HACKER Q&A
📣 mr_coffee

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:

https://www.youtube.com/watch?v=6vMO3XmNXe4


  👤 keiferski Accepted Answer ✓
I know it's not exactly what you asked for, but in my experience, spaced repetition is borderline science fiction -- if implemented properly. It's not as instantaneous or cool as The Matrix, but it does feel like magic when you can remember obscure words or concepts years after entering them into a SRS program.

👤 martythemaniak
No, we are very far from anything like that. I would watch Neuralink's recruiting events for a sense of where things are. They are trying to make a robot that will insert dozens of threads with several hundred tiny electrodes on these threads. This will allow reading a high-resolution picture of that part of the brain's activity. You can then hook this up to a computer and train yourself to map your brain's signals to some desired output (typing, moving a wheelchair, robot arm, etc). They talk about "writing", but that is just providing feedback, ie they send in signals and you train to interpret them as something.

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.


👤 beardyw
It seems that no one properly understands how information is stored in the brain. I think that needs to be solved first.

👤 ahmed_ds
This reminded me of Sleep Learning. This topic has the oddest Wikipedia page [0]. You would expect this topic to have research, historical background and other sections. But 80% of that page is just references to fictional literature.

0: https://en.wikipedia.org/wiki/Sleep-learning


👤 mdp2021
Not a word about the security implications?

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.


👤 bArray
I have thought for a while now that VR would allow for this kind of training. You essentially control a large proportion of the input and can measure the output. We could start training human brains like we train neural networks - throw tonnes of data at them and then test what parts stick, then repeat until the training set is learned.

On a different note, the new Matrix really failed to captivate and imagine like the old movie series did.


👤 Dumblydorr
Cybernetic implants into the mind are in their infant stage. We have electrical and magnetic stimulation treatments being tried all around our healthcare system, but these are to modulate the brain and help it normalize for dementia, depression, etc.

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".


👤 gmuslera
Think in how the Pentagon and CIA shaped movies into propaganda (https://worldbeyondwar.org/the-pentagon-and-cia-have-shaped-...). Now think how something that puts directly information in your brain, without any critical analysis from the one that receives it. Or just plain old advertising or marketing.

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.


👤 GenerocUsername
We have a high bandwidth input already known as eyes and ears

👤 ufmace
I'm not any kind of neurologist, but intuitively, the ways in which people seem to learn things varies so much from person to person, it seems highly likely to me that there is no universal format in which our brains store knowledge of facts and movement patterns. This bodes rather poorly for ever doing something like downloading knowledge to the brain Matrix style. Even assuming you had a way to "write" whatever neuron structure stored knowledge, you'd have to first carry out a super-detailed analysis to determine exactly how to encode that knowledge in each particular brain. And God only knows what'll happen if you get it wrong.

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.


👤 motohagiography
No, but a philosophical argument that showed how it was impossible would be almost as useful. You can just give someone a magic feather and let them believe they have become something, which is sufficient for many purposes.

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.


👤 xg15
Not a neurologist, but my hunch is that even if we had perfectly working brain implants, this would be impossible the way they do in the movie - or at least would require lifelong monitoring of your brain structure before it could be done (which, arguably, would be feasible for people spending their life in a tub like in the movies - but infeasible for us)

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.


👤 teatree
Wouldn't it be easier to understand how the brain works and transfer intelligence and consciousness into silicon (by emulation?) rather than try the reverse?

We know our bodies age and die whereas sustaining a computer for 100s of years via a part replacements is quite doable.


👤 WHA8m
The thread has plenty of replies as of now so I don't feel too bad for playing devils advocate here. Can't you either "download from" or "upload to"? "Download to" just sounds strange. Anyway

👤 sys_64738
Keanu Reeves was also in Johnny Mnemonic which had a similar concept.

https://en.wikipedia.org/wiki/Johnny_Mnemonic_(film)


👤 walterbell
> provide "instant knowledge", like learning kung fu

How quickly we forget, https://www.youtube.com/watch?v=Bg21M2zwG9Q


👤 _448
Few years ago my manager gave me a magazine as a gift(the magazine might still be lying around somewhere in my room). The cover story of that magazine was this very topic. The magazine covered research done by a university in the US on copying and then downloading memory back in a mouse(the animal). The justification for this research was that "it will help patients who have suffered memory loss".

👤 ffhhj
Two examples of skills you can quickly obtain are speed-reading and fast silent running. And actually speed-reading is most similar to The Matrix knowledge download. The mind is pretty impressive at character recognition, we just don't use the whole potential. Also muscular programming allows defining the angle, speed, and strength/flexibilty sequence of each feet to run silently at full speed.

👤 ModernMech
God I hope not. For one, it'll put me out of a job as a teacher. For another, you know that it'll just be used to inject corporate propaganda and advertisements into our heads, not cool skills like kung fu and how to fly a helicopter.

👤 tmaly
I think there is still some low hanging fruit to be had in terms of how things are taught today.

But I have seen nothing remotely approaching this idea of being able to download into the brain.


👤 TrinaryWorksToo
Yeah they're called teachers. Some are good, some are bad.

👤 dusted
I don't think enough is understood about.. understanding to even start thinking seriously on this.

👤 flint
Read "Fall; or, Dodge in Hell" by Neal Stephenson https://en.wikipedia.org/wiki/Fall;_or,_Dodge_in_Hell

👤 bognition
Really great question! Former PhD in Neuroscience here. I spent many years building systems and algorithms to decode neural activity in rats. I've been out of neuro for almost a decade now so some of this is out of date, so take this all with a grain of salt.

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.


👤 ddingus
Infohazard comes to mind...

As compelling as that tech would be, the risks are crazy!


👤 Madmallard
No one has any clue how to do this nor will they (likely) ever.

👤 nonameiguess
Caveat here that I have no specific expertise in this area.

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.


👤 Juliate
It'd be surprising that this category of knowledge would be all/only brain-located.