefrontal cortex ( the part of the brain that uses logic ). AI does not apply emotion when making decisions .
When the judges decide on whether to grant bail , they have the opportunity to see the defendant and sometimes hear from him . AI does not have such a luxury . Or is it a luxury ?
MINIMISING BIAS IN ��E �E�����N-�A��N� �R��E��
Psychologists Matthew Lieberman , David Rock , Heidi Grant and Christine Cox explain that we are all affected by cognitive biases - something the AI system is not . Cognitive bias is a mental shortcut that influences our thinking and decision-making .
It can lead to inappropriate conclusions or decisions . The existing literature tells us that there are more than 150 cognitive biases . This makes decision-making challenging - as we have seen from the example above .
So , how can we minimise the impact of cognitive bias in decision making ? Lieberman et al . have developed the SEEDS Model to help deal with cognitive biases . They identified over 150 cognitive biases and categorised them into five categories . They are Similarity bias , Expedience bias , Experience bias , Distance bias and Safety bias .
Based on their model , we can set up a framework to help minimise such biases in the decision-making process . A typical framework that can be applied is as set out table left :
For a full set of cognitive biases , refer to the Cognitive Bias Codex developed by John Manoogian III and Buster Benson . Manoogian and Benson developed the codex to make understanding the different types of biases easier .
In summary , Lieberman et al . l recommend a three-stage process to deal with cognitive biases , which is as follows :
• Accept that we are biased by virtue of our biology ;
• Label the type of bias that might affect the decision ( by referring to SEEDS );
• Mitigate using the right process . Another summary of cognitive bias is provided by the Cognitive Bias Codex developed by John Manoogian III and Buster Benson . Manoogian and Benson summarised the biases under the following categories :
• Information Biases ( 108 biases )
• Decision Biases ( 111 biases )
• Illogical Biases ( 52 biases )
• Memory Biases ( 55 biases ) In this current age where AI is being applied in many tasks previously carried out by humans , the attraction of AI is not surprising . If cognitive biases can be removed from the decision-making process , the decisions that are made are likely to be based purely on fact with no emotions attached to the process .
The limbic system does introduce cognitive biases into the equation . As the example of the arraignments of defendants in New York has demonstrated , even the most educated person is influenced by cognitive bias .
The SEEDS model does help mitigate the issue but how practical is it ? It cannot be applied for every decision that needs to be made . But awareness is the first step to mitigation .
If humans do not accept this , then it is only a matter of time before defendants appear not before a judge but will have the judgment made by a computer system ; recruiting decisions are made by an invisible AI system rather than the Human Resources team ; even a medical diagnosis will be made by an AI system hosted somewhere rather than a doctor examining a patient ( which may already be happening ).
The world is changing rapidly . In time to come , the issue of cognitive biases could cease to be an issue . – The HEALTH