RNA Memory Hypothesis
scottlocklin.wordpress.com
scottlocklin.wordpress.com
There are other weird parts in the blog post, for example claiming that a particular protein kinase is involved in mRNA transcription. That is by definition wrong as a kinase is an enzyme that transfers a phosphate, so a protein kinase transfers phosphates to proteins. And the RNA in this hypothetical memory storage would not be mRNA, as that by definition encodes for proteins.
The person cited in the post is also a crackpot, I've quite a lot of experience with his exploits on the Stack Exchange network. And I haven't found a conspiracy theory yet that he didn't like. This is not hyperbole, I learned of a whole bunch of conspiracy theories I didn't know existed through his posts.
Memory is literally everywhere. There is already evidence that the brain can encode memory in dynamic molecular processes, and that interference of those process can interfere with an animal's ability to recall memory.
Your point about timescales is a good one, and provides a clue: the timescales over which a given medium of information can be encoded to or decoded from and over which it decays should provide a clue about the timescales we might expect organisms to experience it in. Mammals and other creatures with RNA experience memory over a whole continuous range of timescales. Why wouldn't RNA, which is optimized for encoding information, be just one of many mediators of memory?
Of course there's a lot of other stuff RNA can do like catalytic activity, triggering RNA interference or reacting to other molecules (riboswitches). But information storage isn't really anything RNA is known for.
Well there's lots of information there which can be used to infer aspects of previous states. I assume it's more of this aspect that is being referenced, rather than what is encoded in the RNA.
I am personally interested by the 'RNA Computer' hypothesis which posits an additional layer of differentiation in computations performed by interacting RNA.
I have also read of experiments where memories (learned behaviors) were transplanted by injecting ground-up neural material. Flatworms were the subject I think.
Of course this is all near the edge of mainstream, but I think genuine science demands openness to new ideas.
Although I'd again argue that there is no single "true" medium or mechanism of memory.
I'd remind you that DNA is indeed memory. It's memory whose recall is heavily modified and contextualized by its environment, for example methylation, the topology of how it is currently twisted, the kinds of nutrients available for molecular synthesis, etc.
If RNA mediates memory on a sub-generational timescale, it seems that it'd provide a dynamic, contextual bridge between generational memory in DNA and shorter term kinds of memories that play out over timescales less than an individual's life. So that the molecular formation of a given memory would have a semantic basis in the longer term context. A concrete but way oversimplified example of this might work:
You go through a traumatic experience where you almost die. The specific way you almost die is related to a number of factors involving your ability to run fast enough. The long term memories of that experience are then encoded as proteins by a mosaic of genes related to certain muscles, their shape and size, certain metabolic pathways, etc. The way we might unpack or interpret that memory at any given time/context could be as a gutteral impulse that results in such realizations as "I need to get more exercise" or "my grandparents had heart disease, maybe I should go to the doctor" or "wow I'm sure hungry for fish", etc.
Btw I'm not even arguing that it's as clear cut or simple as that, just sharing how I intuitively make sense of all this evidence and these hypotheses.
So as an simplified analogy, the DNA would be like a dictionary of common words that could be used to encode a meaningful story about something the organism has experienced. The RNA would be like the handwritten copy of each word as the story is written down on a piece of paper. The paper with the handwritten story is analogous to the proteins that have been shown numerous times to be involved in long term memory storage and recall.
Yes, it contains information. But you can't write to it.
My hunch is that you're thinking of memory as being equivalent to RAM in a computer. So I'll remind you that read only memory is still memory.
Some microorganisms, like viruses, do directly modify DNA. That's where CRISPR came from.
I'd encourage you to check out the full range of definitions of the word "memory".
An imprint of history is not memory, in much the same way that fossils, stars, the environment is not memory. Human DNA has not been written to. It is edited in only very specific circumstances - DNA repair, TCR/BCR recombination, meiosis and interestingly, and possibly relevant - LINE-1 transposons in neural tissue.
At any rate, DNA encodes useful information related to how an organism's ancestors were able to survive their environment. If you prefer not to call it "memory" for semantic reasons, I don't really care too much, although my opinion is that memory is too deep of a concept to be narrowly confined like that. Nevertheless, even if you don't call DNA memory, it doesn't change my point in the comments above, which is that the information encoded in DNA is directly relevant to any survival related memories, and so it would make a lot of sense if such memories were formed in direct, physical reference to that information. Seeing as how it's been shown that protein formation is involved in memory formation and recall, and seeing as RNA is how proteins get their molecular structure, it's not a far reach to see the possibility that RNA is a contextual bridge for memory to be encoded in reference to survival related information provided by DNA.
None of this requires that DNA be writable in the same way that transistor based me let is writeable. DNA is written through a probabilistic, distributed process, but it still encodes information about the molecular tools its carriers successfully managed to survive and reproduce with.
Viruses integrate into the human genome, then decay and eventually become fodder for new regulatory regions and genes.
During human life, your genome is mutated, and your body repairs it, sometimes making mistakes, which are permanent (and passed down to your children if it happens to the germ line).
I could go on...
If you're saying it's theoretically possible, I would agree that it might be, and that there is a potential for DNA to be memory. But to claim that DNA is memory without a known write process is premature I think.
Whether it's learned or you were born with it doesn't really change that the information had to be stored and transferred which is...memory?
I remember hearing about this one really interesting experiment where they would expose mice to a specific scent (I think it was citrus), and then give them a shock. Obviously after a while, when the mice smelled that scent they would get nervous and try to hide (or something along those lines, this is from memory so forgive any erroneous details). They then bred those mice and separated the offspring from the parents so that there was no way they could learn about the smell from them.
When they exposed the offspring to the smell, they reacted in the same way their parents did. Without ever having been exposed to it before, or knowing about the shocks. Pretty interesting stuff.
Another attempt:
Straw man:
Phones are too slow and too dangerous to be used as a real time vehicular traffic indicator. If drivers were to text their location back to google every 10 seconds, they would have too many accidents. It isn't even possible to look up the location on google maps and then text a geocoordinate in 10 seconds. Drivers just wouldn't do this anyway because there is no incentive for them to do so. Google would have to pay drivers $2 or so for each location bulletin. That would cost billions and google would go bankrupt.
My first reaction was to downvote your comment for the ad hominem, which doesn't contribute to the discussion. But I can see you are giving consideration to the article, and want to discuss it further - it's just that you haven't got enough out of it to make much of a start, and so this is your conversation starter.
I'm assuming you've labelled this a "fringe theory" by a "crackpot" not to be nasty, buy in order to start a conversation.
Let's start with your statement:
> I don't think there's any real doubt that memory is stored in neurons
The question is "how?". The article is postulating that each neuron is a computer, and RNA is a storage medium.
You say "my first objection would be that RNA would be far too slow for this purpose." That's because your thinking about RNA is limited to what you understand about RNA. You've learned about mRNA, which is one application that uses this molecular format, but it's not the only possible application.
The article suggests that the data transmission protocol which leads to thought and/or behaviour could be electrical, through simultaneous aggregated neuronal RNA based data processing.
Read the section he quotes from Ron Maimon, subtitled RNA ticker tape. What do you make of this statement?
"The RNA in a cell is the only entity which is active and carries significant bit density."
DNA certainly encodes for a significant amount of information as well. And why not proteins, there are a lot of proteins in a cell and you can also modify them in various ways. A cell has an enormous amount of state, any of it could theoretically store information like memory. There are a lot of different molecules active in a cell, in various modification states, at various locations, and all of that could encode information.
What this blog posts doesn't provide is any experimental evidence, this is a pure thought experiment. If RNA were to actually store memory, that would be great. Because determining the RNA sequences inside a cell is something we can do.
True!
> What this blog posts doesn't provide is any experimental evidence, this is a pure thought experiment.
Yeah, so?
mRNA and tRNA move, so "active". Ribosome is also (re)active, and "the ribosome is itself composed of both RNA and protein, likely reflecting its own descent from an RNA world." [1]
[1]: The Ribosome Moves: RNA Mechanics and Translocation
It appears that the brain cells in particular actually do develop significant changes to their DNA.
Why? Ad-homs aren't necessarily bad, it's contextual.
More informative to state the views you disagree with and say why.
> I don't think there's any real doubt that memory is
> stored in neurons, this is really a fringe theory.
All theories are fringe theories at first, that's where they start (as hypothesis before there's any experimental evidence or mathematical proofs).
I would agree that there is little doubt that memory is stored in neurons. However, that does not exclude some memory being stored elsewhere. It's generally accepted that the brain stores memories first in short term memory, and then after a time some of those memories are stored in long term memory, with the memories chosen according to some criteria.
I could see RNA memory as a third stage of memory storage. Memories that the brain views as high impact to the species according to some criteria might get copied from long term memory into RNA memory to be passed on to future generations, accessible only by subconscious.
If RNA is not simply the conduit between DNA and protein, but the original biological operating system, from which many functions have been refactored out to DNA and protein it's fascinating to think about what other functions may have existed and may still exist entirely in RNA.
I like to assume that RNA is the original biological information store, the original structural material and the original catalyst. DNA is the long term cache version of RNA information, and protein the improved structural and catalytic material. The information in RNA can be translated directly to and from DNA, and there is an interesting idea that the mapping from RNA codon to amino acid may have began directly on the RNA strand, some RNA sequences seem to have an affinity for certain amino acids.
But it is a thing to write in GPT style.
GPTs want to be human like, for funding.
Chess AI's still can't play like a human (but people are working on it), chess AIs were designed to solve problems.
The easiest way to pretend to write 'human like' for a computer is to say as little as possible with tiny contractions and poetic words.
Some humans similarly do this, if you get paid by the word it makes sense.
OP style fails on this topic because it's so technical.
GPT-3 wasn't even designing to mess with our heads, but it does, wait till someone designs it to. It's very possible, because it's not about intelligence, making it worse is easier than making it better. If you use engagement as a measure of 'success' you can see you can evolve GPT-3 quickly to a worst nightmare. (It is already a nightmare)
The claim as I understand it is that there is no “memory” at all. What we experience as memories are simply input parameters that signal our brain to begin a particular function. In the way a neural net does not “remember” the training data but carries out what it “learned” anyway. I have problems explaining this idea as I think it is full of holes and would require some fundamental changes in how we think about feedback structures (I mean this neurologically as well as mathematical/computational).
Either way if most people don’t doubt memory storage in neurons then they are very quite about it. I’ve rarely ever read an argument for memory storage mechanisms in the brain. A good reason for this is simply the question: Where is it then? Neurons individually can’t really provide this sort of expansive data storage. And what they seem to be able to account for is too little and too widely spaced apart. Interestingly this spacing also provides a counter argument against the complex processing that would be required if there wasn’t a memory mechanism.
Personally I think the dismissal of memory comes with a lot of hubris attached, as well as a total lack of understanding information theory. Information theory itself seems to be popularly bemoaned by neuroscientist as well. But it’s hard to see the alternative and is a bit paradoxical since the only way I can seem to understand the “no memory” claim is by using a computer neural net metaphor.
Book: [0]Memory and the Computational Brain, C. R. Gallistel and Adam Philip King
You can read the main thesis here https://onlinelibrary.wiley.com/doi/pdf/10.1002/978144431049...
Review: https://www.researchgate.net/publication/262093114_Man_as_Ma...
Also please take all my comments with a grain of salt. I am not expert, I'm just trying to restate what I've read and understood.
[0]Memory and the Computational Brain, C. R. Gallistel and Adam Philip King
You can read the main thesis here https://onlinelibrary.wiley.com/doi/pdf/10.1002/978144431049...
Human language is like that, full of ambiguity, of implicit baggage.
If you feel "memory" means a recording of an event, or a photo book, then faced with the discovery that brains actually don't work like that you may want to state "neurons don't store memories!".
Are you claiming there is a large subset of neurologists how do not believe in long-term potentiation and other plasticity processes driving long-term retention of memoranda? If so, could you please show me to these crowds?
Long-term potentiality is not the point of disagreement. The disagreement is that individual memories are not recalled, as they are not stored, per se. The idea is that neurons acclimate to inputs, so that when we "recall" them we are not retrieving a "memory" from a storage location but instead are being activated in a way that allows us to preform a learned action.
I imagine those marble/cardboard diagrams of logic gates, where the cardboard flap will be left in a particular position after a marble has passed. In this way you can add numbers by dropping marbles into the appropriate input slots. Yet in this way, one could not derive prior calculations, as it does not store this information anywhere.
[1] https://www.cell.com/cell/fulltext/S0092-8674(19)30448-9 [2] https://www.cell.com/cell/fulltext/S0092-8674(19)30552-5
As is often the case, this article conflates some ideas that are truly remarkable and interesting (how are memories stored?) and mixes them with very strong statements that are clearly wrong. In this case, it is clear that synapses (connections between neurons) HAVE to be involved in memory, as the only way that a memory can ever be expressed is if a neuron triggers another neuron, etc, down the chain to a muscle. (OK, thoughts don't necessarily involve muscles, but to show your thoughts to the world.) And if synapse *dynamics* are not involved in memory, then either they never change, or they change randomly, both of which will lead to problems. (That's an exercise left to the reader.) So dynamic synapses are involved in memory, definitely, no question.
The article mentions a couple of things that are interesting misdirection, like invertebrate models where "memories" are transmitted via serum. These are super cool experiments, but these are memories in the same way your immune system "remembers" a virus. Despite Tonegawa's best efforts, I think most researchers are quite confident that recalling your memory of last Christmas uses a very different mechanism than your immune system does to generate imagined tastes and smells and warm feelings.
The article also talks about RNA ticker tapes. There is some super cool neuroengineering research around the idea of making a designer protein that would record a neurons activity in nucleotide chains as a readout. Definitely worth a dive if you're interested.
And how neural activity drives synapse dynamics is a very big open question. It's known to involve protein synthesis (pkm-zeta, anyone?), so RNA is definitely involved.
Finally, for those curious about how memories work, look up the "cognitive map" or the "index theory of memory".
Scientists have already modeled memory retrieval using current neural network models. There's literally videos on youtube demonstrating this. I'd be surprised if RNA was used to encode memories to the extent OP is suggesting: there would have to be a mechanism for storing this RNA inside cells or there would be evidence of random strands of RNA floating around in the body. We would have been found it a long time ago, and scientists would also be able to decode it if these strands of RNA were standardized and could be transferred person to person.
Though the way evolution works and how complicated the body is, it could be occurring in a small sub process in a neurons somewhere in the body. though this would likely be tied to hormonal responses or something equally boring.
Biswas, S., Manicka, S., Hoel, E., and Levin, M., (2021), Gene Regulatory Networks Exhibit Several Kinds of Memory: Quantification of Memory in Biological and Random Transcriptional Networks, https://www.cell.com/iscience/fulltext/S2589-0042(21)00099-7
This indication is the opposite of worrying to me, it's reassuring.
Everything be replaced? What does that have to do with brain emulation?
Also I think you're implying that brain emulation will lead to more "unimaginable monsters", when it could very easily be the opposite: if you can't emulate a brain then an AI has to be even weirder and harder to understand.
* Prodigies which happen to be good in a field related to what their parents are good at. Often the wording used indicates a memory-like reason. For example, finding the rules of chess "familiar."
* Embeddings in machine learning, which allow encoding of high-dimensional information in a low-dimensional space. This would imply that humans have a somewhat-shared embedding space at some "layers". It's interesting to wonder how similar this space would be across various animals, as it would effectively be an evolved language.
* Other more subtle things, like the "cultural memory" of a society.
One question would be what types of time scales these memories survive across. Maybe some forms of instinct are deep memories that have been directly encoded in DNA.
https://www.scientificamerican.com/article/prehistoric-virus...
I also wonder if and to what extent this information can transfer between individuals?
I'm reminded of the Jungian collective unconscious.
You're going to think I'm a kook now, but, there's also a panspermia angle on this. Big ifs here, but. If panspermia is true. If evolution on earth is tied to evolution of life in the wider universe via transfer of genetic material. Then it could be possible that the code in which memories (or a learned manifold, encoding useful generalisation) are encoded is compatible with life here today. You could conceivably get not just morphological/physiological changes in terrestrial organisms through viral horizontal gene transfer, but also transfer of cognitive strategy via the same mechanism. X-files kinda stuff. All kind of unlikely, but an interesting possibility.
> The role of viruses in evolution is almost mainstream now, it is much more likely that virus was just a method for ancient extinct organisms to exchange genes
Of course it is.
You seem to be arguing with a straw man.
I think this could be equally well explained through 'nurture' rather than 'nature'. If a child's parents are skilled at chess and play it often, the child is going to be exposed to it at a much earlier age and for more often than a child of non-chess-players, making it more 'familiar' and 'intuitive'.
To be honest I'm not super familiar with the research in this area but my sense was that there's some amount of experimental evidence for both genetic and environmental factors playing a part together, rather than just one or the other. It's very interesting though -- might have to go do some reading.
In reality, we've had an alternative explanation for about 20 years. The chemical receptors that respond to the molecules E. coli wants to get away from activate and generate a cascade that causes the flagella (the stranded "tail" molecules that allow them to swim) to start rotating in the opposite direction. This causes them to become tangled, and this results in the bacterium moving around randomly rather than making any progress in one direction. If it heads away from the noxious source, the receptors stop activating, the flagella start rotating in the correct direction, and it heads off in a straight line again.
But what about the memory? If the bacterium didn't remember what happened, isn't there a high chance that it would just end up doing the same thing again? Yes, but we don't need to bring RNA into this. One of the molecules involved in the cascade is activated by phosphorylation - once phosphorylated, its sensitivity changes. Over time, another molecule dephosphorylates it, but this altered sensitivity means the bacterium is more sensitive to the noxious molecule than it was previously and biases the bacterium to move away.
So "memory" in this case is actually just a modification of the gain in the response to a stimulus. Does it demonstrate that responses can be influenced by something that happened in the past without requiring a nervous system? Yup, absolutely. Is it something that gives us a better understanding of how rich long term memories are stored in the brain? Not really.
(Source: my first attempt at a PhD was working with https://en.wikipedia.org/wiki/Dennis_Bray who did a bunch of the computational modelling that demonstrated we didn't need anything overly complicated to understand what was going on here. Absolutely wonderful scientist, but I fell out with departmental IT staff over network security stuff. Ironic with hindsight. Also, this is my recollection from almost 20 years ago, so details may be inaccurate)
In a very real and observable sense, the 'blueprint' for the target worm is stored in the tissue system rather than as a 'plan' stored in the DNA itself. I think it feels surprising to see something that seems a lot like 'memory' influencing the action that results in the expression of the complete organism via the cellular level processes in this way.
I find it easy to postulate from the example of planaria that analogous 'memory encoding' systems probably exist in all sorts of cellular systems ... though I think the information theoretic reasoning referenced by the participants in that podcast for how life creates and operates these systems makes it likely that 'the existence of information storage and retrieval' in a cellular system is not at all required to imply that 'this kind of system is therefore the storage model for long term cognition'.
[0] https://www.preposterousuniverse.com/podcast/2021/02/01/132-...
It's a small scale, but it's not empty; ask yourself how many atoms are in a human cell, then check this link for an answer and see how close you are: https://www.thoughtco.com/how-many-atoms-in-human-cell-60388...
See also this talk by Prof. Michael Levin, same guy as your podcast - https://news.ycombinator.com/item?id=18736698 - titled "What Bodies Think About: Bioelectric Computation Outside the Nervous System", including from the top comment by keithwhor:
" - Organism morphology seem to be highly dependent upon large-scale electrical potential differentials between cells
- These electrical networks are primarily regulated by cell-to-cell gated ion channels; simple chemical pumps (the same types of membrane proteins that control faster-acting electric potentials that enable muscle contraction, neural activity)
- These networks and patterns of bioelectric signalling have stable memory (once a pattern between cells is induced, it remains stable) and are responsible for driving "subroutines" of large groups of cells -- morphological gene, protein expression downstream of this "
For the better part of the past week, try as I might I couldn't remember either his first name or last, but I had a vague feeling about the "fit" or "taste" for lack of a better word of his name. After a few days of conscious effort, yesterday I remembered his first name spontaneously and then suddenly his last name too. Almost like a AWS Glacier request.
A background process completing and notifying the result to the frontend.
How? It’s not something you can teach them. They say it’s an innate thing. But what does that even mean.
I think babies are born with this programmed into them. Like a genetic type of memory.
Then over time, the neurons establish more connections, and learn more advanced things. But the spark for that was set off at birth.
Interestingly, patients with brain lesions in certain places or with diseases like parkinsons sometimes lose the ability to suppress these reflexes.
More info: https://www.ncbi.nlm.nih.gov/books/NBK395/
This would be a distributed memory system, as obviously not all RNA molecules in the body could transit from memory state M0 to M1 spontanuously. So there has to be a consensus and propagation mechanism. Wouldn't it be cool if it turned out to be something like CRDTs?
Edit: 10 hours... https://bionumbers.hms.harvard.edu/bionumber.aspx?id=104747
"there’s no obvious way for all that sensory data to be captured in synapses as long term memories"
...just ignores these findings. The model described in the article on the other hand, has never been observed. At the least, if you are going to propose an mRNA hypothesis don't use "T" as one of your RNA bases.
Here is a nice review article:
https://royalsocietypublishing.org/doi/10.1098/rstb.2013.013...
Synaptic weights is our computational model of neural networks that seems to work pretty well. But this is a model. No one has discovered the biological mechanism by which synapses are "weighted". The actual computation going on in a cell is totally based on chemistry.
There are a lot of different specialized networks in human and animal brains that don't necessarily translate easily to "weighted synaptic networks" all that well. For example, networks that process audio information have been shown to recognize individual frequencies (in fact, they pretty much solely do this!) Such networks seem to be primarily based on the firing rate across a synapse. For such networks is seems more reasonable to think of neurons as signal processors whose internal computation is more based on the time domain, e.g. some kind of Fourier transform-equivalent.
In short, I think you should put less weight (pun intended) in our computational model of neurons (link weights) and look more at biology.
After gene expression, mRNA's are left floating as debris in the cytoplasm, until they are cleaned up and expelled. If the gene expression continues chronically, the same mRNA sequences will populate the same cells, transiently, for as long as the gene expression recurs. What if the mass of the mRNA sequences influences the electrical conductivity of the cell e.g. functions as a form of antenna. In the aggregate, cells with mRNA populations e.g. neutrons active in particular thought patterns, might have stronger antenna, or differentiated electrical characteristics, and hence influence thought more or less in the network, for the duration of the gene expression.
If RNA is the memory mechanism, and it floats around for a while, then it gets cleaned up over time, this collection might allow long term memory recall and replay.
Function 1: Surface protein encodes the pattern of incoming synapse firings on RNA, then releases it. This pattern would have a start and end block indicating it is "new".
Function 2: Relatively rare protein finds RNA marked "new" and replaces start and end blocks with "old" markers.
Function 3: Common protein grabs one strand marked new, and one strand marked old and compares part of the tail of "new" to the head of "old" by sliding them past each other.
If a match is found, the older strand is marked "match" at the start.
Function 4: Match strands are used to trigger the next several firings of the neuron, then marked "hold"
Function 5: "Old" strands are gradually recycled.
Function 6: "Hold" strands are copied into new "old" strands for comparison to "new" strands by function 3
Function 7: Something to do with continual refinement of "hold" strands with incoming "new" strands.
...
Anyway, this whole little system just popped into my head based on the small amount I know about crispr and cellular machinery (which honestly isn't much), and I wanted to share.
Maybe the ancients are right and the soul resides somewhere in the liver. Don’t laugh; people have led normal lives with giant pieces of their brain removed, but nobody has survived the death of their livers. The former fact; normal people getting by without much brain tissue, at least, ought to be the end of the argument: purely Hebbian models of the brain are obviously false.
This is interesting.
So if you transplant a liver, you get memories of the donor’s eating habits? LOL.