I find the assumption built into this article's incredulousness odd and I doubt I would ever had made it myself. Why would we ever expect qualia to be based on specific and static combinations of neurons? The fact alone that memories last a finite amount of time should be reason enough to conclude (given our limited number of them) that that's not how neurons work, at least insofar as the qualia of memory and present experience are the same, which seems like a sane null hypothesis to me. I think there is a strong cognitive bias in the present culture of thinking of the brain like a computer or hard drive. It's not, and I would have been more surprised if the scientists here had made the opposite finding that they did: that the constant qualia (in these mice's smells) was linked to constant combinations of activating neurons. That would contradict existing observations on brain damage and imply uncomfortable things about electrically programming the brain to produce known outcomes.
I hate to quote the article so directly, but to make the headline a little less click-baity:<p>> Put it this way: The neurons that represented the smell of an apple in May and those that represented the same smell in June were as different from each other as those that represent the smells of apples and grass at any one time.<p>> This is, of course, just one study, of one brain region, in mice. But other scientists have shown that the same phenomenon, called representational drift, occurs in a variety of brain regions besides the piriform cortex. Its existence is clear; everything else is a mystery.
When anything like this gets published, I always wonder "what would it look like on a 6502?"<p><a href="https://storage.googleapis.com/plos-corpus-prod/10.1371/journal.pcbi.1005268/1/pcbi.1005268.pdf?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=wombat-sa%40plos-prod.iam.gserviceaccount.com%2F20210610%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20210610T111033Z&X-Goog-Expires=86400&X-Goog-SignedHeaders=host&X-Goog-Signature=aba6b0b5aec2a0ae9134b8719786a66615f59c155697a9073e4767ab50696e02725e8f848f44ce41958e4b644fb03a7a7c1a2fca79ae9095d1e1a808d6f81a5ad96c529fded34e3bc7be76f2c1416047297334ce6f8cc02b25132073daf96dc0ba71866d00c4db12c404def6e0687e1b2b0f6d55bd07e83361cca5a979729803bc94646ff4895594ba54a852e4e98b9ae61b75dc08d73f3b36462931a2a72285bfa4bbda01d4c421a2985bc75876f333cfd2eb672adc2fb0ad4be96524157ba62c38f7b19389c33ae7c0eb7b53f4d39406e76e7bc01b94a5fe890bc25ead8c12498080d01248469bb437f3d4433ef137d16523cc0c199b157eadc275bfbc7acf" rel="nofollow">https://storage.googleapis.com/plos-corpus-prod/10.1371/jour...</a>
<i>"Scientists are meant to know what’s going on, but in this particular case, we are deeply confused."</i><p>What? I'm a scientist, I often don’t know what’s going on! I see it more as my duty to help others understand that that is ok, and often even a good thing because that is when the fun starts. Science is a method, not a “state”. States are for religions, they “know” what’s going on because "it has been written!". Us scientists are always writing.
It seems to me that this phenomenon of "representational drift" fits well with the results that show that every time a memory is remembered, it is essentially deleted, and re-recorded. Given what we know about how memory and experience make use of large swaths of the same neural pathways, it would make sense that this is the case with re-experiencing sensations too, not just re-remembering them. And if that is the case, then it would make sense that it's possible for these representations to be re-recorded in a different place or in a different way. For example mixed together with other neural wiring that is correlated but wasn't there when the original representation was made (basically de-duplicating the file system). It would also suggest that either pathways are mutating, or different neural pathways are "competing" for the right to be the ones to process a given input (and if it's the latter, then it also implies that there must be some kind of fitness function "how well did this pathway process this input" that allows a new pathway to win out over a previous one).
Sounds like defragmentation are first sight. Seems likely that the storage algorithm for human memory could use periodic optimization, especially since we recontextualize memories over time.<p>For example, let’s say we smell a lychee for the first time in a shop, and the brain stores that memory using a given set of neurons. Later, a hawker starts selling lychees in the local train in Mumbai we commute on everyday. Now memories of the smell of lychees and the smells and experiences of Mumbai local trains become strongly linked. At that point, in a simplified model of the brain, I can imagine why the brain may want to move those memories so they are using neurons that are closer together. Defragmentation!<p>I am likely terribly wrong, but would love to learn in what way I am wrong :)
Reminds me of how if you take out the speech center of the brain immediately, you permanently lose the ability to speak, but if a slow growing tumor takes out the same region, the brain has enough time to slowly shift over to peripheral areas.<p>As an aside, I wonder if the drift seen in neurons represents thought. Perhaps drift has to occur for the brain to continually think and self improve, even with limited inputs. I wonder how different drift rates in regions of the brain relate to things like better long term memory or creativity.
It’s good to study the piriform cortex to understand memory formation / neuroplasticity, as (IIRC) it’s one of the most primitive “plastic” parts of the brain. In some species (e.g. fruit flies) the piriform cortex is the <i>only</i> part of the brain that changes at all after birth; i.e. it’s the only place where learning could possibly be going on.<p>So, even if the it’s not intimately linked with human memory, the piriform cortex may be the origin of the cellular template that later diverged into whichever other brain tissues are responsible for other types of memory.
I have a similar reaction as bejelentkezni – why would we expect a specific sensation to always activate the same set of neurons over time? It would seem to me that we are observing patterns in a dynamic system. Their relations to other patterns seem more relevant than the specific substrate members participating in the pattern.<p>While too simple, I think by analogy of a whirlpool current in a stream: we would not be surprised at all to observe that a given molecule of water participating in the whirlpool pattern at time (t) has only a (p)% chance of being in the pattern at time (t+1).
TLDR: Representation of the world doesn't seem to be locked to specific neurons or groups of neurons; over time, they drift around to completely different groups of neurons.<p>Someone here likened it to how programs can move around in computer memory. If neurons are the "silicon of the brain" or the "PHY" layer, the biggest problem in neuroscience IMHO is that we have virtually no understanding of what is the next layer up, (i.e. the "logical layer"). What is the equivalent of program counters, CPU instructions, etc. in the brain?<p>The brain is a nonlinear recurrent dynamical system and we are really in the dark as to how to break the dynamics down into understandable subcomponents.<p>To get an idea of the complexity, this paper by Randall Beer analyzes the possible behaviours of 1 and 2 neuron circuits:<p>Beer, R. D. (1995). On the Dynamics of Small Continuous-Time Recurrent Neural Networks. Adaptive Behavior, 3(4), 469–509. doi:10.1177/105971239500300405<p>Obviously this kind of analysis doesn't scale to 10^10 neurons.
It’s worth reading the linked paper on the causes/consequences of representational drift: <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7385530/" rel="nofollow">https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7385530/</a>
What I like about this, and what I like about existence in general is: While we continue to learn new things sometimes rapidly sometimes slowly, there is so much we truly don't know and it's both frightening and exciting. Constantly updating our priors.<p>What I don't like: so many humans think they know everything or at least position themselves that way. FYI this isn't an attack on experts.
My bet is reinforcement decay and strengthening exists to promote novelty rewards. Smell 1000 roses, it's a big "meh" but the identification of them becomes easier and better with strengthened pathways. This would be adaptive for the organism to:<p>1. seek new, potentially-massive rewards, stimulation rather than do the same thing over and over again without ever trying anything new<p>2. improve processing of stimuli that are more frequent
What is so confusing about this? The first time I read a book, it impacts me in a certain way. Every successive time, it hits me differently. I take different life experiences to it, and I understand it more deeply because I've read it before.<p>Why would a smell be any different? Each time I smell an apple, or taste wine, or examine a painting, I am a different person, and I also discover more nuance about the subject.
Well, I do find it’s easier to do significant refactors by copying bits of code into a new file or project rather than trying to make changes in-place.
> If representational drift can happen in the piriform cortex, it may be common throughout the brain.<p>Instead of looking at individual neuron response properties, look for population codes. This would be actually an interesting experiment with an artificial network, to observe how the population activation drifts over time (but remains identifiable) as the network is fed with more and more data
Isn't Neuralink (and other brain interfaces, I assume) also based on the idea that you can pinpoint which neurons are responsible for each function in the brain? If I've read the article correctly, it seems to suggest that they're a moving target. How can permanent implants interface with a moving target?<p>Might the entire Neuralink concept be based on a flawed assumption?
Imagine if we discover most of the brain's functionality isn't in the brain or any other identifiable part of the body. Rather the brain just connects to something else, that's outside the known timespace.
I find the whole spin of this article a little odd. Isn't it enough to talk about how interesting representational drift is without spinning fairy tales about science?
> They needed to develop surgical techniques for implanting electrodes into a mouse’s brain and, crucially, keeping them in place for many weeks.<p>Poor poor mice. Vivisection is done for the 'greater good', but not for the greater good of the millions (billions?) of critters who have suffered it. Yes I know that modern physiology is premised on it as method, the knowledge of organisms as they live not as they are dead, but what a toll.
Someone have found a long post title that can't spoil anything from the article content, so HN users have to click to understand what it is about :/
The real question here is: what insights do computing professionals, whom work with neural networks or parallel distributed processing as their day job, bring to the table?<p>The sort of reasoning that neuroscientists are struggling with is that very same kind of reasoning that computer scientists and practitioners use when thinking about ANN/PDP and deep learning.
There are theories [1] that part of the brain's function might work at the quantum level. If true, we probably won't be able to really understand what happens by measuring it this way...<p>[1] <a href="https://en.wikipedia.org/wiki/Quantum_mind" rel="nofollow">https://en.wikipedia.org/wiki/Quantum_mind</a>
As titles go this one is pretty special. It's my understanding that Neuroscientists themselves claim to understand next to nothing about phenomena in the brain. Meanwhile large amounts of what they think they might understand has such large error bars on it won't surprise them to have it all turned over in the future. They claim to have profound second level ignorance, not to even know most of what it is that they don't know yet.<p>That's kind of what makes the field so exciting, what we don't know is almost all of it! Plus it's not some peripheral thing, it's right there at the core of everything we are individually and collectively. Sadly it's had more than its fair share of charlatans and credulous followers. Sure, the Atlantic, nobody expects them to be any good, right? Well we probably shouldn't but sometimes we still do.
Wouldn't this be similar to how weights for individual terms in a interpolation polynomial can change when adding a new point, while it still (to some approximation at least) captures all the old?
That little line is so toxic and terrible.<p>If my boss came round and said, "This is adenozine, he's supposed to know what's going on," I'd be pissed. That's just dumping blame.
yes, it is. it adapts. firings get less strong as we get used to stimulus, and neurons that fire together wire together. so different connections form on more important info.