Mind Uploading is Closer Than You Think (with Nick Bostrom and Randal Koene)
there is this concept where we can achieve immortality by uploading our mind to a computer as an engineer I'm interested in the technology that's involved and while putting this video together I talked to many experts working in the field and what I've learned through these conversations is just mind-blowing it appears that recent advances in the scanning technology as well as in artificial intelligence making the possibility of Mind uploading more and more real day by day let's break down the problem of Mind uploading at first we have to scan the brain at a very high resolution in order to obtain the images of the Cross sections of the brain and then based on these images build a digital model this model will run on a powerful supercomputer where a digital version of me will continue to live on Happily Ever After this what we call mind uploading and it can get really confusing here because it sounds like we have to create a high bandwidth connection of the brains to the machines similar to what neuralink is working on but in reality it's about building a digital model of the brain that's somehow equivalent to the actual physical brain I am personally excited about this technology because it can eventually free us up from the limitations of our physical body which will not just allow us to live indefinitely but also to invent to engineer things without any limitation of time um of course mind uploading is also a way that people think of to get towards what's called substrate Independence and therefore a longer lifespan and then going beyond that this is really the question for Humanity as such is the ability to grow beyond our current limitations you know beyond the place that we evolved into to for example live in spaces where Humanity didn't evolve to be in such as in space or in virtual reality or on other planets I think there are some obvious advantages to being digital as opposed to biological um they might not appeal to everyone but just if one looks at some basic functionalities of the digital substrate first you have the easy ability to make backup copies just as with any other software you can just store it to file that's hard to do with a biological brain so we are kind of as biological creatures subject to to accidents and even if you could stop aging it would like eventually get run over by a bus or something whereas if you had a backup copy just stored you know every day or something like at most you would lose 24 hours the first problem is actually scanning the brain and here we're interested in both structural and functional aspects of it we have to perform a very detailed scan that captures all the neurons and the connections between them and ideally the momentum as well which pulse is sent at which frequency and I was surprised to learn that out of all Technologies involved in mind uploading this scanning technology at the moment is the most advanced we've already scanned four cubic millimeters of the mouse brain and the whole fruit fly brain now there are actually several methods to do this one of this involves using fmri scanning basically using strong magnetic field to create detailed images of your brain while it's still working fmri technology is quite useful for the medical applications but not so useful for the Mind uploading and the reason for this is simply as a resolution you can't see much the the smallest voxel size on an fmri that has a you know a Tesla level that's not too high to survive is is so rough it's it's so big that you're really only capturing very Global activity in the brain in contrast to fmri the usage of electron microscopy is the standard nowadays it's basically done using a very powerful microscope that uses electrons instead of light to produce a magnified image of the brain structure just this year scientists scan a whole Mouse brain in the most detailed fashion and this was a big breakthrough to summarize we're already getting pretty good at collecting the necessary data the main challenge is interpreting this data and then building a digital model based on it clearly this will still take quite some time that's the reason why the startup nectom is working on the technology to preserve human brains the idea is to preserve it for the future sure so whenever they mind uploading technology arrives we can scan and upload it to preserve the brain nectom has developed a chemical solution which can be injected in the body and preserve the brain by turning it into a glass but this would have to be done while a person still alive this could work for example for ill patients who decided to take part in this project this idea obviously generated quite some controversy nectome of course is interesting because they at least care about how to prepare a sample and how to prepare a brain so that it can be stored for a period of time and get to that time when these Technologies are really ready so that's that's an essential company for that main reason anyway nectum technology is for sure relevant for the Mind uploading but it's not the Mind uploading itself um but the companies that do exist that actually work on it so very very fresh very young is our own company NeXT up neuro we want to solve the problem of going from data to a model and we also want to be able to validate that this model is actually what we were looking for and uh and neuralink is working on neural implants so on BCI which is great especially for today's medical needs but it's also not directly related to the problem of open emulation because it's not obvious you know if you have a brain computer interface you just connect neurons in your brain probably not all the neurons but some neurons in your brain with a computer so they can talk to each other how does that make your mind exist in that other machine or somewhere else that's not clear at all you know we've had lots of kinds of BCI already where that's definitely not what happens from the discussion with Randal and various research papers which I read it seems that obtaining the data is not a problem the problem is building the Digital model from this data and then validating the model that this model is actually doing the same thing as the actual physical brain what's interesting when I was working on this video and researching companies working on this technology I was not able to find many and that's because it's still mostly the research which is happening across various labs and universities and in general the term mind uploading is not used in Academia because it sounds like science fiction and that's why there are not many companies working on this just yet because companies don't prefer to jump into the research but jump into the market ready technology instead the places where most of this is going on is not companies it's because this is very much still research it's Labs I think you really should look at for example the laboratory of Theodore Berger and Dong Song they have two Labs at USC University Southern California then there's also a group led by Professor Orel Lazar from Colombia he runs the flybrain lab and he has a lot of experience in this area of so-called system identification system identification is actually what we're just talking about but it comes from electrical engineering from I for example analyzing electric circuits or chips and trying to reverse engineer them and and he's applying this in neuroscience the whole flow of Mind uploading looks rather complicated right now so that's why I was thinking what if there is a complete the different way to do that what if there is a shortcut what if even without complete understanding of the brain a neural network can learn to emulate it meaning that it might derive the function based on all the outputs and that's exactly how neural networks learn anything they basically observe a lot of output data which is coming let's say from a black box and then they will reverse engineer the function itself based on this output data it is conceivable that even without looking inside the brain but just looking at the output traces of say human that the text they have produced Maybe video feed and recordings if you had enough of that uh like like a couple of Decades of like if you imagine recorded everything the the particular individual did and then you had a super intelligence looking at all that information and trying to infer from that what would the brain have been like that would produce this data set it seems conceivable that you might get quite close but what if we can use it for the Mind uploading to solve the biggest bottleneck the modeling problem so for instance um there is another group that is trying to create a company just like we created next up neuro who specifically care mostly about using llms large language models and similar Machinery tools to try to work on on over emulation and and that's not wrong I think that's a great path to take because llms and other machine learning tools definitely have a very important role to play in making Neuroscience more systematic faster and better when we eventually come to a digital model maybe with the help of AI the next huge problem would be to validate this model let's say a neural network has created a certain structure in a brain to some function but a neuroscientist is not certain what exactly its function is it's just some type of a neural activity so and this is the biggest problem how do you validate something which you can't understand so in the case just getting back to to why this is a problem or not a problem why you just need to be aware of what you can use it for and what not in hibernation is that these llms are generative Transformer models are very useful in filling and missing data where they know what to do so they can for example if you do an electron microscope scan and there's a part where there was noise or something went wrong with a tissue it can help fill that in with what was probably there and that's quite useful but it gets a little risky if you rely on the llm to to make interpretation stations or to generate inferred output for the whole of an individual brain and that's because the llm will have a tendency to erase individually unique characteristics of the data which of course is what you actually want you want to have an emulation of that individual human brain not of a general human brain or a likely human brain you want that specific unique human brain and then if you do what they call adding some temperature to make this llm more creative then it has a tendency to confabulate to make up things that are satisfying to the user or the scientist in this case without actually being testably correct mind uploading possess a massive technological challenge but let's say we figured it out and we managed to replicate me my brain one-to-one digitalia the next question is will it be conscious or not if we replicated all the cognitive capabilities of my brain in there it's hard to say why Consciousness would not emerge what is so special about it and that's really hard to answer because in reality we don't really know what Consciousness is well I think I'm a computationalist meaning I think in principle conscious experiences can be implemented by a wide range of different substrates it doesn't have to be hydrocarbon chemistry it could easily be silicon substrate and that what makes something conscious when it is conscious is the kind of structure of the computation that is being performed you have to have not just a structure but you have to actually run the structure like run the computation so just having a file of like a scan of a brain stored you know on a hard drive somewhere like wouldn't be continuously having conscious experiences knowing that it's just sitting there just think about that if the same computation on the computer which is twice as fast as me I could have more mental experiences in the same 24 hours or I can produce more results or I could have more fun and this sounds appealing one hand could be a path towards Advanced forms of machine intelligence I mean basically you have two paths you could either do artificial machine intelligence where we create sort of synthetic programs and learning algorithms or you could just try to copy and paste from biology if you could transfer a human mind into a computer then if you then run that mind on a faster computer or if you make a lot of copies you could potentially have a powerful form of machine intelligence now it does look in the world we are actually in we'll probably get to machine super intelligence first through the artificial path given the remarkable breakthroughs we've seen in AGI in recent years and then I think this machine super intelligence will maybe develop the technology to enable mind uploading it's clear once mind uploading is here it will be huge it will change the society forever and hopefully we'll have a lot of positive outcomes but let's think for a moment what happens if it goes all wrong or this technology is being misused because we know that there is not a single technology which cannot be misused so the question really is if you're thinking about ways that could go wrong you know just like with other Technologies say for example someone discovers splitting the atom how can it go wrong there are many different ways that can go wrong so hope Innovation mind uploading could also be misused if it is not set up up in the right way if there isn't regulation if there isn't oversight if there are no security measures put in place then you're going to see people abusing it one way or another whether it's too like play advertising in your mind or read your mind and use the things in your mind for their benefits or whatever or clone you copy you for when you don't want to be copied another which I also actually described in you know my book super intelligence which is now almost almost 10 years old uh there's a chapter there about a world where machine intelligence is produced through um mind uploading and and you then have a big population of uploaded Minds that are competing in an economy and competing to make more uploaded Minds um and where in order to actually succeed in this competitive environment they need to streamline and optimize their minds so it turns out that for many jobs maybe you don't need certain parts of the brain and so you wouldn't have competitive pressures to sort of extirpate and remove all Superfluous elements that maybe are valuable from from from the perspective of what we want and care about but that might just cause you not to be competitive in in this kind of hyper economy of uploads so just like this artificial intelligence we would need regulations let's say what will be the social status of this mind and whom the rights to this mind will belong to let me know your thoughts in the comments I think at the current pace of development it might take us another half a century but this does not mean that we don't have to prepare already now in terms of legislation and understanding the social impact if you look at it in terms of the time spans that become commonplace in Silicon Valley then it's still very much in the research stage and so consequently it's very far away um something I sometimes say is that I'd be very thrilled to see it in 20 years but I would be surprised and disappointed if Humanity couldn't achieve it by the end of the century and of course the advances in artificial intelligence are expected to speed up the pace of the current technological development I think once you have you know radical super intelligence then you really get the kind of telescoping of the future where all those developments that maybe we could have pulled off in the fullness if we had you know 40 000 years to work on it we would have probably perfect virtual reality and uploading and cure for aging and space settlements and whatnot right but all of that progress in 40 000 years might then happen on digital time scales you know in four months or four years once you have a super intelligence doing the technology work another very interesting point here once mind uploading is successful it will help us to gain some clarity with respect to the simulation Theory as you've likely heard there is this theory that we all live in a simulation in a beautiful simulation which may or may not be true however if we can make a mind upload and then run a simulation this would increase the probability of simulation Theory a lot and this would be cool I think it would um feed into the simulation argument by strengthening uh our reasons to accept the third possibility there which is the simulation hypothesis that we are ourselves currently in a computer simulation um because it would demonstrate that we had the technology for creating conscious digital people with experiences like ours like whole brain emulation would be one way of doing that then you would have the technological capability to do this then if you combine that with the assumption that there would be some civilizations that were actually interested in using resources um to produce these simulated Minds then then you kind of can check off another box and then you know there's a little more footwork there but basically you would increase the probability of the simulation hypothesis being true the closer we get ourselves to being able to create and I call them ancestor simulations um the greater the probability that we are ourselves currently living in such a simulation even if we can upload our mind the question whether we should remains open let me know if you would like to be uploaded in the comments below and they will reach out to you just joking stay curious ciao
2023-10-06 19:02
