The Grand Locus / Life for statistical sciences



When scientific models fail

Scientific models are more of an art than a science. It is much easier to recognize a good scientific model than to make one of our own. Like for an art, the best way to learn is to look at the work from the masters and take inspiration from them. One of the crown jewels of modern science is undoubtedly Darwin’s Theory of Evolution. I recently realized that I had no idea how Darwin stood against creationism and how he defended his view in regard of the doxa of his time. Digging into this topic turned out to be one the most important lessons I learned about the scientific method... and the lack of it.

Darwin touches vividly upon creationism at the end of “On the Origin of Species”. In his own words, he claims that

It is so easy to hide our ignorance under such expressions as the ‘plan of creation’, ‘unity of design’, etc, and to think that we give an explanation when we only restate a fact.

What strikes me here is that he does not accuse the ‘plan of creation’ and the ‘unity of design’ of not being proper scientific concepts. The real problem according to him is that they do not explain anything. Even more striking is that he empathizes with the illusion of explanation, as if he often found this feeling in himself. In other words, it seems that he considered that the most important feature of a scientific model is that it explains something. Accuracy, credibility and falsifiability come later.

Human after all

Sadly, the illusion of explanation is very frequent among scientists. Most commonly, it originates from an implicit bias that we all share. Fritz Heider and Marianne Simmel, the two psychologists who discovered it, were interested in understanding how we attribute intentions and personality to other entities. Why do we think that a human has intentions but a stone does not? At the time, in 1944, they thought that facial expressions were the key, but they discovered to their surprise that we can attribute intentions and even emotions, based on movements alone.

They demonstrated this using the video below. The test subjects were simply asked to describe what they see as precisely as they could. I recommend watching the video, so that you can experience it for yourself.

Almost all the participants described the scene as a “fight”, using words like “scared” for the circle and “bully” for the large triangle. Nobody used factual descriptions like “triangles moving parallel to a line” etc. If you have watched the video, you probably realized that this interpretation comes naturally, and that we immediately forget that these are just circles and triangles that cannot possibly have intentions. What Heider and Simmel discovered is that we can attribute intentions to almost everything, without even noticing that it makes no sense.

In 1965 another duo, Herbert Jenkins and William Ward published a famous study where they investigated how much control we think we have over a situation. In their experiments, they asked the test subjects to press one of two buttons in order to influence a dual outcome (“score” versus “no score”). At the end of the session, the subjects were asked how much control they think they had. Jenkins and Ward summarize their results by saying that

The main finding of these experiments is that the amount of judged control was a function of the frequency of successful outcomes rather than of the actual dependency of outcomes upon responses.

Without technical jargon: subjects were unable to tell whether they had any control because the only thing they paid attention to was the amount of successes. This study is sometimes used as a lesson of life coaching to emphasize that we have less control than we think. People somehow miss the fact that Jenkins and Ward also included spectator subjects who had to evaluate how much control the “actors” had. Their answers were exactly the same. In other words it’s not just about us, it’s about everybody.

We see intentions everywhere, even in triangles and circles. We see causes behind every intention, even when there isn’t any. Coming back to the ‘plan of creation’, it is hard not to see this bias at work. Things are the way they are because there is a plan. And this is a good explanation because intentions are valid causes.

What? A demon? Are you kidding?

It’s easy to bash against creationists, but the best examples of the illusion of explanation are in our own home. One of the most striking examples is Maxwell’s demon. In a nutshell, James Clerk Maxwell proposed a thought experiment showing how the second principle of thermodynamics may be violated. The idea was that a “demon” small enough to see the molecules could open and close a tiny door between two chambers A and B containing a gas initially at the same temperature. Then this

(...) being, who can see the individual molecules, opens and closes this hole, so as to allow only the swifter molecules to pass from A to B, and only the slower molecules to pass from B to A. He will thus, without expenditure of work, raise the temperature of B and lower that of A.

The discussion that ensued was very important for the foundation of thermodynamics. Physicists put some serious efforts into solving the paradox and made some important progress in doing so. For instance, taking into account the entropy of the demon, some showed that the temperature in the compartments A and B would not change. Some others mentioned that the demon would need to spend energy to measure the speed of the molecules, so the temperature would not change.

Systematically, progress on this question was achieved by dehumanizing the demon. What fuel does he use to open the door? What does it mean to “see” the molecules? How does he couple perception and movement? The original problem is a paradox only because we implicitly accept that there may exist a tiny entity with intentions, that perceives its environment the way we do, and that moves the way we do. But the cracks in the argument show up as soon as one tries to construct the demon from the laws of physics.

Any non-human scientist would probably not see the point. It is surprising that “are you fucking kidding?” was never considered a good argument against Maxwell’s demon. And it is even more surprising that we have learned so much from seriously worrying about this problem. Our human biases mask the obvious, but because we all share them, it is worth to go with them for a while, just to see how they prevent us from seeing clearly.


Good scientific models provide good explanations. We humans have a huge bias towards interpreting causes as intentions and vice versa. Many models just comfort us in this illusion and provide a feeling of explanation. The best way to fight this natural bias is to construct the actors of our models and think deeply about how they can be implemented.

« | »

blog comments powered by Disqus

the Blog
Best of

the Lab
The team
Research lines

Blog roll
Simply Stats
Ivory Idyll
Bits of DNA