Reverse Inference by Knowing Neurons

The recent unveiling of the BRAIN Initiative promised a greater understanding of cognition through improved imaging techniques.  However, the biggest problem in neuroscience is not the limits of technology, but the way inferences are being drawn from that technology.

Reverse Inference by Knowing Neurons

The usual inference drawn from functional neuroimaging studies is that when a certain cognitive process occurs, a particular brain area is active.  However, on closer inspection, Poldrack noted an ‘epidemic of reasoning’ occurring, with researchers falling foul of the fallacy of reverse inference (2006).  This is a process of induction entailing reasoning backwards from the observed brain activity to a particular cognitive process not directly tested, but perhaps linked to the task used, drawing on other research implicating that brain area with that cognitive process.

Such inferences are thus weaker than ones drawn from a direct engagement of a cognitive process, and are not deductively valid, because the brain area in question will almost certainly have more than one function.  They are often used to interpret unexpected results, but Poldrack states that despite their low predictive power making them relatively uncertain, it has become common for a study’s conclusions to rest entirely on a reverse inference.  Aside from leading to unsteady conclusions, it may also influence the research that others conduct, wasting large amounts of time and funding while researchers test the wrong questions.

A spectacular recent example of the misuse of reverse inference is that of James Fallon, who used a brain scan to diagnose himself as a psychopath.  Fallon noticed that a scan of his own brain, taken during a study on dementia, had abnormally low activity in some frontal lobe areas.  Other research has implicated these areas in empathy and moral reasoning, which are allegedly abnormal in psychopathic individuals, leading Fallon to conclude that he too was a psychopath.  Nonetheless, research on dementia does not typically use empathy tasks, so it is unlikely that the reduced activity in Fallon’s frontal lobes reflected pathological empathic ability.  Fallon made a reverse inference – and a silly one at that.

However, despite this grim picture, Poldrack argues that reverse inferences can be used to generate hypotheses for future research, and there are ways to improve the confidence researchers can have in them.  Firstly, increasing the specificity of the brain region of interest will narrow down possible cognitive processes, because complex processes (such as empathy) may have been associated with niche areas within a larger structure (such as the frontal lobe) and not with others.  A further tactic discussed by Hutzler is to consider the nature of the task used, and only suggest cognitive process which are likely to be involved in that task (an approach Fallon certainly didn’t take) (2013).  This tactic technically reduces the risk of false alarms, because it reduces the number of alternative explanations, although there still may be several options.

So perhaps reverse inferences can be acceptable, but their blatant misuse is certainly something to watch out for if neuroscience is to truly advance further.


Written by Lexie Thorpe.



Hutzler F. (2013). Reverse inference is not a fallacy per se: Cognitive processes can be inferred from functional imaging data, NeuroImage, 84 1061-1069. DOI: 

Poldrack R. (2006). Can cognitive processes be inferred from neuroimaging data?, Trends in Cognitive Sciences, 10 (2) 59-63. DOI:

Stromberg, Joseph. “The Neuroscientist Who Discovered He Was A Psychopath.” SmithsonianMag. Smithsonian, 22 November 2013. Web. 10 January 2014. <>.

Image made by Kate Jones and adapted from Hutzler F. (2013).


  1. Hi Lexie, I agree with most of this, and I’m happen to see you saying it, but I have to call out the statement that “the biggest problem in neuroscience is not the limits of technology”. The biggest problem in neuroscience, in my opinion and that of many otherpeople, is our inability to record the activity of large numbers of individual neurons at the same time. Correcting the misuse of fMRI, or any other statistical abuse, is not going to solve that problem.

    Best regards, Bill

    1. Hi Bill,
      I agree that being able to record from bigger areas in finer detail would open up exciting possibilities in research, but being able to do that would not resolve the misuse of statistics or the fundamental issues in research design and interpretation that are prevalent now, any more that correcting misuse of MRI would enable greater technologies, as you rightly said. I guess we simply disagree on which should happen first!

      I think that although improved technology should, in principle, tell us more about the workings of the brain and cognitive processes, but the statistics are only as good as the research questions behind them, and a flashy machine can’t stop researchers from asking the wrong questions.


Leave a Reply