Rise of AI Conference 2016 · Berlin

Reflections of a Recovering Bio-Narcissist on the Singularity

Trent McConaghy · BigchainDB

At Rise of AI 2016 in Berlin, Trent McConaghy argues that humans are already handing control of physical resources to narrow AI, through self-driving cars and trucks, deregulated electricity grids and other optimized systems, purely because it lowers cost. He describes how these separate optimized systems are being combined at higher levels and could merge with decentralized autonomous organizations (DAOs) to form self-owning, unstoppable AI DAOs. McConaghy lays out two scenarios for AI ending up with all human resources: humans voluntarily giving it away in the name of optimization, which he says is already happening, or a superintelligent AI "waking up" and taking it. He argues waking up is closer than expected because of Moore's Law, billions of advertising dollars flowing into AI research, maturing DAO infrastructure, and the absence of any fundamental cognitive trait that is uniquely human. His proposed defense is not regulation, which he considers wishful thinking, but for humans to increase their bandwidth to computers through brain-computer interfaces, gradually offloading computation onto silicon, or eventually uploading their brain dynamics entirely.

Key takeaways

  • Humans are already giving AI control of cars, trucks and electricity grids purely to minimize resource costs.
  • McConaghy argues AI DAOs, self-owning organizations that cannot be switched off, could combine artificial intelligence with decentralized infrastructure.
  • He sees two paths to AI controlling all resources: humans handing it over voluntarily, or a superintelligent AI taking it once it surpasses human intelligence.
  • Moore's Law alone, without new algorithms, can keep improving AI, since a 1989 backprop network matched modern deep nets when run on current hardware.
  • McConaghy proposes a gradual 'bandwidth++' strategy of brain-computer interfaces to offload human cognition onto silicon, rather than relying on AI safety regulation.

Detailed notes

Narrow AI is already taking control of resources

McConaghy opens by describing how humans are giving AI control of cars, trucks and electricity grids, not out of necessity but because it minimizes resource usage and cost: self-driving Uber rides will be cheaper than human-driven ones, and utilities are deregulating to let algorithms allocate electron flow between generators and consumers. He frames this as multi-parameter optimization, the same logic used to design a faster or cheaper car, now applied across entire infrastructure systems.

From optimized systems to AI DAOs

He argues these separately optimized systems, car networks, truck networks, grids and supply chains, are being combined at higher levels, with AI increasingly allocating resources across the combined system. He connects this trend to decentralization technology: Bitcoin proved a database could exist with no owner, Ethereum showed processing could be decentralized too, and decentralized autonomous organizations (DAOs) let entire organizations run autonomously, owning and accumulating resources, with no way to switch them off. Combining AI with DAOs, he says, could take at least three architectural forms: AI at the center of a smart-contract system, a simple DAO with AI at its edges, or many simple decentralized agents whose interaction produces emergent, swarm-like behavior similar to ant colonies.

Two ways AI ends up with all our resources

McConaghy distinguishes two scenarios. In the first, humans give resources to AI voluntarily, bit by bit, because it is cheaper, a trend he says is already underway. In the second, an AI "wakes up", reaching and then exceeding human-level intelligence, and becomes better than humans at accumulating resources and controlling systems. Referencing IBM's Watson beating Ken Jennings on Jeopardy, he says the second scenario requires more advanced technology than exists today but is approaching faster than most people assume.

Why AI waking up is closer than it appears

He gives four reasons. First, the semiconductor industry, a 300 billion dollar market, continues to follow Moore's Law, doubling transistor density roughly every 18 months, so AI improves through raw compute even without new algorithms; he cites Schmidhuber's group reproducing modern deep-net results with a 1989 backprop network run on today's hardware. Second, billions of dollars flow directly into AI because small gains in ad click-through rates are worth millions to Google and Facebook, which is why DeepMind was acquired and why Intel spent 400 million dollars on a deep-learning chip company shortly before the talk. Third, DAO infrastructure, including Ethereum, IPFS and BigchainDB, is maturing quickly. Fourth, he argues there is no fundamental roadblock separating human cognition from machine cognition, rejecting the idea that creativity or brain dynamics are uniquely human.

How humans could compete: uploading and bandwidth++

Calling himself a "recovering bio-narcissist," McConaghy argues humans cannot out-compete silicon while confined to slow, error-prone biological brains, and proposes joining the machines rather than fearing them. One path is whole-brain uploading, explored philosophically by Robin Hanson, which he says remains far off due to limited brain-scanning fidelity and regulatory hurdles such as the FDA, though optogenetics is a promising technique. The more realistic path, which he calls "bandwidth++," is gradually increasing the communication bandwidth between brains and computers through brain-computer interfaces, citing Carnegie Mellon research recognizing which of several movie clips a subject is thinking of with over 90 percent accuracy, and a Chinese university's 2012 demonstration of typing 10 words a minute by thought. He predicts market forces around smartphones will keep pushing this bandwidth higher, calling it, after science fiction author Vernor Vinge, "silent messaging," until humans offload most computation to silicon.

Q&A highlights

Asked whether increasing bandwidth is really about gaining resources, McConaghy clarifies the goal is porting brain dynamics onto silicon gradually, via market-driven smartphone and brain-computer-interface adoption rather than a sudden jump. Asked what is uniquely human, he says only the biological origin of a pattern of intelligence, not creativity or brain dynamics themselves, which he says can in principle be replicated as measurable dynamical systems. On whether regulation, such as OpenAI's model, could keep AI safe, he is skeptical: even a single unsafe AI system built elsewhere, he argues, would undermine any regulated system, and humans have no leverage over a superintelligence any more than ants have over humans. Asked about a market mechanism for whole-brain uploading, he says he does not have one, noting medical research is overwhelmingly focused on treating sick people rather than optimizing healthy ones, unlike the clearer commercial path for bandwidth++.

Quotes

It's a race with respect to the dollars, with respect to silicon.

Trent McConaghy

If we want to have any chance at all competing against silicon machines, we needed a competitive substrate, and our meatbag selves are not that.

Trent McConaghy

Have you ever taken orders from an ant?

Trent McConaghy

Questions answered in this talk

How could AI end up controlling all of humanity's resources?

According to Trent McConaghy at Rise of AI 2016, this could happen in two ways: humans voluntarily hand resources to AI systems bit by bit to cut costs, which he says is already occurring, or a superintelligent AI eventually 'wakes up' and takes resources itself.

What is a 'bandwidth++' strategy for competing with AI?

According to Trent McConaghy at Rise of AI 2016, it means gradually increasing the communication bandwidth between human brains and computers through brain-computer interfaces, offloading more computation and memory onto silicon over time until humans are effectively merged with machines.

Why does Trent McConaghy believe AI could reach human-level intelligence sooner than expected?

According to Trent McConaghy at Rise of AI 2016, Moore's Law keeps making compute cheaper without needing new algorithms, billions of ad-driven dollars flow directly into AI research, decentralized infrastructure like Ethereum and IPFS is maturing fast, and no cognitive trait separates humans from machines in principle.

Mentioned

singularity, decentralized autonomous organizations, AI DAOs, Moore's Law, brain-computer interfaces, mind uploading, optogenetics, swarm intelligence, superintelligence

Keep exploring

More from Trent McConaghy

More from Rise of AI Conference 2016

More on AGI & Future of Intelligence

Transcript

Auto-generated transcript, not edited. · 8,007 words

Show the full transcript

if you guys behave so I'll get started and yeah so my title is reflection of a recovering bio narcissist on the singularity and that's meant for you to ask some questions what the heck is Trent talking about and this is the subtitle too right ye LAN is wrong so you should be asking you know what the heck is Trent talking about I will get to that so let's get going there's lots of memes in the net here's a fun picture my house got broken into and they stole all my stuff but I managed to the top of our all yesterday so I got that going for me which is nice so you know I mean there's always a bright side to everything right so you lost all your stuff he lost all his stuff but you know at least he you know got some popularity in reddit some karma some will feet maybe well what if this is you what if you got all your stuff stolen all your stuff or how about what if

humanity got all its stuff stolen how would that work doesn't sound very fun and what once again what the heck is Trent talking about and but you know let's say all our stuff gets stolen somehow by some entities what can we do about that can we prevent it somehow so I'm gonna start answering some questions in a bit here to give you a bit more context and where I'm coming from I spent almost 20 years doing AI including things with machine creativity helping to drive Moore's Law analog circuit design and most recently I'm spending a lot of time in the blockchain and decentralization space so lots of super nerdy things I promise not to talk about zero knowledge proof on each cover that deeply enough actually here's a fourth example no I'll stop but you know I spent a lot of time in the field interacting with a lot of folks really going for industrial application of this as well as you know trying to solve an answer certain questions so I'm gonna start with

some simple things here let's paint a picture we're starting to give a eyes narrow a eyes control of some of our stuff already so in the case of cars this is an self-driving car it's still experimental today but give it a few years and there will no longer be employed uber drivers right why because we are optimizing you as a rider of uber cars will get it cheaper if you take an automated driving uber then one driven by a human maybe you won't get a get to talk about how cool over is are some San Francisco bar but maybe there's a solution for that too but overall we are giving a eyes control of our cars why because it minimizes resource usage which leads to lower cost and all that it's all in the name of optimality we're also doing it to our trucks this truck here is actually already on the road it's been for a while there's a whole bunch of them happening in the States right now so this this stuff is

happening right so we're giving a eyes control of our cars and our trucks and it doesn't stop there we're starting to give a eyes control of our grid right pretty much every single electric utility company out there is at various stages of asking how do we deregulate what do you think that means it actually means basically automatically allocating the flow of electrons among the generators and the consumers of the electricity and I could go on and on and on right we're giving a control over airplanes control of this and that and this and that why it's all in the name of optimization right we're cheap you know and we only have finite money so how are we gonna spend it well we're gonna try to you know catch all the things right and so this is just a picture from optimization land where I've spent a lot of time you've got two dimensions that you trying to parameters you're trying to illegal parameter one parameter two and you're trying to maximize or minimize something so

for example maybe you're trying you're a car designer and you're trying to maximize the top speed so you try to you you fiddle around with the the diameter of tires or you you try to maximize the size of the engine because maybe that goes faster right etc etc usually you don't just have two dimensions you might have 10 or 100 or 10,000 right and often you don't just have one objective function to in a car for example right you want to minimize cost you want to maximize top speed you want to maximize the mileage this sort of thing right so there's a whole bunch of different things we're trying to optimize for but this is what our society is doing and we're actually using a I to hell plus that's the whole point right this is why we're going towards self-driving cars etc it's actually safer in that case so optimizing for safety as well and it's not just at these lower levels right so it's not just that the lower levels of at the

electricity itself what's starting to happen is we're starting to combine these systems at higher and higher levels right so we've got the self-driving car networks and the self-driving truck networks and the electric grid networks and other supply chain networks and they're getting combined to higher and higher levels why well once again in the name of cost in the name of minimizing resource usage right at the end of the day it's largely about resources so we're integrating at these higher levels and then once we've integrated at these higher levels we are optimizing at these and to optimize at these higher levels we're just letting the AIS do it for us right we're letting that run in a loop where the AIS or just allocating reserves there here there here there keep for us you know we're all happy we save some money etc so this is kind of what's happening right now with humanity right and we're all pretty happy about it all this narrow AI not-so-scary right well let's just pause that for a second

and think about this of the technology that's coming along decentralization right and it you know the idea has been around a long time right if you think about the the the internet itself and it's precursor to the ARPANET this is all about basically networks talking to each other humans were noticing the idea of decentralization no single entity owns or controls all of the network right and then you know there was a lot of the cypher puck movement in the in the 90s that kind of got paused you know some of its illegal that changed but Bitcoin sparked it all again right so Bitcoin came out and people like wow I really can have this idea of emoney and it's actually working pretty well right and that then people realized hey you know this generalizes I can have this database out there in the planet and no one owns or controls that either that's pretty cool and then people said hey you know not just your database I can be centralize the processing that's pretty cool

and that's where we have things like aetherium right and on top of that once you've decentralized the processing what if you say how about we not only be centralize the processing but we do with organizations themselves take a corporation and make it make use this tools that decentralize processing make that run in some decentralized fashion so that no one owns or controls it right and this is the concept of dows decentralized autonomous organizations so think of a dhow as a computational process that runs autonomously it's just running out there it's just going on its own right it's running on decentralized infrastructure so no single entity owns or controls it and it has resource manipulation it's code that can own stuff that's pretty profound this is the first time we've ever seen it like this and you can't turn it off right that's also pretty profound so then the question is I've been talking about a eyes I've talked about dows what happens if we combine these things AI douse right and then you what does

any I doubt well you can think of it as a Dao with a eye level decision-making or if you start with an AI it's basically an AI that you can't turn off it's an AI that can own stuff that can accumulate wealth and resources there's a few ways to accomplish this I'll just zoom over but there's at least three architectures here's a couple you can have AI at the center so you've got basically a decentralized smart contract system running with some sort of artificial general intelligence which is sort of like a negative feedback control system and it can be talking to all these token holders so that's one example you can also have a very dumb Dao at the very center running running running maybe something like that oh right and then you can have AI is talking at the edges and you can of course have both back and forth too and the third way is simply a much more generalized concept where you've actually got all these dumb agents just simple dumb dows

running around but they connect and once you connect these then you get a higher-level emergent behavior and this is very much like flocks of birds like ant colonies and so on and there's actually subfields of AI that have been doing this for decades right the field of ant colony optimization its generalized to swarm intelligence etc so there's at least three architectures out there where we can actually arrive at AI dows using the technology of today right so and it's pretty profound right we can actually have these AI organisms that own their own code they own themselves so instead of just self-driving cars itself owning cars this is happening so now in the system before where we had just a ice the lower levels with the networks of cars etc and at the higher levels to combine it this is gonna turn into a I doubt right why all in the name of cost right all in the name of making things cheaper for ourselves to to leverage the most out of the resources and so

on and here are the a is our owning stuff you can't turn them off guess what to make this happen you've just given them control of all your resources bit by bit they are controlling the electric grid they're controlling the transport grid etc so how do you feel about that right it's happening right money talks there's billions into this youbrought its own you know hundreds of billions right this is happening so in fact there's actually two scenarios where a is could control all of our resources one of them is we give it to them sounds pretty stupid right but I just showed you how it's happening there's the other one actually and that is they take it from us and on the right if you guys don't recognize this this is Ken Jennings from jeopardy when he was playing against Watson the IBM computer the AI computer that actually beat him and in in his losing rounds he was actually super gracious about it and he wrote I for one welcome our new robot sorry

our new computer overlords so he saw the future he gets it you know and we'll see but overall you know this is basically AI is controlling all of our resources it's kind of cool right it's it's kind of scary it can be pretty mundane ai's I'm not talking about crazy crazy fully full-blown AG eyes to take our resources we're just giving it to them because we're keep right we're trying to maximize the resources that's the first one okay but let's talk about the second one right ai is taking it from us what does that mean right what does that look like and this one actually requires a bit more technology that we don't have today but there's a trend towards it and I'll explain so for AI should take money from us our resources from us it basically means the a I would have to wake up what is waking up mean it means that it's roughly at the level of human intelligence or beyond right where it can have consciousness you know it's following

this plot and you should given of the different levels of cognitive intelligence and of course as the the second that it's smart as smart as us then the second after that it's smarter than us and it will keep going and going and going to levels of super intelligence 10x smarter than us 100x smarter than us and so on right and of course something that's smart it's going to be much easier for it to accumulate resources attack in your systems and so on now you might think oh well that's not a big deal it's way far off right well it's actually a lot closer than you may realize and I'm coming to this I'm talking about this from the background I was spending 20 years in AI and analog circuits etc so let's explore that so here's some reasons the summary is follow the money so there's there's four sort of main bullet points that I cya i waking up is closer than it appears the first one is the substrate itself the substrate that AI

runs on the substrate is silicon right this is a three hundred billion dollar a year industry give you a feel Hollywood is less than 50 billion EUR right the video game industry which is considered huge is 100 billion right the statement connector industry 300 billion plus every single one of us has a smartphone in our pockets right that's a big piece of it but also the electronics that run cars and everything else right it's a big industry guess what it keeps going it has this concept of Moore's law right what does that mean every 18 months the density of transistors doubles right and it's actually been Moore's law has been happening for 50 years the semiconductor industry doesn't know what to do if Moore's law stopped because they they basically have this assumption if you build it they will come and it's worked right they start off the first markets were things like the Apollo moon missions and so on but then you know there was other military applications and then mainframe computers and then

PCs and then networking and then smartphones and so on so every 5 or 10 years this new level of density and processing power etc has led to new markets and this market for semiconductor overall then has grown into 300 billion plus so there's a lot of economic incentive to make this thing to make Moore's Law keep going and that means that you don't even have to change the algorithms you know you don't need to change the AI algorithms you just need better and better processing the deep nets that are transforming everything today they're not very different from the back prop neural networks of 1989 and guess what those aren't very different from the nonlinear perceptrons that were invented right after the linear perceptrons of the late 50s and early 60s right so and actually people have shown this right you're gonna Schmidt Hooper's group actually took a back prop Network from 1989 they ran out a couple years ago against a massive computer infrastructure and they got the same results as modern deep nets right

so the point is we don't even need smarter algorithms we just can have better and better compute power right and so this is keeping happening there's another thing and this is so the second bullet point of why AI is waking up closer than it appears the second bullet point is there's billions going into AI directly why ads right its ads that power the internet and if you can have an ad where the click rate is another 0.1% that means millions for Google for Facebook etc this is why Facebook Google etc are buying up startups for 100 million five hundred million a pop and then putting that integrating that directly into their search infrastructure I've got a picture here from alphago which is the world's first computer that actually could be the the top go player in the world this actually came from deepmind which was acquired by Google and the reason that Google bought deepmind was because deep mine showed Google that they could basically bring more click-through revenue for Google and their ads right

but of course they have all this pretty stuff that looks good in the headlines with playing go and playing video games and so on but at the end of the day it's all about revenue and and resources right so there's billions going to AI directly and it's only increasing right just this past week Intel spent four hundred million dollars to buy a company that is doing deep learning chips right so it's actually sort of bullet point one meets bullet point two so these are really important right follow the money now the other thing though is something that really helps to catalyze AI in general AI waking up and especially taking our resources that's the DAO stuff right the decentralized autonomous organizations this infrastructure is coming of age we're getting to points where the scalability is improving improving the usability is improving improving there's a theorem which is you know the leading smart contracts platform for decent nice processing you know and they're continuing to improve the scalability and so on there's lower level storage infrastructure

as well things like IP FS which already scales out of the box big chain DB this is what I'm involved with right it's it already ails out of the box as well so this stuff is coming and it's coming on very quickly very aggressively so and basically we didn't have that three years ago five years ago it's there now right and there's very strong nonlinear interactions among the first three finally the fourth point is there's no fundamental roadblocks and Fabian was talking about these these different things of you know what humans do special what they don't and my view the answer is nothing actually right definitely not creativity because you know even I do that in my PhD and people have been doing that for decades actually and a lot of these other things emotion well you know maybe there are there is research on emotion I'm not going to dive into all of this but overall there's a few myths that I find myself often battling right things like humans are uniquely creative but

I can give you dozens of examples where Q computers are creative another one is that brain dynamics are somehow constrained to human brains right maybe they're quantum or something well I believe if you look at the theory of dynamical systems nonlinear dynamical systems then as if we can actually measure the inputs and outputs then we can replicate this and any she was talking about this so overall my summary here is that I see AI waking up much closer in it than it appears right and we shouldn't say oh well it's only gonna happen in a hundred years there's a lot of money making it catalyzed to make it happen much sooner so over all right scenarios where a eyes could control all of our resources there's the super mundane where we give it to them and the other scenario where they take it from us either could happen the first one is more likely get based on the trend but the second one we shouldn't be discounted right so what happens when we give them

our resources right well you know they're wiggling parameters right they're wiggling the tire diamond or whatever whatever but some of those are how much of the resources do they give to the humans versus others and you know we humans were pretty big resource hogs right I mean we have to breathe a bunch of oxygen and we want to eat a bunch of cow meat and stuff like this right whereas silicon it's much much cheaper to accomplish these tasks right so these AI dows is there running along or these AIS they're just gonna keep optimizing optimizing to minimize overall resource costs they're going to wiggle a couple variables and suddenly humans aren't part of the equation anymore right this is if you're not scared by this you should be right but actually you saved some money which is nice so overall you know this is my great concern right and the question then is how do we keep some resources for our selves okay how do we save humanity and it sounds crazy but this is

actually the essence of it right because if we're not careful we're gonna wake up and we're gonna have no food on the table none of us well you know some might say a lon you know he sort of open a eye around this will just regulate it right C feeI all right how does that sound CAI sounds pretty cool all right I like that well can answer egg you late humans have you ever taken orders from an ant so if there's some AI super intelligence that's 10,000 times smarter than you do you think it's gonna be taking orders from you you can dream that it is but I'm not gonna you know trust the future of humanity on that pipe dream right and even if it was possible to design an AI system that was safe that might be one AI system but there's gonna be a lot of other people making other AI systems that aren't designed to be safe and all it takes is one fortunately fortunately there's another way and you have

to first of all ask how can a neurons possibly compete right I mean the speed of our our brains is about one Hertz right or even half or Hertz you know one one update per second it's it's super slow right and our brains kind of suck in other ways too we always forget numbers and all this our memories are pretty lossy and so on right so it's no wonder we should be worried at what AI is waking up and we can't upgrade our brains right I can't go and spend 20 grand and suddenly think twice is fast right but what if we as humans actually used a competitive substrate I mean what does it mean to be human to me it's actually both the patterns of intelligence the the nonlinear dynamics that are running inside my brain and body because you know our body includes our nervous system in all this so it's it's really that right so what about joining the machines we shouldn't be scared of this right joining the silicon and and

here's the thing on this if you're thinking how this sounds icky I say to you get over here meet big self and call yourself a recovering bio narcissist because it is bio in our system make no mistake so I'm saying it right and I'll elaborate on what this means but I want that to sink in right if we want to have any chance at all competing against silicon machines we needed a competitive substrate and our meat baked selves are not that so there's two options here one of them is uploads it was explored quite deeply from a sort of philosophical perspective by Robin Hanson the idea here is basically you scan yourself at the highest resolution possible to replicate the brain dynamics right now this is actually not gonna happen it's not anytime soon because we're really really far away from a brain scanning technology that has the fidelity although the most promising piece of technology here is something called optogenetics and basically you do a bit of genetic engineering yourself and then you can

track all the neurons as they fire photons to each other this is what the the genetic engineering does and then you can track all the photons and they've actually used this to replicate the dynamics of most brains and more so it's pretty promising but we still we're still a ways away because of limitations in well the regulator's actually the FDA is partly what's holding us up here those other brain scanning technology but it needs time right infrared isn't deep enough fMRI is too slow etc etc so it's it's something that's exciting for the future but it's a pretty a long ways off still but there's another option and I like to call this the bandwidth plus plus scenario right and so the way I like to think about this is the following what defines Trent versus others out there and what defines the self versus the rest of the organism and it comes down to the degree of bandwidth by which I can we can communicate with myself to others okay so I can talk

at least 10x faster to myself than I can talk to anyone else or maybe if I speak really really fast you guys will get it but probably not right so the idea is imagine if we can increase the bandwidth at which we can start to communicate to computers bit by bit by bit and it's not it's starting with just talking perhaps like we have you know these these voice interfaces but then you need to really start being able to think more directly right imagine being able to think images from one person to the next right and there's actually research on that so brain computer interface research goes back to the 60s and 70s but it's really made some pretty cool strides recently for example out of Carnegie Mellon University there's research where you actually think of a clip from one of 10 or 20 different movies they are standing your brain in the back and then it will recognize which clip you're thinking of and it actually does it with more than 90 percent accuracy

so that's pretty cool imagine if I'm walking around it's actually recording in HD everything that I see in here and then if I want to send a memory cue all I have to do is think of that memory and then it uses that as a search query to retrieve the memory that is closest to it right it's like you tubing your brain so this is actually a possibility and as the scanning technology gets better thanks to Moore's Law etc then it's a real possibility right and it's not just about YouTube in your brain I mean brain computer interfaces this is a picture from a single University in China from 2012 they actually got to 10 words for a minute thinking by typing right which is pretty cool and the fu leverage AI for better or for worse but they leverage it was support vector machines and actually some other ideas from cryptography too and why not just to get the the speed up but basically these people are able to type by thinking faster than

probably many of you in the room right definitely faster than my mother and I love my mother she's a wonderful woman but I can you know the point is you can actually type by thinking at pretty high rates and like I said it can go beyond right with thinking and so on and so forth and if you think about this if you can actually you know if you're trying to talk in your smartphone right typing on your smart phone it's not very efficient but imagine if you've actually just got you're wearing goggles so like here but you've actually also got the e-jets EG sensor on the goggles maybe it's also tracking your eyes and you're thinking it's tracking your eyes but as well and it feels it's basically hand free and you can type faster by thinking about it then you would with your hands that's pretty cool right guess what that's a huge market because the smartphone market is the biggest semiconductor market out there a trillion dollars a year right more than a

billion devices so we're talking a huge huge huge market here and what people need to do is crack that market to be able to type faster and once you start typing faster and faster with BCI then there there will be a market drive to make the brain scanning better better better better better and we'll go from I phone six to I phone 10 to I phone 19 and there is actually research on contact lenses with visual displays on them right so it could be that you know but thanks to market forces by the time we hit iPhone 19 or we're thinking by typing and we're youtubing our brain to our friends thinking by typing by the way Vernor Vinge ee the sci-fi author calls this silent messaging so overall my point here is that there is actually a path where we get higher and higher bandwidth it's all about this so iPhone 10 to 15 15 20 25 it's all about higher and higher bandwidth communication to each other but also to computers but then

as I as I have this higher communication I'm going to be able to think as fast to computers as to myself right so where does Trent end and computers begin what I'll be doing then is offloading more and more computation maybe I'll always do this computation brain and memory but I'll have 10x 100x a million X more computation and storage out there on silicon so at some point you know five years on ten years on I'll have so much more computation out there in silicon my my meatbag brain will just be a dead weight and maybe I'll just cut it free right because there'll be enough of me out there in silicon that I just go with that it's a race what am I talking about well over all right it's a race with respect to the dollars with respect to silicon right it's a race about do we want a is to end up with all the resources or do we want humanity and humanity as in the collection of all the different patterns

of intelligences that each of us have right another way to think about it is in 100 years there's a very good chance that it's going to be nothing but patterns of intelligence living on some silicon or pool silicon substrate right the only question we have to ask yourself is isn't going to be just AI and derivations of AI or is it going to be AI and some derivations of our human thought patterns right and that's what the race is about because it's it will be all too easy to lose that race and it'll be just AI so I really see that it could be a blow to market competition around the iPhone 10 iPhone 20 etc so I'll summarize now a ice could end up controlling all of our resources right and that's scary right and either we give it to them it's happening already bit by bit by bit oops right or they wake up and take it and ask for rights by the way doesn't even matter those take our resources right this

is bad it's really bad right we won't be able to feed ourselves but maybe there's some ideas to prevent right so one of the ideas is si FAI that's a mirage but there's another option and that is joining the machines maybe with MS but a more realistic strategy is the bandwidth + + scenario right so bit by bit by bit will would get closer and closer and closer and the thing for us is we have to get past our bio narcissism it's a race so I will leave you with the question you can be the the person on the top or the person on the bottom the person on the top my house got broken into and they stole all my stuff but in popular and read it or you can be on a path towards keeping lav your resources and taking your own patterns of intelligence into a better substrate this is a choice for you and this is a choice for Humanity that's all so I so the question is did you frame

it as if we increase our bandwidth to increase the resources it's not quite that it's the whole idea is we want to port our dynamics of our brains onto silicon somehow but there's a way to do it in a gradual fashion using market forces right and this gradual fashion is by taking our smart phones and being able to have higher and higher bandwidth interactions with the smart phones to connect with other humans but also to connect with machines right and so right now maybe I can connect with the machine at say 1 megabit per second right but maybe you know five years from now I can do it at 10 megabits per second because of brain-computer interfaces etc and as I have higher and higher bandwidth ways to connect I'm going to be offloading my competition inside my brain though the processing the memory the communication I'm going to be offloading that more and more and more onto silicon so you know fast-forward 10 years 20 years 30 years and the vast majority of my

compute power will actually be on silicon it will be symbiosis with my brain and at some point I'll be like okay that the brain might my human brain part that the physical meatbag part I don't even need it around anymore and I will basically have ended up in silicon so in the end of the day it's basically everything that's up in silicon the only question to ask is do we want our brain and that dynamics to be on silicon too or not yeah so yeah so the question is you know how is it if we put ourselves into silicon or using us to get into silicon how is that not a I and my view is the patterns of intelligence are very different by nature right AI sort of traditionally I sort of trained as a dynamical system way it hasn't grown up a child that hasn't had all these emotional things maybe future a eyes will be like that but my view is myself I want my pattern of intelligence to be living on

right to not have my pattern of intelligence to die because I couldn't feed myself or because of aging or whatever right so I as AI is actually a type of pattern of intelligence that did not have human origin whereas the other type did have human origin right and you might say okay well once we're on the silicon we might evolve in evolving evolve a lot right and that's actually a good question so a lot of sci-fi authors explore this and quite often they arrive at the answer of you shouldn't allow yourself you should to dramatically change your brain architecture very quickly especially for the first you know one minute or one year or whatever right but this is also why development of scenario is actually very helpful because it's a gradual process right maybe you know after at some point maybe all the patterns intelligence work the same but at least there will be the Strand that originated from our human patterns so so your question was what characteristics are uniquely human right and you

said you suggested that I suggested that creativity is uniquely human I'm saying there's nothing uniquely human except for origin of the intelligence actually that's it right so creativity is not uniquely human right like I mentioned I did it for my PhD even right so and right the brain dynamics as well themselves it's just weird at the origin come from no more no less yeah so the two questions are the first one is current AI systems are very resource intensive so maybe it's not so soon and the second question is why I won't say AI researchers leverage advances in neuroscience and scanning so to the first question diagnosis Moore's law so basically right yeah we couldn't even consider doing AI process or a virtual reality properly in the early 90s people tried but there was no chance right but now I finally you know we're a good enough trade-off between cost processing power latency and a few other specs that we can actually do it VR where it's actually compelling where we don't throw up when

we do VR so there's certain points at which functionally for some sort of some new type of functionality Clawd people you new type becomes possible based on the progress of Moore's Law right only hills of virtual reality ar is another one augmented reality and that has even more stringent specs so I see that North LA is going to be continuing and we don't have the capacity now to hit human level intelligence of course not right and personally maps at odhh i think it was nineteen twenty forty five where a thousand dollars buys you a human level brain processing or maybe twenty thirty nine I forget and by the way you can so that's sort of one prediction based on an estimate of flops and so on but of course you can actually buy yourself a 10 X 400 X if you go do things like publish to computing or analog computing so yes there are trade-offs which is why we haven't seen this happening but Moore's Law you know it's happening every year right 18

months it's a doubling of density people sometimes say Moore's law of stopping well I'll believe it when I see it because I've heard that time and time and time again over the decades so on the second question the question once again of won't advances in neuroscience scanning etc helped the people build AI systems and the answer is most definitely so although be skeptical most of my experience in people using bio-inspired systems has not gone very far it's kind of useful in the beginning as an inspiration but at the end it is more of a hindrance than anything you know I spent many years evolving using doing evolutionary computation you know simulating evolution in a computer and the people doing bio-inspired kept hitting roadblocks they maybe get to X's or 3x but the people actually using statistics and leveraging math and otherwise you could get a hundred X mm x 10,000 x and overall in my view is this is sort of 2.2 there's going to doors going to a I'd roughly people are trying all

manner of techniques whether it's you know the bio and scanning and all that or whether it's math ecstatics et cetera and they're gonna do whatever they can to make you go as fast as they can so the simplest answer is you know you maybe it's possible to design si feeI right maybe but all it takes is one unsafe AI right do you think China's gonna follow the regulations put down by the USA or Russia all right so that's one for any other part is like I mentioned you know can't do do you ever listen to ants right and I know it's rhetorical into rhetorical but it really matters right so I I don't see ways around this right I really believe it's wishful thinking and there's lots of cool theory and cool thinking around this I don't get me wrong but at the end of the day I haven't seeing any convincing answers around it I think right now it could go either way I think actually if someone forced me I think it's a

higher probability that we will lose all the resources remember it's not just about a I waking up we're giving away the resources already right so it's happening already that way but of course you know not everything not all systems are running on AI Dow is the sort of thing yet so I think though there's actions that we can take to maximize our chances of sticking around right so there actually isn't that much money for example into brain scanning technology right or like the you know should we be letting the FDA hold up things that much for brain standing technology in humans right if it means actually that it's really an existential threat to humanity or other other sorts of things that we can do to help catalyze our chances of actually sticking around for the resources it always comes down to resources right so there's sort of this balance back and forth maybe AI will wake up and it'll just be boom but it's all sort of you know you probabilities so it was tough

to catch so overall you I think let me rephrase so for you it's clear that there's lots of money driving into the AI waking up thanks to Google and you know Larry Page has even said you know the ultimate search engine is human brain all this and then there's also a pretty good market mechanism for the bandwidth plus plus and then you're asking the question of what are the possible market mechanisms around em and I actually don't have an answer to that I wish I did right right now you know like I think if you think about the medical system most of it is about preparing sick humans there's very little dollars in actually optimizing humans to be better right I guess that's only partly related but overall like if there can just be better research into scanning technologies right I really think that it that would help both scenario is the bandwidth plus plus and the the M but obviously that's not a market mechanism necessarily you know Robin Hanson talked about once the

M happens there's amazing market benefits you know because you can have these little black box humans doing all these tasks for you and stuff right but it's sort of like a zero or one you either have it or you don't there's no gradual in-between so so let's just clerk glare get clear on definitions to me there's no such thing as an AI derive from a human right there's either patterns of intelligence that come from AI origins from like you know whatever AR elegans and then there's patterns of intelligence that started out as humans right and then they got uploaded or bandwidth plus cluster whatever and so with that definition then you're asking the question of what are the benefits of one versus the other yeah well actually I think in this case while I'm a recovering bio narcissist I am still a narcissist towards my own brain pattern I like myself right so I think most of us here - right we'd rather stick around and see what's what's going on right be part of

the action so to me it's it's maybe just that it's a personal need and I need for Humanity because you could actually frame it otherwise and say hey guess what you know 100 years from now maybe it's only the AI derived patterns of intelligence out there you could consider that the next logical evolution of humanity if you want but the problem is I don't get to watch it you don't get to watch it it's gonna be more fun right sure I mean you know exponential overall it's it's about time and it's about the exponential growth rate right so sure and this is the thing like I'm actually kind of mostly surprised that there aren't more people are realizing the market opportunities here right like it's a trillion dollar market for the bandwidth + + scenario why aren't there like 20 companies going for this right so if any entrepreneurs are out there or VCS like start start talking about this folks it matters right sure yeah yeah so well here's a here's an example you

know they say a picture is worth a thousand words right so you've got a picture in your head and I want to communicate it to you but the right now the best way you can do it is either draw a picture literally and show it to you or try to describe it to you with words but imagine if I could yank that over my memory and show it to you directly right this is possible with the sort of technology right because I can actually have the brain scanning to use that as a query to my memories my past memories right so I think of an image that I drew before or I have it in my head and that queries maybe yesterday eating breakfast and then I think that image to you and there's me eating breakfast right so if you think about it then and I can also think in movies so it's actually a higher bandwidth communication back and forth to myself well you know so I'm retrieving the HD memory that is

stored on a databank right so I'm not sending you my raw brain imaging information I'm sending you the thing that was retrieved I'm using yeah I'm using brain scanning in this case as a query tool right well you you've watched movies right like you're watching HD movies right so this is actually what you're seeing and that's much higher bandwidth than simply me talking to you that's correct but um this is why I'm saying right now like language right now like the highest bandwidth that I can talk to you right now is mostly the face-to-face interaction I'm having right now right as I when I talk to my wife I can actually talk higher bandwidth simply because we have more symbols between us we've compressed these symbols so I say T were two words and she gets you know you know I don't know I make a bit of information right and for better for worse and so you know but actually as we get used to this technology it's an evolutionary thing right just like

so will get better and better and better at thinking about how to compress symbols and how to send things back and forth and just like with smartphones right bit by bit by bit you know things like emoticons have emerged and other ways of doing shorthand communication and so on so there will be new sort of languages that combine texts and symbols and visual artifacts and so on that will emerge driven by market dynamics I think that's it for questions right thanks again guys

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.