Strong inference and the distinction between soft and hard science, part II

Continuing our discussion of Platt’s classical paper on “strong inference” and, more broadly, the difference between soft and hard science, another reason for the difference between these two types of science mentioned but left unexamined by Platt is the relative complexity of the subject matters of different scientific disciplines. It seems to me trivially true that particle physics does in fact deal with the simplest objects in the entire universe: atoms and their constituents. At the opposite extreme, biology takes on the most complex things known to humanity: organisms made of billions of cells, and ecosystems whose properties are affected by tens of thousands of variables. In the middle we have a range of sciences dealing with the relatively simple (chemistry) or the slightly more complex (astronomy, geology), roughly on a continuum that parallels the popular perception of the divide between hard and soft disciplines. That is, a reasonable argument can in fact be made that, so to speak, physicists have been successful because they had it easy. This is of course by no means an attempt to downplay the spectacular progress of physics or chemistry, just to put it in a more reasonable perspective: if you are studying simple phenomena, are given loads of money to do it, and you are able to attract the brightest minds because they think that what you do is really cool, it would be astounding if you had not made dazzling progress!

Perhaps the most convincing piece of evidence in favor of a relationship between simplicity of the subject matter and success rate is provided by molecular biology, and in particular by its recent transition from a chemistry-like discipline to a more obviously biological one. Platt wrote his piece in 1964, merely eleven years after Watson, Crick and Franklin discovered the double helix structure of DNA. Other discoveries followed at a breath-taking pace, including the demonstration of how, from a chemical perspective, DNA replicates itself; the unraveling of the genetic code; the elucidation of many aspects of the intricate molecular machinery of the cell; and so on. But by the 1990s molecular biology began to move into the new phase of genomics, where high throughput instruments started churning a bewildering amount of data that had to be treated by statistical methods (one of the hallmarks of “soft” science). While early calls for the funding of the human genome project, for instance, made wildly optimistic claims about scientists soon being able to understand how to make a human being, cure cancer, and so on, we are in fact almost comically far from achieving those goals. The realization is beginning to dawn even on molecular biologists that the golden era of fast and sure progress may be over, and that we are now faced with unwieldy mountains of details about the biochemistry and physiology of living organisms that are very difficult to make sense of. In other words, we are witnessing the transformation of a hard science into a soft one!

Despite all of the reservations that I detailed above, let us proceed to tackle Platt’s main point: that the difference between hard and soft science is a matter of method, in particular what he refers to as “strong inference.” Inference, of course, is a general term for whenever we arrive at a (tentative) conclusion based on the available evidence concerning a particular problem or subject matter. If we are investigating a crime, for instance, we may infer who committed the murder from an analysis of fingerprints, weapon, motives, circumstances, etc. An inference can be weaker or stronger depending on how much evidence points to a particular conclusion rather than to another one, and also on the number of possible alternative solutions (if there are too many competing hypotheses the evidence may simply not be sufficient to discriminate among them, a situation that philosophers call the underdetermination of theories by the data). The term “strong inference” was used by Platt to indicate the following procedure:

1. Formulate a series of alternative hypotheses;
2. Set up a series of “crucial” experiments to test these hypotheses; ideally, each experiment should be able to rule out a particular hypothesis, if the hypothesis is in fact false;
3. Carry out the experiments in as clear-cut a manner as possible (to reduce ambiguities of interpretation of the results);
4. Eliminate the hypotheses that failed step (3) and go back to step (1) until you are left with the winner.

Or, as Sherlock Holmes famously put it in The Sign of Four, “when you have eliminated the impossible, whatever remains, however improbable, must be the truth.” Sounds simple enough. Why is it, then, that physicists can do it but ecologists or psychologist can’t get such a simple procedure right?

If Platt’s strong inference sounds familiar, it should: it is related to Francis Bacon’s method of induction, and Platt explicitly invokes the British philosopher in his article. The appeal of strong inference is that it is an extremely logical way of doing things: Platt envisions a logical decision tree, similar to those implemented in many computer programs, where each experiment tells us that one branch of the tree (one hypothesis) is to be discarded, until we arrive at the correct solution. For Platt, hard science works because its practitioners are well versed in strong inference, always busy pruning their logical trees; conversely, for some perverse reason scientists in the soft sciences stubbornly refuse to engage in such a successful practice, and as a consequence waste their careers disseminating bricks of knowledge in their courtyards, rather than building fantastical cathedrals of thought. There seems to be something obviously flawed with this picture: it is difficult to imagine that professionally trained scientists would not realize that they are going about their business in an entirely wrong fashion, and moreover that the solution is so simple that a high school student could easily understand and implement it. What is going on?

We can get a clue to the answer by examining Platt’s own examples of successful application of strong inference. For instance, from molecular biology, he mentions the discovery of the double helix structure of DNA, the hereditary material. Watson, Crick, Franklin and other people working on the problem (such as twice-Nobel laureate Linus Pauling, who actually came very close to beating the Watson-Crick team to the finishing line) were faced by a limited number of clear-cut alternatives: either DNA was made of two strands (as Watson and Crick thought, and as turned out to be the case) or three (as Pauling erroneously concluded). Even with such a simple choice, there really wasn’t any “crucial experiment” that settled the matter, but Watson and Crick had sufficient quantitative information from a variety of sources (chiefly Franklin’s crystallographic analyses) to eventually determine that the two-helix model was the winner. Another example from Platt’s article comes from high-energy physics, and deals with the question of whether fundamental particles always conserve a particular quantity called “parity.” The answer is yes or no, with no other possibilities, and a small number of experiments rapidly arrived at the solution: parity is not always conserved. Period. What these cases of success in the hard sciences have in common is that they really do lend themselves to a straightforward logical analysis: there is a limited number of options, and they are mutually exclusive. Just like logical trees work very well in classic Aristotelian logic (where the only values that can be attached to a proposition are True or False), so strong inference works well with a certain type of scientific question.

Yet, any logician knows very well that the realm of application of Aristotelian logic is rather limited, because many interesting questions do not admit of simple yes/no answers. Accordingly, modern logic has developed a variety of additional methods (for instance, modal logic) to deal with more nuanced situations that are typical of real-world problems. Similarly, the so-called soft sciences are concerned largely with complex issues that require more sophisticated, but often less clear cut, approaches; these approaches may be less satisfactory (but more realistic) than strong inference, in that they yield probabilistic (as opposed to qualitative) answers. Soft science, then, is soft because of very good reasons intrinsic in the nature of the object of study, certainly not because of the intellectual inferiority of its practitioners.