Survivorship bias shows how we can make poor decisions when we only look at the cases that reached the end. In communication and business, this happens when we only analyse clients won, proposals accepted or successful projects.

Perhaps Hitler would have thought of a plan B if he had known what Wald was going to discover later.

The son of a baker and the grandson of a rabbi, Abraham Wald studied mathematics in Vienna, where he worked as a researcher. With the German occupation and a new Nazi director, emigration became inevitable.

In New York, he joined a secret statistical research group that contributed to the war effort. The group’s contribution was made through “equations”.

The problem of survivorship bias

That day, they received a request from the military. They needed to armour aircraft against enemy fighters. They thought a mathematical equation could find the right point.

They were shown aircraft that had returned from missions, covered in bullet holes in some areas. The distribution of those holes was not uniform. There were many holes in the wings and almost none in the engine. The logic was, therefore, to add more armour to the area that had been hit, right?

Armour makes aircraft heavier. Heavy aircraft are harder to fly and use more fuel.

Too much armour is bad. Too little armour is also bad. How could they make the best decision??

Abraham Wald’s solution

Wald disagreed. The reinforcement used to arm the aircraft should be placed where there are no bullet holes.

“Where were the missing bullet holes? They were in the missing aircraft.”

The reason why aircraft returned with a few bullet holes in the engine was simple: those that had been badly hit in the engine did not return.

So they reinforced the armour around the engine. How did Wald see what others did not?

Because the others started from an assumption: that the aircraft that returned were a random sample of all those that had carried out missions. For a mathematician, the bullet hole problem is seen as a phenomenon called “survivorship bias” – a distorted view in the observation and analysis of the facts identified.

“How to apply survivorship bias to communication

When we want to develop a project, a brand or a service, it is common to study the profile we are attracting, the audience we have won over. But what about studying those who do not stay? What information can they give us?

How many ideas were put aside because they did not work straight away? What criteria did we use to conclude that they had not worked?

It is important to look at what goes well, to focus on the clients we attract and on the projects where we have been successful. And it is equally important to ask the questions that help us see our own survivorship bias.

Whenever I send a proposal, I ask for a reply. It may come some time later, but I need to know whether a decision has already been made and, if another option was chosen, what the main reason was. I manage to get one 90% of the time. Every reply helps me understand where I stand and where I need to reinforce the armour. This requires organisation, resilience and a lot of focus. It is easy to leave it forgotten at the bottom of everything else.

Even in the past, during recruitment processes, I always asked for feedback, to understand whether the choice had been financial, based on experience, training or something else. What I received was always useful to help me understand where I stood and guide the strategy I had.

It is common to look at what worked. In phases like the one we are living through, it is important to understand whether, with a different kind of “armour”, other things could have worked too.

What were you able to learn from what did not work? Are you focusing on the right people? Is the strategy aimed at you or at the audience you want to reach? It may seem clear to you, but is the message just as easy for others to understand?

Questions to identify survivorship bias

Before concluding that an idea does not work, it is worth asking: what data was left out? Which people did not reply? Which proposals did not move forward? Which clients did not return?

Survivorship bias is not only useful for analysing aircraft, numbers or statistics. It also helps us understand where communication may be failing and which signals remain invisible because they never reach us.

Now that many of the things that “worked” can no longer “work”, what are you looking at?

Sofia Almeida Bernardo

In the context of the 4th Industrial Revolution, entertainment, education and work itself will be completely different, says the author of the book Digital Minds: The Science Redefining Humanity. According to Arlindo Oliveira, this is a reality for which small companies are better prepared than large ones, and where a major investment in training will be necessary.

“Emotions will eventually become a natural development of computational systems, and there will be systems with emotions,” says Arlindo Oliveira.

What do computers, cells and robots have in common?

All the complexity that exists in today’s world came, for millions of years, purely from biological systems and cells.

The world we see now is about 200 years old and would be completely strange to someone who had lived 500 years ago. Cars, televisions, aeroplanes and telephones would have been difficult to imagine.

We are used to the reality created by the Industrial Revolution. The 3rd Industrial Revolution brought computers, information technology, mobile phones and new ways of interacting with the world.

The 4th Industrial Revolution will be the moment when computers begin to behave intelligently. It is likely to bring even deeper changes.

What will this 4th Industrial Revolution be like?

It will happen when we talk to computers as we talk to people, and computers begin to take care of parts of our lives for us, such as shopping or tidying the house.

When the Industrial Revolution happened and we began to build skyscrapers, railway lines, airports and aeroplanes, we changed the nature of the planet.

Now, with these 3rd and 4th Industrial Revolutions, we are going to make an even deeper change.

What examples do you see of technology supporting humanity’s evolution to the next stage?

That is a longer-term question. This environment, in which people are constantly interconnected through their mobile phones, is something that will become increasingly intense.

It will increasingly affect every area of activity, from education to leisure, shows and entertainment. All these areas will change a great deal.

Technologies will probably evolve, although we cannot predict what they will become.

What is the role of emotions?

They are very important in the human species. They are not there by chance, but because they developed through the evolutionary process, creating a series of actions and reactions in human beings. Fear, for example, leads us to run away from dangerous things; love leads us to reproduce. Emotions evolved because they give human beings a competitive advantage. Some people say emotions are the most advanced form of intelligence because they are the hardest to automate.

“Although I am not anticipating that one of these ‘assistants’ might wake up one day in a bad mood and answer rudely, it seems reasonable to me that if we shout at it, it may react with surprise.”

Can they exist in these computational systems?

At the moment, we tend not to associate emotions with computational systems. However, there are already many systems capable of detecting people’s emotions.

As we seek to improve the interface of these systems, it is natural that they may be able to have basic reactions and emotions. For example: “I can see that you are tired” or “I can see that you are upset”.

The second step is to empathise with those emotions. I am not anticipating that one of these assistants might wake up one day in a bad mood and answer rudely. That does not seem reasonable to me. But it does seem reasonable that, if we shout at it, it may react with surprise. That would make the interaction more natural.

It would be strange to have a system that, for almost all practical purposes, is human, but then does not get upset, does not get angry and always answers in the same tone. Emotions will eventually also exist in these systems.

How will they be programmed?

Current systems learn through interaction, rather than being programmed in detail, and many may be trained to display emotions that make human interaction easier.

Very often, emotions are on the side of the person perceiving the emotion. For example: animals do not speak, we do not know what they think, but it is very easy to attribute emotions to them – angry, hungry.

Although emotions are perceived by the person observing them, experiments show that robots can be trained to relate to humans in a way similar to animals, leading people to attribute emotions to robots and empathise with them. It is believed that these systems will have emotions that are more controlled and less variable than those of human beings.

Will we be partners or rivals of computational systems?

I hope we will be partners. It makes no sense for us to be rivals. If we reach that situation, it means we have done something wrong. But it is not impossible that, when designing systems, we give them goals that conflict with our own. For example, we might ask a system to prevent global warming, but the best way to prevent it may be for us to walk everywhere and, in that case, there may be a conflict of goals. I hope that, when very complex systems are designed, it is guaranteed that humanity’s interests always come first.

What examples do you see of this alignment of goals?

In autonomous cars, the highest priority must be the protection of human lives, although choices will still have to be made. That is where we need to guarantee alignment. Imagine a car worth 100 thousand euros, programmed to avoid destroying itself. It may have to decide between destroying itself in a ravine or running someone over. Those decisions have to be made.

These problems also exist when we are drivers, but we normally do not have time, and our reactions are instinctive. Computers, because they are much faster, will almost always be able to decide and make the choice of whether or not to run someone over. We need to guarantee that the decisions made by the computer are aligned with human interests and values.

Why does this topic matter to companies?

The evolution of computers and artificial intelligence is not just a technological topic. It also changes the way we work, communicate, learn and make decisions. For companies, this reinforces the importance of training, adaptation and clear communication in contexts of change.

Excerpt from an interview published in ISQ Tecnologia & Qualidade 5 magazine, in January 2018.

Sofia Almeida Bernardo