‘Robots are far better judges than humans’
Should computers make all our decisions? In a new book published today, three top scientists argue that human beings are useless at making choices. Machines would do a much better job.
Modern society is run by experts. Think about your own life: in school, you are taught by a professional educator who makes decisions about what you should learn. Perhaps you go to the doctor with an illness, and they decide what to prescribe you. And if it is a very bad day, you might find yourself pulled up in front of a judge, who determines whether you should walk free or end up in prison.
But this kind of system only works if we know that the professionals are making the right decisions. Noise, a new book by Daniel Kahneman, Olivier Sibony and Cass R Sunstein, suggests that it happens far less than we might like to believe.
The authors claim that professional decisions that are supposed to be objective are actually warped by tiny, unnoticeable influences in our environment, which they call noise.
For example, a doctor might give you a different diagnosis depending on what the weather was like on the day of your appointment, or whether they had eaten breakfast that morning – or had a good night’s sleep.
This noise has serious consequences for society. It means experts can end up making very different decisions in similar cases. We like to think that in a fair legal system, for instance, two people who are found guilty of the same crime will be given identical punishments. But history suggests that this is not the case.
In one study, in 1981, 208 judges were asked to hand down sentences in 16 hypothetical legal cases. They were only given the most basic details of the case so there was no room for bias.
Yet in only three of these cases did all the judges agree that a prison term was appropriate. And the average difference in prison time handed down by any two of these judges was a staggering 3.5 years.
This might not matter if we were more aware of how noise influences our decisions. But most professionals overestimate their ability to make accurate, fair judgements and so are unable to adjust their choices to account for it.
Kahneman, Sibony and Sunstein suggest it would ultimately be better to entrust all decisions to algorithms. Unlike human beings, computers are not affected by tiredness, weather, hunger or emotions.
That means their judgements are much less noisy. A computerised doctor could identify illnesses more accurately. An AI judge’s verdicts would be perfectly consistent.
But others are wary. They point out that while human experts do make mistakes, we can hold them to account for this by appealing a legal decision or disciplining a doctor for negligence. But a computer’s decision is supposedly objective – and therefore final.
And sometimes the human capacity to process environmental factors is important. A judge might feel compassion for a young offender and decide to give them a lenient sentence in the hope that this small act of kindness will persuade them to change their ways. An algorithm cannot show such empathy.
Should computers make all our decisions?
Yes, say some. Human beings are much too fallible to be trusted with important decisions. A person’s life can be ruined because a judge forgets to eat their lunch, or their doctor is going through a messy divorce or because the weather has turned cloudy. Using algorithms to make decisions is the only way of ensuring fairness, consistency, and accuracy in our everyday lives.
Not at all, say others. An algorithm is only as good as the person programming it, and that person is just as susceptible to noise as any other expert. Besides, human variability is a gift, not a curse: it is what makes us creative and critical beings. And it allows us to recognise failures and injustices in our systems and work towards changing them.
- Think about some of the things in your life that might count as noise, and consider how they affect the decisions you make.
- Why is it important for people to feel that experts in positions of authority are making consistent decisions?
- In pairs, take it in turns to draw a portrait of each other surrounded by things you think might count as noise in each other’s lives. Swap the pictures back at the end and compare them.
- In groups, write a short play about a machine that makes a wrong decision and begins to feel guilty about it. Act it out for the rest of the class.
Some People Say...
“In a properly automated and educated world, then, machines may prove to be the true humanizing influence.”Isaac Asimov (1920 – 1992), American writer and scientist
What do you think?
Q & A
- What do we know?
- Most people agree that human beings are impulsive creatures. History has often been made by poor decisions. Ancient Troy was destroyed when its people decided to trust their Greek enemies and bring the wooden horse into their city. In 1812, Napoleon Bonaparte launched an invasion of Russia: he lost more than half a million soldiers and was deposed two years later. In 1919, the USA, the UK and France decided to punish Germany with the Treaty of Versailles, which led indirectly to World War Two.
- What do we not know?
- There is some debate over whether or not AI can really be free of random external influences. Certainly, a computer programme can only work with the data that it has been given. However, researchers have found that this makes AI very rigid: it tends to assume that the data it has been given is a perfect representation of the world. As such, in order to train AI more effectively, they have to introduce some noise deliberately into the data that they give it.
- Daniel Kahneman
- An Israeli psychologist who wrote Thinking, Fast and Slow, which explored the relationship between quick, intuitive thought and slow, rational thought.
- Olivier Sibony
- A French writer and academic who specialises in the strategy of business decisions.
- Cass R Sunstein
- An American philosopher most famous for co-writing Nudge, which argued that human decisions are best influenced by light, non-invasive interventions, or nudges.
- Like noise, bias also affects people’s decisions, but in a different way. Noise refers to random background influences, whereas biases are beliefs that systematically distort a person’s decisions.
- At its most basic, an algorithm is a list of rules to be followed in order to solve a problem. They are used in computer programmes and, it is hoped, eventually AI that is capable of teaching itself.
- Failure to take proper care of something. Medical professionals have a duty of care to their patients. If they break this, they can be accused of negligence.
- Lighter or more merciful treatment than expected.