I recently had the pleasure of giving a lecture on mirror neurons at UC San Diego which is a very active locale for folks working on the human mirror system. I expected a lot of push-back on my critical views of mirror neurons, and I wasn't disappointed.
One of my major points of emphasis is and has been that if the mirror neuron system is really important for action understanding, then damage to action execution should result in action understanding deficits. I have pointed out that this prediction doesn't hold, either in apraxia or with more force in aphasia.
A typical response to this argument is that "the mirror system" involves lots of areas working together (George Lakoff, who was at the UCSD lecture even seemed willing to include the STS in this network) and so it is no surprise that damage to fronto-parietal areas doesn't result in the expected action understanding deficits. Along these lines, some folks pointed out that mirror responses have now been found in lots of brain areas, not just the frontal and parietal areas where they have been documented in monkeys. In other words, the mirror system is expanding.
At this point in the talk, Pat Churchland, who was my host, jumped in and said (and I paraphrase here), "Now wait a minute. If mirror neurons are all over the brain then don't they lose their explanatory power? Aren't we now just back to our old friend, the How Does the Brain Work Problem?" Very true, I think.
I ran into this same issue in my debate with Fadiga at the Neurobiology of Language Conference in Chicago. I said, here are several examples of preserved speech perception in the face of an absence of speech production ability, and the mirror neuron proponents said basically, it's a large system that's too complicated to succumb to the loss of the motor system.
A new review by Rizzolatti and Sinigaglia (2010) also pushes back against some of the critiques that have been raised lately. One concerns the observation that mirror responses in a TMS paradigm can be re-trained such that they no longer mirror and dissociated from understanding. RS: "The reason is that, in the task, there was nothing to understand: the investigated movements were meaningless." Another is the "surprising" claim (mine) that damage to the fronto-parietal motor system should be associated with deficits in action understanding. RS:" As clearly shown by electrophysiological mapping, there are motor sectors in the monkey inferior parietal lobule (and even in area PFG) with and without mirror neurons. Thus, dissociations between motor deficits and action understanding deficits can and do occur." Wait, I thought the mirror system got is magical powers from being part of the motor system. If the motor system is functionally disrupted, shouldn't the action understanding system also be disrupted? Or is there a parallel mirror motor system that can't control movement but can understand action (kind of sounds like a sensory system to me).
I think the mirror neuron folks have a serious problem on their hands: there is apparently no empirical result that can falsify the theory. If a mirror neuron shows up in an unexpected place, it is a new part of the mirror system. If a mirror neuron's activity dissociates from action understanding, it was not coding understanding at that moment. If damage to the motor system doesn't disrupt understanding, it is because that part of the motor system isn't mirroring.
Can someone from the mirror neuron camp come forward and provide us with an example of what kind of empirical result would falsify the theory? Because if you can't falsify it, it's no longer a scientific theory, it's religion.
Rizzolatti G, & Sinigaglia C (2010). The functional role of the parieto-frontal mirror circuit: interpretations and misinterpretations. Nature reviews. Neuroscience PMID: 20216547
You say in your blog; "Because if you can't falsify it, it's no longer a scientific theory, it's religion". This point of view is pretty common amongst people working in science, but if you don't mind, I'd like to take the chance to spiral off here into an argument against the position that the "falsifiability" of a theory is any sort of metric of its scientific importance. In fact, I'd argue that a straightforward falsification of any theory is a logical impossibility.
The argument for falsificationism rests upon two crucial assumptions;
(1) That it is possible to obtain ‘secure’, infallible statements – i.e. that we are able to obtain some veridical measure of the state of something by observing or measuring it in some way.
(2) That these statements (or measurements) are theory-independent – i.e. that the reliability of these measurements do not rest upon additional auxiliary assumptions about the presence or absence of confounding factors.
If (and only if) these assumptions are met, can we falsify a theory. In this case, if we discover observations that run counter to predictions made by a particular theory, we can safely reject the theory as false. If, however, the above assumptions are not met, then when we discover contrary observations we have a choice - we can either; reject the theory, reject the observation as being false, or blame the contrary observation on there being a problem with one or more of the auxiliary assumptions. In such a case, there is absolutely no a priori logical basis for selecting any of these alternatives over the others, the decision will depend upon the circumstances of the case under consideration.
(continues in next post)
It should be immediately obvious that both of the above assumptions of falsificationism are invalid. This is primarily because the accuracy with which we can observe something depends upon the methods of observation available to us at the time of observation, and because there is no way of measuring anything that does not rest upon a whole raft of auxiliary assumptions.
Observation statements are always made in the context of some theory or other, and are only as precise or accurate as is the theoretical framework under which they are made. A few years spent working in brain imaging should usually be enough to convince anyone of this, but it’s also clear if we look at examples from the history of science. Copernican theory, for example, was ‘falsified’ before it was even proposed, because under a Copernican system, the relative size of the planet Venus, as viewed from Earth, should change during the year as its orbit took it closer to, and further away from the observer. With the naked eye (the best technology available at the time) this prediction was shown to be false. It was only with the invention of the telescope that we were able to show that the prediction was supported. What was actually false was the auxiliary theoretical assumption that the naked eye gives an accurate measure of small point-light sources.
I’m not arguing here just that we need better, more accurate measuring devices, I’m saying that if you find an observation that appears to contradict a theory, then it’s logically valid to either find possible flaws with the observation (your machine doesn’t have the correct spatial resolution), or to modify some of the theory’s auxiliary hypotheses (mirror neurons aren’t just in motor cortex, they’re found all over the brain).
It is impossible to truly falsify any theory. To attack a theory for being unfalsifiable, then, is to miss the point. To my mind, what is really of importance for scientific theories is their predictive fertility and their explanatory power. This is really the angle to use when evaluating a theory. Rather than ask “Does it make predictions that can be falsified” (to which the simple answer is always ‘No’), we instead should be asking “How many predictions can you derive from this theory about the way the world works, and how well are these predictions supported by the evidence?”.
In the case of the mirror neuron theory of action understanding, your answer to this question would probably be “Not many, and not very well”. However, of crucial importance here is the idea that there is no absolute ‘truth’ in science – theories can only be judged in relation to other, competing theories.
Evolutionary theorists have never been able to “provide us with an example of what kind of empirical result would falsify the theory”. Does this make Darwinism a religion? I don’t think so. We (generally) all accept the validity of the concept of natural selection because it explains the data better than any other competing theory. As a scientist, I hope I’d be ready to abandon it if something better came along.
What are the competing theories of the neurobiological basis of action understanding? Are any of them better models of the available data? Do they make more, or deeper, predictions about how the brain might work? I have no idea, but maybe the mirror-neuron thing is the best current theory of how the brain does action understanding at the moment. If so, and even if it’s not very good, you’ll need to come up with something better to convince people that it’s not true.
Man, it would have been so much faster to just say "here, look at this page"
Even if the Q-D Thesis is correct (and it certainly seems like it should be), all empirical science proceeds by assuming isolability on various scales (e.g. "neither the current state of the LHC, nor this patient's slightly enlarged hippocampus should affect the outcome of my TMS experiment"), which means that all hypotheses derived from empirical work (and probably most "strictly theoretical"---whatever that means---work) come with great, big, (usually, but not always) implicit ceteris paribus clauses. Of course, going back and checking these assumptions now and then is wise (cf. H&P on task effects in various neuroling results).
But the presumption of isolability means that we scientists have been acknowledging the existence of Quine's "web of science" all along. Basically I think Popper's claims about falsifiability should also be read as being about the ceteris paribus versions of whatever hypotheses are under consideration.
As for "truth" in science, if there's no such thing, then the corollary is that we should all be instrumentalists, and while that can be a useful position in the early stages of a research program, it feels like a kind of "weak tea" version of science. Larry Sklar's book Theory and Truth is good w.r.t. Truth (and whether Science can be seen as asymptoting toward it) and how science presumes isolability (in a physics context).
Of course, everything I just said might be wrong and/or based on a radical misunderstanding of tom's point.
I agree with you both, and of course you are right in the limit. But here we have two theories: the motor system is the basis for action understanding vs. the perceptual system is the basis for action understanding. What test can we devise using the accepted assumptions of the field that would lead us to prefer one over the other? If there is no result that mirror neuron theorists will accept as evidence against the hypothesis, then we have a scientific problem. And this is exactly the situation that has developed in the field.
Damage to the motor system doesn't cause action understanding deficits? No problem, there are "other means" by which action understanding can be achieve.
Mirror responses are found in the hippocampus? No problem, the hippocampus must be part of the motor system.
People are ASSUMING the hypothesis and then using logical arguments to explain the findings. This is backwards and is inhibiting progress.
To restate the points I was trying to make;
1) For the reasons given above, I think it's a mistake to criticise a theory on the grounds that it's 'unfalsifiable', as Greg does in this blog entry.
2) If you disagree with a theory, for whatever reason, it's just not good enough to pile up examples of instances where the theory doesn't seem to fit the data. As Greg has been telling us, this doesn't seem to work. It can just lead to proponents of the theory either criticising some the assumptions you've made in interpreting the evidence, or modifying some of non-core hypotheses associated with the theory in order to accomodate the evidence. Moving the goal posts like this might seem sneaky, but it is perfectly valid.
3) The only way Greg (or anyone else) will be able to win this argument is by displacing the mirror neuron theory of action understanding with a better one. This theory will probably have to provide a more parsimonious explanation of the existing evidence, and also make more exciting predictions about how the brain might 'do' action understanding.
I suspect it's the first of these points that you disagree with most strongly. I don't really buy your defence of Popper, but this probably isn't the right place for us to argue about that. In re your points about realism, I hope no-one out there really thinks that cognitive neuroscience is at a stage where it is 'asymptoting towards the truth'. It seems to me that we're in a pretty early stage of the research program. We don't yet even know exactly what it is we're measuring with our scanners.
I agree that it's now got to the point where the existence of mirror neurons has become a framework for research, rather than a theory. Scientifically speaking, I don't actually have a problem with this. If they exist, and are the basis for action understanding, then useful work is being, or will be done. If they don't exist, or aren't actually the basis of action understanding, then the journey down that road will come to a halt pretty quickly (and maybe some useful stuff will be picked up on the way).
Pragmatically speaking, I suppose it could be argued that when resources are limited, it might not be wise to fund lots of research projects that are grounded in a theory that does not yet have convincing evidence behind it.
Maybe my unfalsifiable critique was presented a bit out of context. Here's the context:
There is an alternative theory that fits the data better:
"Perhaps the activity of mirror neurons simply
reflects sensory–motor pairings" p. 1240 (Hickok, 2009, JoCN, 21:7, 1229–1243)
This explains why mirror neurons fire both during action execution and action observation and the fact that action execution and action understanding dissociate.
Given that there is a very simple explanation of mirror neuron behavior, the point of my Eight Problems paper and all of my critique posts here is that all of the evidence cited for the more complex claim in fact doesn't support the theory at all.
In this context, to modify the theory every time a new inconsistent result rolls in is scientifically problematic. They simply stipulated a theoretical conclusion (which has no empirical support) and interpret every finding in light of this stipulation. Nothing can falsify the theory because the theory is assumed a priori. This is not good science, and is the basis for my 'unfalsifiable theory' post. I feel like I'm arguing for evolution with a bunch of creationists.
I think Greg is saying that the mirror neuron theory is now extraordinarily un-parsimonious. It has become a tangle of Ptolemaic epicycles.
The cliche answer to "what would convince a scientist evolution is not true?" is "a rabbit skeleton in undisturbed pre-Cenozoic strata." Of course, that assumes that one can identify the skeleton, date the strata, and determine that it is undisturbed.
It might help a little if you could define what you mean by action understanding? Previously I have found these sort of discussions frustrating as people often mean different things when they say "action understanding".
That's another problem, James. Action understanding is not clearly defined by the mirror neuron folks (there are different loose definitions in different papers), nor is it ever measured. Check out page 1230 in my 8 Problems paper for discussion.
you will find this interesting. A theory only makes sense within a "school of though" which has its own lexicon. Falsifiability is only useful when that context exists. Other than that scientific progress is subject to the same evolutionary forces as anything else. Order out of nonorder:
So if there is confusion of what "action understanding" means, then when you say "..then damage to action execution should result in action understanding deficits. I have pointed out that this prediction doesn't hold" what is the deficit that is not seen that leads you to make this conclusion?
In the human literature it has been operationalized as gesture recognition deficits in apraxia or speech recognition deficits in aphasia.
but isn't there data that patients with ideomotor apraxia also have gesture recognition deficits?
Yes, but they also dissociate so you can have a movement execution disorder without the receptive component.
following up on what Paz said - surely you can't demand that every movement execution disorder has a receptive component? If you had damage to primary motor cortex, you would have trouble moving your fingers but not understanding actions.
Similarly, looking at your figure 1 in your 8 problems paper, you complain that there are participants in the lower left corner of the plot, who have poor action recognition without IFG damage. If these patients had damage to primary visual cortex (an extreme example), that could cause poor gesture recognition. I think if you want to claim that IFG damage is not related to action recognition, you need to find patients in the top right corner of that plot - with substantial IFG damage AND normal action recognition. That corner looks pretty empty to me.
Thank you for the link. I am determined to use the phrase 'Kant on wheels' as some point before I die. I would argue that the 'lexicon' of cognitive neuroscience is absurdly vague and inconsistent at the moment. The cognitive bits have been inherited from disparate branches of cognitive science/psychology/psychophysics, and the neuro bit is very poorly understood. There is not yet any real 'school of thought' with agreed assumptions, principles or definitions under which we can practice.
Yes, that's exactly the sort of argument I have sympathy with. My (very long-winded) point was that there's no need to drag falsification into it. The challenge is really to come up with a more parsimonious explanation that makes more and better predictions.
I bet if someone (Greg?) did find "a rabbit skeleton in undisturbed pre-Cenozoic strata", they would have a fight on their hands.
This is precisely what some theorists are claiming, that the knowledge of an action is encoded even in M1. Of course, I would not expect damage there to produce any kind of receptive deficit.
Notice now, though, that once you admit that low levels of the motor system aren't coding action knowledge, you are on a slippery slope. The basis of the mirror neuron claim is that understanding is achieved via motor simulation. If you can no longer simulate the movements because your motor system is damaged (at any level), don't you lose that knowledge? Response: no because you can simulate it at a more abstract level where you are not coding specific movements, but movement "goals". But now what is a goal? And how do you simulate it? It is simulated in the motor system? Or at some more abstract, conceptual, level?
Re: the figure you are correct to point out that there are not patients in the upper right corner. I have two responses to this observation. One is that the point of the figure was to illustrate that the claimed correlation doesn't really exists when you look at the actual data: the fraction of IFG involvement does not predict gesture recognition deficits. The second is that IFG, more anterior portions in particular, may be involved generally not because it is critical to gesture recognition, but to response selection. So called "cognitive control" functions have been linked to anterior portions of Broca's area.
Duhem-Quine, theory laden observations, error of observations etc, Have no bearing on what falsification means.
These are red herrings thrown out by justificationists/verificationists/inductivist philosophers who either don't understand Popper or to mislead people and scientists as to what Popper said and solved.
For instance, Popper handled what later was called the Duhem-Quine problem in the 1930's
To understand critical rationalism and all these and supposedly even more serious problems to falsification then read David Millers "Critical Rationalism" and his "Out of Error."
Dear Mr. Hickok, what do you think about this article: V. Kosonogov. Why the Mirror Neurons Cannot Support Action Understanding, Neurophysiology, 44, 2012?
It criticizes the mirror neuron theory of action understanding and proposes another role for these neurons.
Post a Comment