A SEEMING PARADOX: Learning is associated with brain growth, but also shrinkage
Image: ktsimage
-
The Wisdom of Psychopaths
In this engrossing journey into the lives of psychopaths and their infamously crafty behaviors, the renowned psychologist Kevin Dutton reveals that there is a...
Read More »
With age and enough experience, we all become connoisseurs of a sort. After years of hearing a favorite song, you might notice a subtle effect that’s lost on greener ears. Perhaps you’re a keen judge of character after a long stint working in sales. Or maybe you’re one of the supremely practiced few who tastes his money’s worth in a wine.
Whatever your hard-learned skill is, your ability to hear, see, feel, or taste with more nuance than a less practiced friend is written in your brain. But where, and how, exactly? What are the biological pen strokes that spell perceptual expertise?
One classical line of work has tackled these questions by mapping out changes in brain organization following intense and prolonged sensory experience. In rough overview, many of these studies support a model of learning that might be in line with your intuition. Namely, the parts of the brain allotted for discrete sensory skills - hearing the note middle C, feeling a piano key on your thumb tip - expand when those skills are repeatedly called upon. Or, shamelessly dispensing with the biological details: practice makes bigger, and bigger means better.
But don’t adopt that slogan quite yet. In a recent study from the University of Texas at Dallas, Dr. Michael Kilgard’s lab questions the tidy relationship between altered size and enhanced skill. Studying the auditory cortex of rats, they found that the expansion of a ‘skill-specific’ brain area with training is only short lived, even when changes in ability are long lasting. Instead of working like a muscle, where training adds size and size begets prowess, learning seems to involve some heavy duty trimming as well. In fact, if Kilgard’s theory of learning holds up, both the biology of learning and our experience of it share a common principle: skill must be culled from a string of mistakes. Lots of them.
If you were to look at the side of someone’s brain, focusing on the thin sliver of auditory cortex, it would seem fairly uniform, with only a few blood vessels to provide some bearing. Functionally though, it’s more like a patchwork of distinct neural territories, each of which ‘hears’ only a limited range of sound frequencies. Imagining for a minute that the auditory cortex is the United States, low frequency tones would be preferentially processed in California, high tones in New York, and intermediate tones in between.
One of the great neuroscience findings of the past several decades is that the ‘state lines’ of the auditory map (as well as many other sensory maps) are redrawn after training. Work from Dr. Michael Merzenich’s lab in particular showed that if monkeys were trained to make difficult sound discriminations - say, between two very similar low frequency tones - then low frequency regions of the auditory cortical map became enlarged (California would get a bit bigger, and encroach onto what was formerly Nevada). A host of other studies built on this basic idea, showing that blocking cortical expansion also blocks learning, and that more expansion often correlates with better learning. Cortical expansion and skill learning seemed to be deeply intertwined.
And yet, some aspects of this theory invited skepticism. Does learning really require such large-scale remodeling of our cortex? How do we maintain a large number of skills if there’s rather limited space to store them? Wouldn’t new learning start to overwrite old learning after a time?
To address these problems, Kilgard and his team used rats in a new test of the dictum that size matters. Instead of altering the auditory cortex map through training - the already proven route - the researchers sought to refashion it directly, if somewhat artificially. By doing this, they could isolate the importance of map size in learning: if you simply make a sensory map bigger, what does it buy you in terms of performance?
Directly ‘writing in’ changes to the auditory cortex is of course challenging, but the researchers had a trick. Their approach was to electrically stimulate a brain region (nucleus basalis) involved in motivated learning to serve as a surrogate for training. By pairing this stimulation with repeated low tones played from a speaker, the scientists were able to take aim at the low-frequency portion of the auditory map, and encourage its expansion.
Later, when these rats were tested on tone discrimination tasks, it initially seemed like a score for the classical remapping idea. Indeed, having a bit more cortical bulk devoted to low tone processing was associated with significantly faster learning. Rats with larger low frequency maps learned to discriminate between similar low tones within three days, while control rats took more than a week.
However, when the rats were tracked over time, the researchers found that the enlarged areas of the auditory map began to shrink, or ‘renormalize.’ In fact, within 35 days, the artificially expanded areas had returned to their original size. Critically though, despite reverting to cortical status quo, the rats retained their sharpened perceptual skills. Similarly, when Kilgard’s group tracked map changes in rats trained on tones the normal way - with weeks of hard work and no artificial boost - they observed the same basic phenomenon. Maps got larger, tone discrimination improved and lasted, but maps then settled back to their original organization. Whatever separates a virtuoso rat from a less-discerning one, it’s probably not a difference in gross brain organization.
So what does change? Although newly learned perceptual skills don’t show up in a bird’s eye view of the cortex, they must have some neurobiological basis. Kilgard suggests that learning probably results from a few parsimonious tweaks at a more microscopic level, involving relatively small numbers of neurons and synapses.
In one sense, this view isn’t so different from other ideas about learning and neural function. The brain is an ultra fine loom, and many of its functions are regulated on a mind-bogglingly small physical scale. Perhaps it’s not surprising that small- and hard-to-track physical changes could add up to changes in skill or behavior.
Still, there’s a big question lurking here. If a newly learned skill, in the end, has only a tiny footprint in the brain, why not tread more lightly during the learning process? Why go through the trouble of having large functional regions of cortex expand and shrink, only to arrive at the few small neural differences that make a perceptual difference?
It might be because your brain is no smarter than you are. Just as you don’t know how to move from novice to expert in a few optimal and wisely chosen moves (how could you?), your brain doesn’t either. In a provocative theory that describes his findings, Kilgard speculates that the expanding cortical map is like a search committee. It’s generating a huge range of candidate solutions to a problem the brain has been tasked with, but doesn’t yet know how to solve. (How do I discriminate these tones? How do I get the ball in the basket? How do I solve that tricky calculus problem?) Once a good solution is found, the search committee is disbanded. Efficient changes that impart skill are retained, and the non-meaningful changes are winnowed away as the map shrinks.
There’s something empowering in the idea that learning might ‘look’ the way it feels. Standing on the perch of mastery, it’s easy to look back and say what the straight path to expertise is. But neither we nor our brains get that free ride. Perhaps we just have to make a lot of moves - many of them redundant, roundabout, and just plain bad - to ensure that we stumble on the few good ones.
Are you a scientist? And have you recently read a peer-reviewed paper that you would like to write about? Please send suggestions to Mind Matters editor Gareth Cook, a Pulitzer prize-winning journalist at the Boston Globe. He can be reached at garethideas AT gmail.com or Twitter @garethideas.
See what we're tweeting about
11 Comments
Add CommentHi,
Reply | Report Abuse | Link to thisNice post.
This is really similar to what happens in the motor cortex during skill learning (Xu et al., Nature, 2009, doi:10.1038/nature08389 and Yang et al. in the same issue). Early on, there is formation of many synapses early in the learning and synapse formation correlates with learning. Later on, the number of synapse per neurons decreases with more practice. I guess this trimming improves the signal-to-noise ratio. Anyway, perceptual and motor learning appear to share similarities and that's interesting.
Thanks for the nice blog
JJ
So, that looks like a "sound" mechanism for energy saving. I'm assuming that mapping is occurring all the time (in some sort, checking for any changes on the stored 'arrangement'); if this is correct, would this mapping consume less energy than a constant expanded cortical area (before trimming), or is it a part of a bigger picture, where (re)mapping of the auditory cortex crosses over other cortical areas (hence, consuming more total energy but still less per area unit)?
Reply | Report Abuse | Link to this@jjodx I think those refs you listed are spot-on. I linked to them in one of the paragraphs above (‘parsimonious tweaks’... ‘microscopic level’), but didn’t have room to describe them in any detail. Thanks for pointing them out. It would be cool if this is some sort of general (or even the main) engine that drives learning. Next step is to try to tag the stably-maintained population of synapses/spines to see if silencing them can selectively snuff out a sensory/motor memory.
Reply | Report Abuse | Link to this@joset Good points. The Kilgard paper also mentions that expansion followed by trimming is probably more metabolically efficient than having to maintain some large population of new or altered synapses. The brain is working with a tight energy budget, and that fact can account for certain properties of neural activity (spike dynamics & typical firing rates... check out A. Hasenstaub et al PNAS 107(27), 2010). Would be cool if expansion/renormalization is a ‘cost effective’ solution to learning, relative to other plausible schemes. Energy per unit area/volume is also probably an important constraint (don’t want to overheat!).. not sure how it would favor one mapping/remapping strategy over another though.
Interesting blog. If the findings are correct, could we be on to a neural explanation for the question, "Why it is easier for people to perform 'adult' tasks than 'commonsense' tasks?"
Reply | Report Abuse | Link to thisReduction in neural allotment in order to achieve a certain task repetitively or successfully may be boiled down to the conservation of energy. Simple is best, even in the human brain and simplicity may be measured in joules required to attain a goal most efficiently.
Reply | Report Abuse | Link to thisI speculate that the winnowing process seen in these learning experiments are related to the winnowing process that occurs in the brains of newborn babies? I've read, can't recall where or how often, that newborns have substantially more neural connections than toddlers, and that as the babies age the neural connections decrease in number. Seems like the same process is at work in both cases - at least to this interested lay observer.
Reply | Report Abuse | Link to thisThis confirms my own experience. I have had many, many different jobs in my life; manual and mental. Always in the beginning I am criticized for going too slow, then, always finish up as one of the faster. It is good to see that there is an actual mechanism that accounts for my experience. I'm amazed how clever evolution has made the human brain. It is self optimizing.Computer science has a way to go.
Reply | Report Abuse | Link to thisAs I teach people how to master physically moving out of pain and stiffness using the cortical process of a pandiculation vis-a-vis somatics exercises...
Reply | Report Abuse | Link to thisI recommend that erring into correction will take place especially when we release ourselves slowly from an intended direction of movement. The hesitating, jerky quality of an overly contracted muscle releasing is apparent to both the observer and the experiencer.
A few repetitions is all it takes to reset followed up by days of repeating the process until it smooths itself out and transfers to the cerebellum.
Like updating our computers, resetting our movement software like healthy vertebrates allows us the freedom to move like an animal and not be lame and get muscle spasms merely by reaching for something whereas how many times does the cheetah pull a hamstring running at 60mph.
The 7-10 morning tweaks, followed by the 50 or so throughout the day keeps the animals moving quite well.
Throw in the differentiation we can add as the human animal and the ability to reorganize our 17 layers of muscles gets easier and more interesting as we age... we can pickup the subtle cues of the motor-sensory system if we decide to go there and quietly listen as we move.
Like many things about learning, learning to move well doesn't have to end... just keep erring into correction.
Wow, this is fascinating! The implications for this information are numerous. I blogged about this article the other day! http://patientmadman.blogspot.com/2011/06/learning-brain-gets-bigger-then-smaller.html
Reply | Report Abuse | Link to thisI think you'll agree with me, the implications for this information are numerous.
I must've gotten distracted while typing that, my apologies for the repetition.
Reply | Report Abuse | Link to thisJason - excellent summary. I've come across this concept a few times in the past few years. Originally I saw it mentioned in Hawkin's book: "On Intelligence". More recently I've seen suggestion (Goldberg, "The New Executive Brain") as to what might be going on. Goldberg proposes a model of left/right hemispheric distinction is. He claims that the two sides have different functions because they learn at different rates and in different ways. As a result the one side is the generalist fitting problems to existing patterns, the other learns more slowly but creates new patterns to match problems.
Reply | Report Abuse | Link to thisYour thoughts?