Neuroscience’s Inference Problem And The Perils Of Scientific Reduction

In Science’s Inference Problem: When Data Doesn’t Mean What We Think It Does, while reviewing Jerome Kagan‘s Five Constraints on Predicting Behavior, James Ryerson writes:

Perhaps the most difficult challenge Kagan describes is the mismatching of the respective concepts and terminologies of brain science and psychology. Because neuroscientists lack a “rich biological vocabulary” for the variety of brain states, they can be tempted to borrow correlates from psychology before they have shown there is in fact a correlation. On the psychology side, many concepts can be faddish or otherwise short-lived, which should make you skeptical that today’s psychological terms will “map neatly” onto information about the brain. If fMRI machines had been available a century ago, Kagan points out, we would have been searching for the neurological basis of Pavlov’s “freedom reflex” or Freud’s “oral stage” of development, no doubt in vain. Why should we be any more confident that today’s psychological concepts will prove any better at cutting nature at the joints?

In a review of Theory and Method in the Neurosciences (Peter K. Machamer, Rick Grush, Peter McLaughlin (eds), University of Pittsburgh Press, 2001), I made note¹ of related epistemological concerns:

When experiments are carried out, neuroscientists continue to run into problems. The level of experimental control available to practitioners in other sciences is simply not available to them, and the theorising that results often seems to be on shaky ground….The localisation techniques that are amongst the most common in neuroscience rely on experimental methods such as positron emission tomography (PET), functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and magnetoencephelography (MEG). [In PET] a radioactive tracer consisting of labelled water or glucose analogue molecules is injected into a subject, who is then asked to perform a cognitive task under controlled conditions. The tracer decays and emits positrons and gamma rays that increase the blood flow or glucose metabolism in an area of the brain. It is now assumed that this area is responsible for the cognitive function performed by the subject. The problem with this assumption, of course, is that the increased blood flow might occur in one area, and the relevant neural activity might occur in another, or in no particular area at all….this form of investigation, rather than pointing to the modularity and functional decomposability of the brain, merely assumes it.

The fundamental problem–implicit and explicit in Kagan’s book and my little note above–is the urge to ‘reduce’ psychology to neuroscience, to reduce mind to brain, to eliminate psychological explanations and language in favor of neuroscientific ones, which will introduce precise scientific language in place of imprecise psychological descriptions.  This urge to eliminate one level of explanation in favor of a ‘better, lower, more basic, more fundamental’ one is to put it bluntly, scientistic hubris, and the various challenges Kagan outlines in his book bear out the foolishness of this enterprise. It results in explanations and theories that rest on unstable foundations: optimistic correlations and glib assumptions are the least of it. Worst of all, it contributes to a blindness: what is visible at the level of psychology is not visible at the level of neuroscience. Knowledge should enlighten, not render us myopic.

Note: In Metascience, 11(1): March 2002.

How Best to Introduce Scientific Reasoning

A couple of days ago on Facebook, by way of crowd-sourcing syllabi preparation for an undergraduate critical thinking course that includes a unit–three to six class sessions–on scientific reasoning, David Grober-Morrow threw out the following query

What do you most wish that undergraduates (science and non-science majors) understood about scientific reasoning?

This is a very good question. I would suggest that for the indicated demographic and unit-length, the most valuable instruction would be in the centrality of ampliative inference in the practice of science. That is, students should learn that scientific reasoning, besides relying on deductive inference of the consequences of scientific laws, also utilizes, in large part, induction and abduction.

Both these forms of inferences ‘go beyond the data’; they enable the bridging of the gap between observations and the conclusions drawn on their basis. In inductive reasoning, the scientist infers statements about future observations after having made a finite set of observations of empirical phenomena. The classic ‘All ravens observed thus far are black, therefore, all ravens, even those unobserved at this point in time, are black’ formulation of this kind of reasoning leads to Nelson Goodman‘s famous riddle of induction; it is a form of prediction, an inference made about the future. In abductive reasoning, in making inferences to the best explanation, the scientist infers backwards, to the past, about the kinds of events that might/must have occurred to make true the observations recorded. A bridge has collapsed; what must have happened to have made this event occur? This might thus be termed postdiction.

Thus after the scientist has made observations at one point in time, using these forms of inference he is able to look backwards and forwards along the timeline.

Introducing students to these forms of inference leads quite naturally to an introduction and explanation of the centrality of probabilistic forms of reasoning in science, the nature of admissible and inadmissible evidence and the confirmations they permit, the formulation of scientific hypotheses and theories, and the nature of scientific laws. It also shows how deductive inference is a relatively minor part of scientific reasoning, one that follows on the heels of these two forms of inference.

It would be a mistake, I think, to introduce students, in a class like the one described above, to the ‘gruesome’ Goodman puzzle of induction. The fairly sophisticated concepts involved in its clearest explication and resolution are likely to be found confusing by the students in the limited time available. (It also has the unfortunate feature of seeming a bit like a parlor trick, a sure-fire method of turning off a student already convinced that philosophers’ examples are a kind of intellectual sandbaggery.) Instead, I would rely on as many colorful examples as possible to show how science is not the mere routine noting down of data in notebooks, how creative and inventive induction and abduction allow scientists to be, and how much of the impressive and awe-inspiring edifices of scientific knowledge are built on the seemingly tenuous foundations provided by these forms of reasoning.

Enrico Fermi, Abduction, and Slow Neutrons

In his acceptance speech for the Nobel Prize in Physics in 1938, Enrico Fermi spoke briefly and thoughtfully about the theoretical and experimental work which had earned him this honor.  His talk, ‘Artificial Radioactivity Produced by Neutron Bombardment,’ is a little gem of scientific writing, which showcases not only descriptions of the results of the groundbreaking work in atomic and nuclear physics he had engaged in, but scientific explanation as well.

I provide here a little extract to show Fermi demonstrating abduction–inference to the best explanation–in his recounting of the phenomena of ‘slow neutrons’:

The intensity of the activation as a function of the distance from the neutron source shows in some cases anomalies apparently dependent on the objects that surround the source. A careful investigation of these effects led to the unexpected result that surrounding both source and body to be activated with masses of paraffin, increases in some cases the intensity of activation by a very large factor (up to 100). A similar effect is produced by water, and in general by substances containing a large concentration of hydrogen. Substances not containing hydrogen show sometimes similar features, though extremely less pronounced.

The interpretation of these results was the following: The neutron and the proton having approximately the same mass, any elastic impact of a fast neutron against a proton initially at rest, gives rise to a partition of the available kinetic energy between neutron and proton; it can be shown that a neutron having an initial energy of 10^6 volts after about 20 impacts against hydrogen atoms has its energy already reduced  to a value close to that corresponding to thermal agitation. It follows that, when neutrons of high energy are shot by a source inside a large mass of paraffin or water, they very rapidly lose most of their energy and are transformed into ‘slow neutrons.’ Both theory and experiment show that certain types of neutron reactions…occur with a much larger cross section for slow neutrons than fast neutrons , thus accounting for the larger intensities of activation, observed when irradiation is performed inside a large mass of paraffin or water.

This explanation of experimental data–or ‘interpretation’ as Fermi terms it–is, I think, a particularly elegant one. It is concise both in its form and content; it does justice to the observations with very few claims on our credulity; it integrates the new into the old with a minimum of effort. It is dazzling too–as many explanations of that heady time in atomic and nuclear physics were–in the seeming sleight of hand it performs: it takes the broad, chunky, mundane details of macroscopic phenomena and reduces them to the minute interactions of invisible particles. It pulls off that trick that is so distinctive of so many memorable scientific explanations: such sparse data, such elaborate theory.

I first read of the phenomena that Fermi describes in my eleventh-grade physics textbook; it is only recently that I have read of them in Fermi’s own words. The explanations seemed elegant then, but their style is even more acute in Fermi’s formulation.

Quotes from: Emilio SegrèEnrico Fermi, Physicist, Appendix 2, University of Chicago Press, 1970, pp. 217-218.

Rock Arches, Geology and the Wonders of Abduction

During the Pennsylvanian period (300 to 320 million years ago), this area was part of the Paradox Basin, a giant inland sea that dried up intermittently, leaving behind a thick accumulation (5000 feet, 1525 meters) of layered marine salt. Loading by deposition of subsequent Permian through Triassic layers caused the ductile salt to flow to areas of lower pressure–principally along fault lines. Gradually, salt accumulation or salt walls up to 10,000 feet (3050m) thick, three miles (5 km) wide, and 70 miles (110 km) long developed beneath the northwest-trending Moab and Salt Valley faults. The area between sagged in response to withdrawal of the subsurface salt. Approximately 60 million years ago, and long after the salt walls had been emplaced, compressive forces associated with the  ongoing Late Cretaceous Laramide Orogeny warped the region into anticlinal folds….

And so goes the answer to the question ‘Why are there so many arches in Arches National Park, Utah’? (From Geology Unfolded: An Illustrated Guide to the Geology of Utah’s National Parks, Thomas H. Morris, Scott M. Ritter, Dallin P. Laycock, Brigham Young University Press, 2010)

No matter how many times I see it in effect, I remain in awe of the marvels of scientific abduction (post-facto inference to the best explanation).  A response like the one above sometimes resembles nothing so much as magic, a peering back into an  unimaginably distant time, revealing detail with unerring precision. And if it feels like there is some sleight of hand involved, then that is as it should be, for didn’t I just say it felt like magic? (As should be clear, I am convinced by this answer and would accept it had I asked the question. Come to think of it, I did. )

I’ve taught scientific explanation in several classes now: for instance, Philosophy of Science, Scientific Revolutions, and the Philosophy of Biology. Teaching it–sometimes in the context of cosmology, sometimes the theory of evolution–always lets me point out how such inferences, while seemingly pulled out of a hat, and thus subject to the polar extreme of creationist-style queries of ‘How do you know? Did you see it happen?’,  are in fact, eminently sensible in light of our previously acquired knowledge. An explanation works because it makes sense, and it will only make sense in light of how well it fits in with our previously held beliefs. Those beliefs, in turn, have been held and retained by us because of their accordance and sympathy with yet others.  And so on.  (It also helps to point out that such explanation is of the same variety as the kind that we indulge in on a daily basis. Very little of what we claim to believe meets the ‘seeing is believing’  test.)

Underlying the seemingly-magical, reverse-crystal-ball gazing nature of the explanation of the profusion of rock arches in a particular part of the southwestern United States then, is a whole body of scientific knowledge, acquired slowly, painstakingly, and carefully interlocked with other scientific disciplines: carbon dating, physical chemistry, materials science etc. It seems speculative to the uninitiated, but a closer look reveals that this particular house of cards has a very strong foundation indeed.  You could bring it crumbling down, but you’d have to work very hard.

Oh, and by the way, this is what one of those arches looks like (click for a larger image):