Ask HN: Is anyone working on “downloading” knowledge to the brain? In honor of the new Matrix movie, I was curious if anyone is working on or aware of research or methods people are working on to provide "instant knowledge", like learning kung fu: |
Ask HN: Is anyone working on “downloading” knowledge to the brain? In honor of the new Matrix movie, I was curious if anyone is working on or aware of research or methods people are working on to provide "instant knowledge", like learning kung fu: |
As far as I'm aware, all of the current work on brain-computer interfaces are focused on output, interpreting electrical signals from the brain either to control external devices or perform some extremely simplistic form of "mind reading." I don't believe anyone is working on input.
That said, there is no reason in principle it can't be done, but there are additional challenges in that encoding knowledge in the brain involves changing neural architecture, and doing that via attached mechanical device is risky. The closest to this we've ever achieved is electroconvulsive shock therapy to try and treat certain types of psychiatric disorders, which is controversial at least. I have no idea how effective. We don't currently have the means to do this at any finer granularity level and we wouldn't know what we're doing anyway because we don't know in any general way how knowledge gets encoded in the brain (and for athletic activities like kung fu, the entire CNS, not just the brain).
I can imagine at least a line of research where we attach extremely good internal sensors to a person's entire CNS and observe what happens when they train. If it looks sufficiently similar on all test subjects, you can possibly assume what you're seeing is a pattern of electrical activity that amounts to "learning kung fu" and then try to induce that via mechanical device rather than actual training.
Even if this works, however, you're limited by a few factors. One, if you've ever done any serious athletic training, you'll know it can actually fry your CNS, not just cause muscle soreness. Your CNS does need time to recover from learning activity. So you can't just upload kung fu instantly, or in the span of hours like they do in the Matrix. The human CNS can't actually handle that level of continuous input and needs rest and recovery.
Let's grant you can still learn faster than by actually training. In The Matrix, they're only ever performing kung fu in a simulation, not in physical reality, so their level of physical fitness doesn't matter. Performing real kung fu takes more than just knowing kung fu. You need to be sufficiently physically fit, in terms of strength, endurance, aerobic capacity, flexibility, everything it takes.
I can imagine a future where, given these limitations, maybe time to mastery can be decreased from 20 years to 2 years, but I don't see it ever being possible to become a kung fu master in a matter of hours like in The Matrix. At least one major difference between brains and hard drives is brains can only have knowledge encoding gradually changed. We need repetition to learn. It isn't possible to just erase a file and write a new one. There are no files. And the rate at which it is possible to learn is inherently limited by biochemical realities. "Electrical signaling" in the brain involves ion channels opening and closing, physical molecules moving around. It's an energy consuming process and it wears down the cellular machinery that then needs to be repaired by producing and emplacing new proteins to replace broken down. These same things are true of silicon microchips, but there are major differences in that you can just keep it plugged in and give a continuing energy source. Humans need to eat, digest, excrete waste, all of which take longer. CPUs can be much more easily and quickly cooled. Some rigs just blast them with liquid nitrogen, which you definitely can't do to a brain. And they can be replaced. You don't need to rebuild them in-place at the molecular level. You can just keep a fresh stock of excess chips, and when one burns out, replace it. When a hard drive fails, you have backups of the file and write them to a fresh hard drive. You can't do that to a brain. You have to let it recover and rebuild itself in-place.
The same way deep brain stimulation is not an input of information. They're just attempts to modulate neural oscillations in order to alter behaviors and thought processes
To be honest, we don't understand memory itself well enough to build a device for information input. Even our output devices are incredibly blunt and primitive.
Not to say its impossible, people are working constantly on learning how the brain stores information but it doesn't have the sort of funding that cancer research has, so it can be a bit slow going. Also not to say that throwing money at it would solve the problem overnight, the brain as a system is absurdly complex
To "study" you will have to use some sort of (often natural, implicit, unplanned) spaced repetition. But you will also process the information with an eye to make it productive (as opposed to "just stored").
I know that flashcards were a big thing among students when I was in grade school and college (decades ago). I'm sure they helped students memorize pieces of information, but I'm not sure they helped students understand them.
On the one hand, I've downloaded AnkiDroid and played with it a bit. But I have the impression that just "seeing" the card isn't helping me. I need to be forced to write down the answer.
On the other hand, I've struggled to understand what "kind" of things should go in to a card. For example, verbs: I'm learning a language that has a complex verb conjugation structure. What I really need is to study the conjugation table. But instead the usual cards I download just ask the infinive form, and then display the conjugation table in the back - I'm not going to sit down and study the conjugation table when I'm swiping cards. When I sit down and study I do without cards, so I don't see the benefit...?
I feel I'm missing the trick.
See the Cloze Deletion section: https://docs.ankiweb.net/editing.html
2. Try to make the cards into real-world examples. For instance, don't add the conjugation table, create example sentences for each of the conjugations and make each one a card. For foreign languages, I try to add audio to my flashcards, too.
3. Always make your own cards. Downloading premade decks never works and you miss part of the learning process.
4. For things that don't necessarily lend themselves to being on a flashcard, don't just write the information. Try to frame it in terms of a question and/or in a form that mimics the context in which you'd use the information.
For example, let's say you want to memorize the periodic table. Don't just write:
Carbon - Atomic Number: 6, Chemical Symbol: C, etc.
Instead, create cards with questions like:
What is the sixth element on the periodic table?, Is carbon before or after nitrogen?, What element has the chemical symbol C? and so on.
The more "viewpoints" you have of the topic, the better. I've found this to be a solid approach for virtually any topic.
OpenSource and Libre!
If they are successful, you can expect to have an accessible (ie, couple thousand dollars) brain-computer interface covering a tiny portion of your brain sometime in the 2030s. New knowledge would necessarily need new connections between between neurons, ie you have to grow and connect neurons in a very particular and intricate pattern. I think we're still at the "just barely trying to understand wtf is going on" stage.
If you'd like a more in-depth intro to the state of the art, Jeff Lichtman's work on connectomics is a great place to start. Here's a 3-part lecture: https://www.youtube.com/watch?v=MtTOg0mzRJc
You can skip ahead to part 3, where he describes the absolute state of the art machine they are building which will be able to map the structure of a tiny slice of brain <1mm^3 part of a brain using advanced optics, robotics and machine learning. It's quite humbling.
Dayan & Abbott theoretical neuroscience (I found a pdf)
Spikes: Exploring the Neural Code
Slightly more useful might be lucid dreaming. I can imagine that a stable lucid dream would be great for knowledge synthesis.
The easiest reply would be "They are called books", or whatever container and format for information - and already there issues and risks are more than just possible, mitigated by filters one is trained to develop. The more acquisition of information takes the aspect of an "injection", a whole world of issues open.
From "I read War and Peace in one hour: it was about Russia", to all kind of mental poison (from "bad" notions to concretions of intellectual inadequacies): information has to be digested - processed and integrated.
--
So, on the point of view of feasibility: if knowledge implies diffused transformations (as opposed e.g. to installing an independent informatic file, non-integrated with its peers), that idea of injecting knowledge is absurd - it does not work that way. When you come to know that Paris is in that location, in that position on the map, you modify a number of notions: some directly related to Paris, others of all related entities - of the idea of France and its layout, of Montparnasse, of the Second World War, of the treaties there signed etc. Information needs to be digested to be productive.
I would suspect you are right though that the idea of a knowledge injection is absurd.
If an athlete takes steroids they are injecting strength in one sense but not like a movie that they take the injection and are superman 5 minutes later. Surely, there are biological processes that can't be sped up all that much unless we are talking about complete science fiction with total mastery of biology.
On a different note, the new Matrix really failed to captivate and imagine like the old movie series did.
The best practical way to achieve your end? Reform education. We are forced to learn so much useless, outdated, opinionated information in the 20 years of mental training we call education. Let people begin apprenticeships very early on, let them try actual jobs and tasks, not just tests.
A low tech noninvasive way? Phones. If our phones downloaded all the data they needed, theoretically they could make our decisions for us. We'd never need to know things as long as the phone can communicate precisely what to do, without the costly why.
Obviously, I wouldn't sign up for the latter unless it became necessary to stay relevant. I don't like the privacy and freedom implications of a device with all the "right answers".
There will be market forces to put there more than just the intended knowledge you want. And probably it won't be so easy to debug or clean as with plain text.
Can you only imagine what advances we'd make in "information upload" if we invested as much in Teaching Pedagogy as we invest in Neuroscience?
Nothing worse than a system not sufficient trained to deal with high noise environments
Probably something that the brain does very early in life, like around infancy, is to figure out for itself a way to store knowledge and learn how to store new things that it learns in that way. If each one makes it all up for itself, then it stands to reason that they're all different in unpredictable ways. The observed fact that people vary widely in overall intelligence, what they're good and bad at learning, and how they learn it, seems to follow naturally from that.
https://www.reddit.com/r/askscience/comments/jbt562/what_is_...
According to that, the current best-supported theory is that the human brain evolved the ability to think and reason in in a modern way long before they started actually doing it. The ability to actually think like that is only latent, and needs to be exposed to "recursive language" at the right point in the brain's development to actually have it. So if only under the correct stimulus does the brain develop the ability to think, it stands to reason that they don't all work in exactly the same way at a biological level.
Remember that in the Matrix story, a world in which they could download skills from the simulation and instantly become things by essentially just imagining them, so long as they stayed within its confines - was the world they were trying to escape. If you tell a 5yr old that their crayon drawing is beautiful, they become an artist, and for most people, that pattern doesn't change for the rest of their lives unless they become disillusioned and pursue their destiny as an existential hero.
Maybe there could be some neural interface that lets you transmit and replay brain activity through other bodies faster than language does now, or muscular electrostimulation that simulates say, deadlifting, but just as having a music collection doesn't make you a musician, I'm with Feynman in that I don't think we understand what we don't ourselves create.
In general though anything I have read says they work well even if not injured. They have just never caught on the way lifting weights has.
You can erase entire categories of knowledge by damaging the brain in particular areas. There is no reason to believe the opposite is not true.
Therefore it is a matter of modifying the state of the brain to learn or know something.
As a start, what "data format" would that downloadable knowledge be represented in? It would have to be interpretable by your brain and able to reference knowledge and other concepts you already know about - not just explicit knowledge, but also motor memories, etc. Unfortunately, all this stuff depends heavily on the experiences and learnings you personally have made until this point - so this is probably completely unique for every brain.
We know our bodies age and die whereas sustaining a computer for 100s of years via a part replacements is quite doable.
How quickly we forget, https://www.youtube.com/watch?v=Bg21M2zwG9Q
Most US based comics, movies, books, short-stories celebrate the hero's accomplishment _after_ the training period. The hero is already the _the best one around_ in whatever art/skill/etc. Sometimes it's a random event (spiderman, captain america, ironman) or the backstory is quickly brushed off (e.g. jack reacher, even Hiro Protagonist in Cryptonomicon to mention something more involved) in the past.
I believe this approach, gives instant gratification but creates that narrative that _things come for free_ and/or by _design_ (God, nature, randomness, whatever you wanna call it). Not saying one is necessarily better than the other, but IMO there's beauty in widely considered small wins as well, especially when they become consistent.
ps. A counter-example western Hero I can think is Harry Potter. There's too much shine there as well, but Harry ultimately acquires his skillset through struggle. Harry is not as proficient as others, but dealing with dangerous situations through the series grows into a formidable dueller.
Does he? It’s been a while since I’ve read the books but my recollection is that he never actually develops a ton of deep skill, or at least it’s never seen. His biggest accomplishment in magical terms seems to be his ability to do the patronus charm at an earlier age than most. Otherwise most of his success comes from 1) bravery, 2) friends, or 3) luck. At the end he chooses to be an auror, which I thought was odd because it seems like he’s poorly equipped to do that job.
It’s true that the books do not paint his journey as instantaneous (except his proficiency flying the broom). But they also don’t really paint his journey as ending in a place of impressive skill.
But I have seen nothing remotely approaching this idea of being able to download into the brain.
The TL;DR is yes but its insanely hard and we're decades away from knowing how to do this (if not longer). We still lack the technologies to interface with individual neurons. We're good at reading from a handful of neurons, we very good are reading from a population of neurons.
When it comes to writing there are a number of problems that need to be solved.
First we haven't cracked the "neural code". We don't have a unified theory of how information is represented across the brain. We do have some idea of what individual brain region are doing but the codes used by each region can be wildly different. I'm not as caught up on my neuro literature as I would like but its completely possible their isn't a unified code and therefore we'll need to learn how each specific neural population encodes information. The software analogy here is we're trying to reverse engineer a system running in product without the source code and each module was written in a different programming language that we've never encountered before.
Second we do not have mechanisms to write to individual neurons at the population level. The best we can do is akin to blasting a loud speaker at a crowd of people. We can push around the population but we cannot do much more than that. Software analogy here, what we can do its more akin to Steve Balmer yelling "Developers, developers, developers" than a class room instructor teaching students python. Opto-genetics is a potential path forward here but it requires controlling the genome of an individual and really only lets you target specific neural populations rather than any neuron in the brain.
Third, we still do not understand where the memories or information live exactly. We know what brain regions are involved but we're a long ways away from truly understanding how long term information is consolidated to the cortex from the hippocampus and other structures.
Finally, lets assume we know how to do all of the above (in rodents). Translating the science to humans is going to be a Herculean task for a bunch of reasons. Some of these reasons are scientific while others are ethical. The immune system of the human brain is more aggressive than in other animals and tends to reject implants more quickly. Next finding human test subjects is akin to sending someone to Mars. You're putting their life at risk and there's not true way for them to come back. Nearly all of the human subjects in studies that required invasive brain surgery were suffering from medical conditions so severe that the risk of the condition was greater than surgery and the research was piggy-backed on top of a planned medical intervention.
As compelling as that tech would be, the risks are crazy!
every culture values work
the op probably means that learning kung-fu takes _a lot of practice_. In the real world probably years not just a summer camp :-p
So we've got to fight that battle as well.
In my experience it works just fine. For example, using a geography deck to learn the location, flag, capital city, etc. of every country in the world. It also worked well when I taught my kids the multiplication table, the NATO phonetic alphabet, and so on.
> To create a cloze deletion note, select the Cloze note type, and type some text into the "Text" field. Then drag the mouse over the text you want to hide to select it, and click the [… ]
Select the Cloze note type, where exactly? I've searched both AnkiDroid and AnkiWeb (I was thinking that maybe I could creat cards from my computer and use them on my phone) and I can't find any button to make a note a "Cloze note".
Here is a video that walks you through it:
https://www.youtube.com/watch?v=d__MjidDz8c
When you create a card, it should say Card Type or Type at the top. By default, this is on Basic. Click it and you can switch to Cloze. Then highlight the text and click the [...] icon in the toolbar.
Especially with regards to solution architecture, you're often combining or comparing a huge (think factorial!) mix of tools, code, data, etc. Each with their own configurations, settings, gotchas, patterns, constraints, capabilities, hacks ...
Being able to quickly recall and locate URLs, code repos, docs, points of contact in a company or the wider community ... for me, it is the ultimate career moat.
This article on A Cloud Guru kind of encapsulates the sentiment well
https://acloudguru.com/blog/engineering/the-career-changing-...
Anyway, convenient texhnology for spaced repetition plus a second brain is amazing compared to earlier centuries.
In the future this will probably be a core part of OP's question .. adding things to the brain, automatically deriving connections, surfacing things into your "local" memory at the perfect time ...
* programming things that I use rarely enough that I don't memorize them simply from usage, but often enough that there is a time saving from having to DDG what's the exact way to do something.
* a couple personal information items like my credit card number (I don't like having it saved anywhere), my wife's phone number, my businesses registration number, etc
* moves for solving Rubik cube
* Bible verses
This would also enable flashcard/SRS workflows to interoperate seamlessly with other memorization techniques, like memory palaces or mind maps. Foreign language vocabulary is the rare example of a domain that doesn't need these things, so flashcards are perceived as "good enough" for it.