Neuralink AI x Neuromorphic Computing - Tomasz Stanczak

and I'll be talking about neuromorphic computing and neural linked economy and the part where you can expect a bit more insights from my side is the second part the first one is a bit more like exploring together what's there in the market and what's coming to us in the technology but it's not where I'm practitioner so it's not a mind we work a lot with the blockchain systems and built also the exploration of the future of blockchain in a combination with AI. And we're exploring how, in the future, potentially the investment game or economy game or collaboration game can look, if you think about the connection of the brain with the network of brains or the network of agents. So in a naive way, thinking about a full brain emulation, I can think of 80 billion neurons in a full brain emulation. I can think of like 80 billion neurons in a human brain and some of the projects in the world getting grants and solving the issues of trying to simulate all of this. And starting from the very low level in thinking in the traditional computing way of looking at the cellular level.

And one of these projects is the Blue Brain project with lots of criticism but also delivering hundreds of various packages that attract a lot of researchers to build together solutions for thinking about how in the future we can simulate the brain. And the project started years ago, before the explosion of the current success of the artificial intelligence, NLM. And, well, the criticism you think here, the brains in silicon is another research center, a company that builds the hardware solutions for simulation of the neurons. And here they're joking practically about the blue brain project using like 10 million watts while they can use 10 watts for simple simulations. But when we think about the direction years forward, there's this one chart that says, oh, we will take time until 2,110 until we can simulate the brain on this cellular level, like do the full simulation.

And you see the charts with this, like, every 14 months doubling of the computing power and all the Moore's laws and so on. But if we listen recently to the thought leaders and those who work on AI, they will say, look, at the moment we are doubling every three months, three to four months, our processing and the strength of the AI. And also, we hear it on the hardware side, on the software side, the progress is amazing, so maybe it's somewhere much closer than 2110. And when I think about neuromorphic computing, so there are three main concepts. So event-driven processing, so you no longer look at the entirety of data and to process the data traditional sense, but you actually look only at the inputs, at the events that invoke processing in parallel from multiple inputs at the same time.

And we're trying for the neuromorphic computing, we're trying to mimic the behavior of the brain and neurons. So it's different hardware and computation paradigm to say that practically you have all of those neurons connected together or like the hardware devices that represent neurons. And the way they behave, they also introduce the temporal aspect. So they not compute all the time, but they wait for the input to reach some threshold, and then they fire the processing. So that's the event that involves processing.

And you can think of it either in the sense of incoming data, but it can be from the sensors. So it can be visual data as a single pixel on the camera, firing and initiating the processing and reprocessing of just part of the image if the input signal changes, like if there is light falling on the pixel. And this is done for the spiking neural networks. So that's bybased communication between neurons is crucial and the synaptic plasticity that actually allows the system to change its shape or behavior, or strengthen or weaken some connections between hardware parts. And this comes also with lots of tooling that supports the solutions.

You've seen the exploration of the Blaine emulation with the Blue Brain project. Nango is a Python package and set of libraries to actually work with the devices that are working this neuromorphic way. And the two chips that are not commercially available, but are available for the research purposes from IBM and Intel is IBM's TrueNorth and Intel's LoiHe2. And the LoiHe2, there is the HalaPoint, which is the system that is combined of 1,000 of those chips, LoiHe2. And what they say, they look, there's 2 billion neurons representation, which is a bit like an owl brain.

So 80 billion with people, 2 billion with an owl, and if you want to get all the way to elephants, it would be around 260 billion. So after people, after humans, we'll be able to also simulate elephants. And on the right, what you see here, is the actual vision device, so dynamic vision device. It detects... So on the right of this picture, is the actual vision device, so dynamic vision device.

It detects, so on the right of this picture, you see actually how the system sees the changes to what it's recording. So on the left, you have the full image as we're recording nowadays. Like frame after frame, we record the entire visual. And to the right, you see that only the change, like here, the person speaking, as actually CEO of Prophecy, the company that produces those cameras, is moving hands, gesticulating, moving head. And what it allows is, first of all, much more frames per second, like I think something like 100,000 frames per second.

So highly dynamic, like, sensors for industrial use cases. And much better compression, practically, from the, I mean, like, much less data being saved, because you practically totally ignore what's not changing in the scene before you. And also we have the processing, the chipsets from Inatera and Synsense. So these are the companies that are providing these neuromorphic chips. So if you explore those together with the packages, it is like one way to look at new way of approaching the AI and signal processing.

And to my knowledge, it's nowhere near configured for using on LLMs, so this would require much more research, and hard to tell whether it's possible to match the current solutions. But yes, the promise is that actually if you did it, then you can save energy like 100 times, use 100 times less energy for the processing. And since we hear now that's like one of the major challenges for the AI in global use cases is how much energy it will use. Like we hear about all the nuclear power being used by the major companies and also leads to massive centralization of the access to the AI processing. It seems like one of the promising futures.

Now I go a bit further away from this. Since we're talking about the brain and we've been talking about the BCI, the interfaces for connecting your brain, the invasive systems are the ones like Neuralink or Synchron. So about Neuralink, it's the company from Elon Musk that is implanting a device to your brain, connecting the representations. And actually, at the bottom, you see the person with the implant actually is playing a computer game, Civilization, and making decisions on how to progress within the game. And now, a bit of exploration of, are we sure?

Are we sure that we want this? And what it would mean if we start connecting people to the game that normally is in the digital world to the network. And I'll go to the space of finance and decisions on the major game for many people like you can choose the game for relaxation or you can choose the game for many people. You can choose the game for relaxation or you can choose the game for ambition. And this pure game that many people play would be investing.

And then also when you remove the game aspect of investing that actually what you care about is the life aspect of investing. Like why are you investing or why do you manage your money? It's because you're planning your life and planning your wealth. And at the moment, we'll say, well, robo-advisors are nowhere near beating humans in investing, especially, like, the top traders will still be absolutely beating in on the market, but then they will use a lot of tooling to support. And then we see the rise of agentic markets and agents and there's lots of cloud for the agent systems.

So using AI to deploy agents that potentially deploy those agents on blockchain networks to trade for you independently. And within blockchain, we talk a lot about the user interfaces, and practically in traditional finance, the finance as we know, we also care about interfaces, like how we talk to our accounts, to our bank account, how we take decisions on investing, and so on. So this is cumbersome. It takes time. We have to understand complexity of the markets, but sometimes we have the basic the basic intent to to make money or to save money for particular for particular purpose and I think that's Really the if you simplify it the only thing that matters in the end for the regular person or for any person is You wake up in the morning.

Can you wonder how much money I can spend today, how much money I can spend this month, can I travel, when can I have wedding, university, parents can retire and so on. And I would say, well, the moment when you have the BCI, the interface connected to the device that can read your mind, it means that you no longer have to play the game. You no longer have to take decisions. The AI starts reading your intents or even how tired you are, and it starts thinking, no, you don't have to decide for orders execution. I don't have to solve for your intents.

I can read the intents before you even have them, like in a bit of way of minority report. And then the only thing that matters in the end for us to get more wealth, if agent is doing everything for us and knows everything that we need, is to get healthier and to learn more efficiently. So in a way, I feel like that we are going to the brighter future, where the only way to make more money will be to be healthier, to live longer, to be more productive, but not really to try to be better at the markets or to play the game. So we'll be playing games only for pleasure, for the pleasure to feel healthier and to make more money because the agent will know that I can actually do more meaningful work. And the other aspect, maybe like more dark one, is that we think that nowadays we create those digital systems and agents and we're happy that we can hold them as NFTs or like simply represent them on digital.

At the same time, it's really trying to uniquely identify humans, to protect humans, like in the world of AI, so we can say, oh, I'm unique. But when that happens, actually what happens is that finally agents can hold people as NFTs, and they can start to trade people. So I think there's lots to explore. And maybe in the future, ideally, when I'm connected to the network, I'll be able to just say I focus on my life and on my health and somewhere there at the back of my brain subconsciously I execute the work for the network and someone pays me for that work and someone decides of where that money that I get from my work is being allocated, knowing when I want to go to travel. And this will be just a tool, the same way as nowadays, our mobiles and our internet knows much more about us and gives us hints how to leave.

All right, thank you so much.