Daniel Lemire's blog

, 8 min read

Don´t let the experts define science!

We know more than we can tell. We all know what a democracy is… We know that France, Canada, the USA, Japan… are democracies… Russia and China are not democracies. However, I have never seen a definition of “democracy” that fits the actual use of the term. One formal definition is “government by the people, exercised either directly or through elected representatives”. But, of course, both China and Russia have elected representatives. And they will certainly claim to be governments “by the people and for the people”. Formal definitions are less useful than you think.

Science is another term that defies formal definitions.

The philosopher Karl Popper offered us the concept of falsifiability to distinguish scientific statements. Something is falsifiable if it could be proven false. For example, I can say that the Moon was constructed the god Zeus. Could I test this statement and prove it wrong? Probably not. Thus it is not a scientific statement.

Notice that falsifiability is entirely general and has nothing with the natural sciences. For example, I can tell you that the unemployment rate in the US will skyrocket to 20% in 2018. That’s a falsifiable prediction. What is not falsifiable is to say that “because we elected Donald Trump to the presidency, the unemployment rate will skyrocket…”.

So falsifiability is a useful ingredient of science… but it does not define science, not anymore that elections define democracy. For example, there are highly paid financial analysists whose job is to predict economic indicators for the coming years. Their predictions are thoroughly falsifiable, but we do not consider them to be scientists.

People often say that science relies on the so-called scientific method. You start with a hypothesis and then you try to falsify it. There is no denying that it does happen in science… but that’s not what scientists “do”. What most scientists do is to try things out, without a clearly stated hypothesis… then they look at the results… if they find something that they can’t explain… they design more experiments to try to deepen their understanding. The major fault of the “scientific method” is that it assumes too much on the part of the scientists. A clearly stated hypothesis is often the end result, not the starting point.

Let me take an example. Maybe I think that coffee causes cancer. You’d think that it might serve as my hypothesis, but as stated it is not falsifiable. What do I mean by “cause cancer”? I certainly do not mean that drinking coffee will lead you to have a stomach cancer the next day. So I will start to collect statistics about coffee and cancer. Maybe I will setup experiments with mice that somehow drink coffee. Eventually, after years, I might end up with a hypothesis that others can try to falsify. Maybe. So though the concept of falsifiability is central to the sciences, the scientific method is not.

Another relevant scholar was Thomas Kuhn. Kuhn believed that scientific truth is not an objective fact. We have a collection of beliefs that are proving useful and that are robust enough to fit the facts as we see them. So far so good… but then Kuhn added a dichotomy… there is “normal” science where people do boring stuff under the existing belief system… and then there is “revolutionary” science that seeks to replace the existing belief system.

So you had silly Newtonian physicists… and Einstein appeared, threw everything overboard and forced people to rethink their long-held assumptions. Kuhn’s view is compatible with the Heroic Theory of Scientific Development. Scientists collectively stumble in the dark until a hero comes in to save the day.

Kuhn’s revolutionary science makes a good story… but I don’t think that’s how science works. In retrospect, we can tell stories about what happened and how it happened… but that’s not nearly as useful as it sounds.

If really science followed two distinct models, then you’d be able to tell, as it happens, whether this scientist pursues “normal science” whereas this other scientist pursues “revolutionary science”. But life isn’t quite so simple. You see… the reason the Heroic Theory of Scientific Development is false is that “ideas have people” (as opposed to people having ideas)… that is, ideas evolve and spread on their own… As long as the fertile ground is available, strong ideas thrive… and it is only in retrospect, in an illusionary manner, that we pin down when exactly an idea came to be.

Why did most scientific progress come from Europe and North America? It is not because white men are smarter. It is because of the European culture, maybe through Judeo-Christianity, created fertile ground for strong new ideas. That Kuhn was wrong matters because… because if he were right then you could turn around science funding and simply identify the “revolutionary science” and support that instead of wasting time with “normal science”. But, of course, if there was some way to tell apart “normal science” from “revolutionary science”, we would know.

Let us take a recent computer science “breakthrough”: deep learning. Companies like Google use deep learning to translate documents, recognize the content of pictures, speech recognition and so forth. It is the fastest area of growth right now in computer science (or it must be). We have gone back in time and granted great scientists like Yann Lecun the label of “science revolutionary” because deep learning took hold of them some time ago. (And, to be clear, people like Lecun should be celebrated.) But what actually happened?

  • We have had the idea of neural networks since… nearly forever. Though I was far from anything having to do with machine learning, I kept ending up in conferences covering neural networks in the last few decades. The field has grown and there have been lots of interesting theoretical developments… But it can easily be qualified as a whole lot of “normal science”.
  • Our computing hardware and software are a lot better, a lot faster than they were 20 years ago. What would have required a supercomputer worth billions in the 1990s can fit on the desk of any software engineer today. How did we get there? Through a whole lot of “normal science” and “normal engineering”.
  • Companies like Google have a lot more data than any company ever had. It is hard to think clearly about it, but we are talking about orders of magnitude in difference. How did we get there? Again, lots of patient, “normal work”.

So the dichotomy between normal and revolutionary science is about as scientific as the study of history… which is to say, not at all.

If the subversion of beliefs is to be qualified as “revolutionary”, then we may as well give up on science. Science needs the freedom to put in question our beliefs at every step. If you are not constantly rethinking what you think you know, you are not doing science.

There is, however, a useful dichotomy between non-scientific and scientific work. Lots of government-supported scientists, including some celebrated ones, do work that should never be qualified as “science”. Also, most scientists are guilty, at one point or another, of being lacking in ambition. That is, they pursue work that has little chance of being proven useful. But I would not call it “normal science”, I would call it “boring science”.

Most science out there is boring. The interesting parts are done by a small minority of researchers. But ambitious science is not the same as Kuhn’s revolutionary science. There is no need to work on a paradigm shift. For example, you can be a cancer scientist. On paper, your goal is to cure cancer. You can take the easy path and study one of the millions of mechanisms that have some relevance to cancer. You can then publish an endless stream of papers. Or you can set yourself as a goal to actually cure cancers. And then you can aim for clinical trials and so forth. Thankfully, if you know where to look, science is still super exciting. But it tends to happen only on a fertile ground… of which we have a limited supply.

What is fertile ground for science? Feynman described science as the belief in the ignorance of experts. What does it mean? It means that science can only thrive where experts can be questioned. Prior to the emergence of science (in Europe), truth was mostly whatever your ancestors and the highest figure of authority said it was. And that’s not a bad way to go about it. There is a lot of collected wisdom in your culture. But relying on your culture to determine truth is not science.

That’s why people who say that something is true because most scientists think it is true (e.g., with respect to climate change) have little understanding of what science really is. Scientific truth is not established by a vote among scientists. It does not matter how nice and popular your idea is… if it does not fit the facts on the ground, then it must be rejected. For example, biologists taught us that human beings have 24 pairs of chromosomes, up until the 1950s, even though it was visible that there were only 23.

If you live in a culture where questioning the established truth will get you in trouble, then you are not on a scientific fertile ground. So you need freedom (including freedom of speech).

Freedom of speech is a delicate thing, but an integral component of science. When we silence someone, it often only harms the few at first while it can greatly benefit the rich and the powerful. But if we could not stand up and speak up to contradict the authorities, we could not pursue science for long. To sum it up: I don’t think we can define formally what science is. It is a useful cultural construct that emerged out of Europe. It entails falsifiable ideas, and people working in a culture where it is possible to question and subvert widely held beliefs.