Satnav systems have been blamed for all kinds of motoring mishaps. In one notable incident, a German motorist drove straight into the Havel river after following the instructions of his device. The gadget display showed a bridge across the river when it should have indicated a ferry.
Ignoring warnings from passengers, passersby and flashing red lights, the driver obediently steered his BMW off the end of the ferry ramp into 13ft of water. This unquestioning faith in computers is called automation bias.
But oddly, when people make forecasts – of investment performance or sales, sporting events or future income – they seem less willing to believe what a sophisticated computer algorithm is telling them. They prefer to use their own judgement, even if there is clear evidence that the computer’s predictions are more accurate.
Research I carried out revealed that managers in some companies overruled up to 90% of computer-based sales forecasts in favour of their own gut feel. They did this despite needing to make forecasts for hundreds of products each week.
Some of these interventions were probably made for political reasons (higher sales forecasts might please the boss) and others simply to justify the manager’s role. But there is evidence that people generally believe that their own judgements are more accurate than computer forecasts.
In an Australian experiment, participants persisted in relying on their own judgement even when the computer flashed up messages such as: “Please be aware that you are 18.1% LESS ACCURATE than the forecasts provided to you”.
This lack of trust in computers appears to extend to personal financial predictions. In a study I conducted in Turkey, we provided two groups of participants with forecasts of share prices on the Istanbul stock exchange. Both groups received the same forecasts but we told one group that they came from a financial expert (the truth) and the other that a statistical computer algorithm had generated them.
We invited the participants to change the forecasts if they thought they could improve them. Those who thought a computer had produced the forecasts made significantly bigger changes.
So why don’t people trust computers when it comes to predictions? After all, we are prepared to trust them in so many other aspects of our lives. They are relied upon for getting our bank balance right, handling our emails or flying our planes. This is what I set out to explore in my book, Forewarned: A Sceptic’s Guide to Prediction .
I found that while people expect humans to make mistakes – to err is human after all – they have much higher expectations of computers. But this means that when those expectations are not met, the computers have further to fall. Any apparent errors made by the computer are magnified in our perceptions so they erode our trust. Humans on the other hand are repeatedly given the benefit of the doubt.
Things we want to forecast usually consist of systematic patterns, such as trends, which we can predict, and randomness, which we can’t.
For example, supermarkets know that sales of soup will rise in winter and decline in summer. But customers’ whims, changeable weather and a host of other factors mean that demand cannot be predicted exactly.
Because there is no point in trying to forecast the random element of sales, computer algorithms are designed to filter it out so they only forecast the underlying systematic pattern. As a result, their predictions will seldom be the same as actual sales. These kind of differences cause managers to doubt the computer’s competence.
The tendency to see illusory systematic patterns in random events exacerbates these doubts. As humans, we seem to be programmed to seek these patterns – in one study rats were found to handle randomness better than Yale university students.
Once we think we have spotted a pattern the computer has missed, our urge to overrule its subsequent forecasts becomes irresistible. Worse still, we are brilliant at inventing explanations for these random fluctuations – the new sales manager is “working wonders” or bad weather in China is depressing the price of a company’s share. So our confidence in the superiority of our judgement is reinforced.
Satnavs tend to be correct most of the time so the motorist can become complacent and inattentive. In contrast, computer forecasts appear to be (somewhat) wrong almost all of the time, so we become extra vigilant. But this vigilance is likely to be misplaced. The time and mental effort that goes into overriding forecasts usually has only one outcome – an inaccurate judgement replaces a relatively accurate computer prediction.