Alexandre Marcireau Interview Transcript

 

DR. BRUCE MCCABE: Welcome to another episode of FutureBites with Dr. Bruce McCabe, your global futurist, where we explore pathways to a better future, especially technological pathways. And I'm here with Dr. Alexandre Marcireau at the International Center for Neuromorphic Systems. And I'm really excited to talk to you today, Alex, because to my mind, neuromorphic systems, these computer systems that are more analogous to the way neurons work, the way brains work. And I'll get you to explain a little bit more in a minute, but they seem to me to be one of the most, perhaps even the most important field of computer science when we look at the future, because of where we're going. And I want to get into that with you today, but just a quick intro to you. I've got you here and we've been chatting for a little while here at the lab, and your background was both at the École Centrale in Paris.

DR. ALEXANDRE MARCIREAU: Absolutely.

BRUCE MCCABE: And, at the Université de la Sorbonne. Do I pronounce that one...

ALEXANDRE MARCIREAU: It does...

BRUCE MCCABE: My French is terrible.

ALEXANDRE MARCIREAU: Well, that's perfect, actually.

[laughter]

BRUCE MCCABE: And now you're here at the International Center for Neuromorphic Systems at University of Western Sydney. So I've discovered a gem here in this big lab. You've got something like 40 researchers.

ALEXANDRE MARCIREAU: Indeed, yeah. Including PhD students. So it is quite a big team, and it's grown tremendously in the past few years. So I joined almost four years ago now. And before that, the lab was tiny and it's grown very recently.

BRUCE MCCABE: I have a feeling it's going to be growing exponentially over the next 10 years.

ALEXANDRE MARCIREAU: We really hope so. And that's really the goal, to build a very strong center. So the explicit goal is to become the largest research group on neuromorphic engineering in the world. And we're definitely in the top two or three, for the moment.

BRUCE MCCABE: Really. See, this is one of the ironies. I think very few people listening to this will be familiar with the term neuromorphic computing. They'll probably hear a lot of quantum computing. They know the world's excited about that. I'm excited about it, but I want everybody excited about it. I think it's so important. Can you give us a bit of a quick summary of what it is and why it really matters?

ALEXANDRE MARCIREAU: Absolutely. So neuromorphic really means "in the morphology of neurons". So the field started some 30 years ago with the intent of building computers that are still made of silicon, like regular computers, but their structure, instead of being, like a conventional computer, you've got your RAM and your CPU, would instead be more like neurons, tiny units down to the transistor level that would be exchanging something that looks like a spike, which is what the brain uses. And the key idea here is that we've been doing computation all wrong.

[laughter]

ALEXANDRE MARCIREAU: Our computers are fantastic machines when it comes to crunching numbers. So, if one is interested in doing maths, computers are the best option, no questions there. But if you're trying to move in an environment, you may not actually care all that much about the absolute value of temperature or the position, the specific position of certain objects. However, latency is everything. If you spend an hour computing something that you're seeing, but you're too slow to really take action, your computation is pointless. However, if you can do far less with far less power, you're more likely to survive. So that's really where we're going with this whole idea.

BRUCE MCCABE: Yeah. So when I look at some of the applications you've shown me, which I want to run through, there's use-cases in the real world. So much is about moving away from this, the traditional, what we'd call a von Neumann computer architecture, where we're actually wasting an enormous number of cycles doing nothing or processing information we don't need necessarily need, moving to a much slimmed down way of doing things where we're just processing the information we need and we're doing that in a quiet efficient way and as we need it. Is that kind of a nice little summary of...?

ALEXANDRE MARCIREAU: It really is. I think conventional systems really like clocks. Everything is about doing things at...

BRUCE MCCABE: Faster execution...

ALEXANDRE MARCIREAU: Set intervals, right?

BRUCE MCCABE: Faster clocks. Yeah.

ALEXANDRE MARCIREAU: Yeah. And cameras, video cameras are a good example of that. A video camera is just a fast photo camera. Take a picture 50 times per second, and you get the illusion of motion. Problem is that that's very wasteful, especially if not much is changing in the scene. The next picture is very similar to the previous one, but if you give that to a computer, it's still starting from scratch, has no idea that nothing changed, so it still has to redo all the processing, all the number crunching to extract different things in the scene. And quite often the result is a tiny thing. After all this processing, we conclude that the robot should slow down or speed up. So neuromorphic engineering is really about moving a lot of this heavy lifting as close as we can to the sensor, so that we do far less but ultimately make the same decision. So the system still works, it's just that it's not wasting, every step of the way, a lot of resources.

BRUCE MCCABE: Yeah. And a lot of your work's at the sensor end of things, so cameras, audio sensors, potentially all the sensors that we have as humans are candidates for better ways of doing the computing for robotics, for household sensors, for satellites, I believe you're doing stuff in satellites as well.

ALEXANDRE MARCIREAU: Yes. So specifically in the context of space, vision is really the only sense that really works, because yeah, olfaction or audition aren't all that useful in space. However, for more grounded applications, much like one can build a vision sensor inspired by the human retina, it's possible to build a sensor inspired by the human cochlear, so part of the auditory cortex, or to build a neuromorphic nose. So we actually have students that are trying to build one of the first neuromorphic noses, to try...

BRUCE MCCABE: Neuromorphic nose?

ALEXANDRE MARCIREAU: Yes. To try and...

BRUCE MCCABE: So now we have a new way of sensing smell.

ALEXANDRE MARCIREAU: Exactly, yes.

BRUCE MCCABE: All those particles in the air.

ALEXANDRE MARCIREAU: And, yeah. It may not seem like much, but really olfaction covers anything chemical, so detecting fire...

BRUCE MCCABE: It's huge. Industrial leakages.

ALEXANDRE MARCIREAU: Absolutely.

BRUCE MCCABE: I've come across from a medical diagnostics conference earlier this year where there's olfactory sensing for all kinds of diseases, neurodegenerative, cancers, there's a whole new field. How do we do better olfactory sensing, yeah.

ALEXANDRE MARCIREAU: Yes. Well, that's really the question that they're trying to address, and they're trying to address that by saying, well, somehow dogs, have figured out ways to be pretty good at smelling. Let's look at their neurons. Let's look at how they do this processing and just cram that into an electronic chip and see what we get from there.

So more generally, that's the idea that we apply to pretty much every other sense. Take what we understand of the human brain or quite often a much simpler animal, and build a working system out of that. And usually, we run into all sorts of trade-offs. We need to separate ourselves from being an exact copy of what biology does, which is why neuromorphic engineering, or the sense that we attach to the term, has perhaps changed over the years. It used to be very much, ‘let's do a one-to-one mapping of neurons to transistors,’ and it slowly moved towards more of an inspiration. Let's try and understand the fundamental principles in biology when it comes to sensing and build a computer that uses the same principles, not necessarily an exact copy.

BRUCE MCCABE: Yeah, not necessarily copying the neurons in a bee or whatever, but using them as an inspiration for how you might do sensing.

ALEXANDRE MARCIREAU: Exactly.

BRUCE MCCABE: So this is one of the things that excites me so much, that there's two frontiers that we're forging ahead in the world that I'm constantly researching myself and talking to people about. One is in neuroscience, just understanding brains and in medicine, in all of those applications. And then one is here in the hardware side we're pushing ahead in how we do better neural networks, artificial intelligence, different models of AI, different components of AI. And it seems that one is informing the other. And in this lab, you've got people with biology backgrounds, people with silicon-based backgrounds, if you like.

ALEXANDRE MARCIREAU: We do.

BRUCE MCCABE: And all over the world, these fields are learning from each other and pushing each other.

ALEXANDRE MARCIREAU: So the original idea was to say that both fields would benefit from one another. That you'd take from biology, understand it, use that to build a better computer. And by using a better system more generally and using that system in real conditions in the real world, informs your biological modeling. It tells you what the system does well and what it fails at, and perhaps tells you what you need to add or what's missing in our understanding of biology. In practice, that second arrow has yet to really crystallize into something useful, I would say. So it's been mostly us engineers stealing from the biologists. We're still building up that second arrow. It turns out that it's quite challenging.

BRUCE MCCABE: It's a very challenging field all up, I think, but it's so rewarding. A couple of use-cases that I've come across which might help people picture why this is so important. I was chatting to a guy in the Netherlands, a guy called Prof. Guido de Croon, and he's a scientist who builds micro air vehicles, so little robot insects. And what's cool about them is he's using them to inspect crops in glasshouses, in these greenhouses that they have for so much of their horticulture and stuff in the Netherlands. And they've little dragonflies flying up and down. And one of the surprises in this conversation with him was he said, "Oh yeah, I'm sending all my PhDs away to learn more about neuromorphic computing," because all the onboard vision systems for a little dragonfly, if he could make it neuromorphic, and I think they're starting to use that technology, they use so much less power, less electricity, which means they can fly longer and they can fly more missions and do better work! That blew me away, and it seems to be a constant theme, this power thing.

BRUCE MCCABE: And the other one was in medicine. This idea of a sensible skin, sensible as in sensory, for prosthetic limbs, where you could have a surface on a prosthetic hand, for example, and when you touch something, it would now feel if something was hot and perhaps send signals back to the brain if something was sharp or hot or rough or smooth. And that had neuromorphic hardware behind it as well. And so these two, just two tiny applications, but they're actually huge, aren't they, especially the power side of things?

ALEXANDRE MARCIREAU: Yes, absolutely. So the key here, I think, is that by making the sensor smarter in a lot of ways, we can remove a lot of the computation from what would usually be done on a GPU and move that to the sensor. And that makes all the difference, really because GPUs are monsters when it comes to power consumption.

BRUCE MCCABE: Yes. GPUs are those graphical processing chips which... All the graphics cards we now use for AI. [laughter] Huge quantities of heat and energy are just being...

ALEXANDRE MARCIREAU: Power may not seem like much, as in what's the power of a computer really compared to a car or a plane? But at the current rate, in a decade, half of our power production, half of the total power production, will be dedicated to data collection, processing, and storage. And that's problematic. That is not something we can afford.

BRUCE MCCABE: It's mind-blowing. I've not heard that data point before. It's really, really mind-blowing. So half of this planet's electricity use will be used for information processing and storage ...

ALEXANDRE MARCIREAU: Yes. Absolutely.

BRUCE MCCABE: … and collection, all that. So half. Basically, especially with the exponentials going on in all forms of computing but AI especially right now, we're on a collision course with all of the goals we have, to move to renewables, to reduce overall energy consumption, to become more sustainable. This is really fundamentally important to our future.

ALEXANDRE MARCIREAU: It really is. And we really want to keep those techniques. Those new AI tools are incredibly exciting.

BRUCE MCCABE: We don't want to give them up. [laughter]

ALEXANDRE MARCIREAU: Yeah, absolutely. But they really cost too much in terms of not just money, but also power, very fundamentally. And part of the reason why they cost so much is because our sensors are dumb. We just capture everything, store everything, and figure it out later. And as a result, that is still a lot of data that one needs to process, usually to extract a tiny thing out of the data. For instance, if I'm, say, trying to track insects, ultimately what I really care about is the pixel position of each insect. It's like two numbers. But in practice, I'm probably going to capture frames with, say, a full HD camera that's millions of pixels. So I'm really storing millions of numbers 50 times per second just to extract two out of those.

BRUCE MCCABE: Even though you don't need them.

ALEXANDRE MARCIREAU: Exactly. And yeah, ultimately that's wasteful, and we can do better.

BRUCE MCCABE: And we think the human brain works very much that way. It takes what it needs and doesn't over-process, if you like, right?

ALEXANDRE MARCIREAU: Yeah, or at least what we understand of it seems to indicate that the human brain is very efficient in this regard. And more generally, anything in biology. So for instance, a human eye, or any eye for that matter, does not capture frames. We haven't found any species that takes photos of the entire scene at set intervals. They all have change detectors. Their eyes are only sensitive to motion. T-Rex vision, if you will, as in Jurassic Park, where the T-Rex cannot see you if you're not moving, right?

BRUCE MCCABE: [laughter] Oh, the T-Rex.

ALEXANDRE MARCIREAU: We're actually all like that.

BRUCE MCCABE: Yes. [laughter] That's a great example!

ALEXANDRE MARCIREAU: And we compensate by moving our eyes all the time if we want to see something that is not moving. So it doesn't prevent us from acquiring information when we want to, but we're not forced to absorb that information. That's really what we're forcing onto our computers. We're force-feeding them all the data all the time. And yeah, that's really where this change in mindset, in how you build your system, has dramatic consequences in terms of its overall architecture. And it's actually quite challenging. We need to reinvent what a computer is like if you change that very way of acquiring the data in the first place.

BRUCE MCCABE: And if we... Getting back to that power thing, I think you used the figure, when we were chatting offline, that the brain uses how much energy, roughly?

ALEXANDRE MARCIREAU: So the brain is roughly using 20 watts...

BRUCE MCCABE: Twenty watts.

ALEXANDRE MARCIREAU: Of power based on blood flow. That's how we estimate it.

BRUCE MCCABE: So roughly if we were to... Our mission, the AI mission is to try and replicate the complexity, sophistication and power of the brain and exceed it. But if we were to actually be truly analogous in that and create circuitry to do that on traditional computing architecture, someone told me once that it'd be like using all the electricity for the city of Singapore to duplicate one brain. I mean, if we don't transition to neuromorphic architecture, we'll never do that, in a sense. We can only solve small slices of it.

ALEXANDRE MARCIREAU: Absolutely, for something like ChatGPT, which arguably is pretty good at generating text, definitely on par with most humans, if not better than us in a lot of cases, was trained on... I think it cost millions of dollars. And those millions of dollars aren't because they were using a unique computer that was really hard to access. It's just because they were using hundreds of GPUs.

BRUCE MCCABE: Oh, thousands probably. They just buy them. You can't even get GPUs now. There's a shortage because of all the AI people are buying them up.

ALEXANDRE MARCIREAU: Exactly. And indirectly also buying all the electricity that you feed into those models. So it is really expensive and clearly impractical if you want to do, if you want to create a lot of those systems.

BRUCE MCCABE: Which we're going to do. It's going to be exponentials on exponentials.

ALEXANDRE MARCIREAU: Absolutely. And there are cases where moving those computers away from... So you could have like one massive computer in a big data center and then just communicate with that. That's sort of the Google way, if you will. Our smartphones are quite often smart, not so much because they have a lot in there, but because they can communicate with that server. But there are a lot of cases where it's not really a good idea or something we can even do for any sort of medical data for instance. I'm actually quite happy that Google doesn't have my entire medical history.

BRUCE MCCABE: You want to do it more privately and locally.

ALEXANDRE MARCIREAU: Exactly. And it means that the AI that could be analyzing my data to tell me, "Oh, maybe you should go see a doctor because there might be something wrong with you," that AI may need to run on my smartphone if I want to make sure that that data isn't shared by someone I don't know. And that...

BRUCE MCCABE: So this is like a pathway for the local device, your phone, or your camera to do its own learning...

ALEXANDRE MARCIREAU: Exactly.

BRUCE MCCABE: … on the platform without that huge data center being required at the back end.

ALEXANDRE MARCIREAU: I really think so. I don't think that we're there yet, but that's where we want to go.

BRUCE MCCABE: No, but it's a pathway, it's where we want to go, yeah.

ALEXANDRE MARCIREAU: There is another family of systems that cannot rely on the big data center and that's any sort of device that is isolated. So if you have, say, for instance, an underwater exploration system or if you have a spacecraft.

BRUCE MCCABE: Or forestry sensors. You were talking about forestry sensors as well.

ALEXANDRE MARCIREAU: Yeah, or if you're in the middle of the bush in New South Wales and you want to monitor, say, koalas, for instance, using energy to communicate with your data center to do analysis would force you to wire your device. You wouldn't be able to live off of a small battery.

BRUCE MCCABE: Yeah, you need some car batteries or whatever. [chuckle]

ALEXANDRE MARCIREAU: Exactly. So that's very impractical. Or even if you're, say, sending a mission on Mars and you want that system to be somewhat autonomous, it takes, depending on the position of Mars and the Earth and so on, minutes, typically 20, for information to go to Mars and back. And it's not because we can do better. It's just the limit of the universe. That's how fast the light travels.

BRUCE MCCABE: That's it. Speed of light is all we've got. Yeah. [laughter]

ALEXANDRE MARCIREAU: So, the only way to build the systems is to give them intelligence. To have the computations done right there. What we're doing right now is just moving them really slowly, so really we can check on them. It means that we've covered a fraction of the surface of Mars, because our rovers are slow and they're slow because if we make them any faster, we risk breaking them, because they have no clue where they are. They have cameras, but we don't trust them to make the right decision.

BRUCE MCCABE: The future... Let's try and get it in a sentence. So, ‘the future of localized intelligence has to be neuromorphic,’ to do this really well.

ALEXANDRE MARCIREAU: Absolutely. Even a fly, though it's quite simple, is probably better at navigating than the million dollars, billion dollars robot that we put on other planets. So really that's sort of the dream. Can you build something that is as good as navigating a new environment as a cat? Cats are pretty good. They like to explore, they like to jump on things, they're quite clever in this regard. Can we put a robotic cat on Mars? Well, not yet.

BRUCE MCCABE: No. And as soon as you think of that problem, can we make something as good as a cat or even a fly, and do it in a low ... a tightly packaged way with low power consumption, suddenly you've got an explosion of applications. It really changes everything. Because so much computing is based on some set of sensors somewhere, or some remote component. 

Now when we get into AI, I wanted to test my thinking with you because, so many researchers like Yann LeCun and others that I've met, their vision of the future is ... they're tackling AI in different ways. They're doing, lots of training through data sets, so ‘curated learning’ if you like. But then they talk about uncurated learning, sensors that can learn as they sense, which to me is neuromorphic, and things that are more, if we think of a future electronic brain, we might think of it in terms of cortices and we might have parts of it which are much more instinctive. Again, I would say neuromorphic circuitry would be more appropriate for learning straight from the sensors and so forth. Would that be how you picture that future of AI? We're chopping pieces of it off, and some of it would be more traditional computing and some of it might be around the...

ALEXANDRE MARCIREAU: Yes. I think so. I think there's a lot of value in those specialized pieces of hardware. I'm sure that they would disagree but, in a sense, the most recent iPhone is sort of neuromorphic in that instead of having one central CPU and one central storage system, it has all sorts of dedicated chips. It's got a GPU, but it also has machine learning, a specialized component just for machine learning, just for sub types of algorithms. So we are already specializing hardware to do very specific tasks. And the neuromorphic engineering is very much going in this direction as well to say, if you understand what you're trying to do, you can build a system that is always going to outperform something more generic, because you can tailor it to do something.

BRUCE MCCABE: It's really like the beginning of computing all over again. It's like where traditional computing was in 1950. [laughter]

ALEXANDRE MARCIREAU: I really think so.

BRUCE MCCABE: There's so much headspace, opportunity space.

ALEXANDRE MARCIREAU: I actually blame the inventors of conventional computing. They were too brilliant.

BRUCE MCCABE: Too good. [laughter]

ALEXANDRE MARCIREAU: They were so good that the architectures, the ideas they came up with, in particular von Neumann and Alan Turing, it was just... Not just them but they were perhaps the most visible. Their ideas were so good that they worked for decades. Sure, we've iterated on the system but fundamentally we haven't changed what a computer is. And neuromorphic engineering is about that. What if you put those transistors in a different configuration and compute it in a completely different way? What would you then get? And it is hard, but also incredibly interesting.

BRUCE MCCABE: So there's a lot of work to be done on the software side, the algorithms. I understand there's a huge spectrum of activities out there as we learn how to program differently, let's say. A lot of this is analog, which is fascinating. So it's about moving from digital to analog computing, because that's how biology works. And I think you described the dream to me as somewhere along that spectrum because we've got components at the sensory level which we can do. We are moving towards neuromorphic hardware to replace GPUs one day, but the connectivity between the two is hard. How do we do that? So you were looking at more end-to-end, weren't you? Ultimately that's the long-term future dream.

ALEXANDRE MARCIREAU: Absolutely, so what really stands out I think for the brain is that it is a massively parallel 3D structure with those incredible connections. Part of the computation is in the very structure of the system. Our computers by contrast, are far more boring, far more organized in like pipelines. And I think that ultimately, we'd like to build systems that are more like the brain in this regard. That are massively parallel, like what if you connected every single pixel of your camera directly with its own little unit, which can do something and maybe communicate with the units that are nearby but that's it. And it's very local in this regard. And that seems to be how biology is organized, because it means that it can be locally much slower. If every pixel generates, say, one spike per second, that's pretty slow compared to the gigahertz of a computer, but you're still performing a million operations per second. You just distribute them spatially.

ALEXANDRE MARCIREAU: In this regard perhaps not unlike a GPU, except that the GPU then puts everything back in memory and you've got this one point of failure that uses a lot of energy. So really can we build an end-to-end, fully parallel system? And today the answer sadly is no. It's really hard to get the entirety of that system right so, what we've been doing is multiplexing to separate the problems as much as we can. So have a sensor that really pushes everything onto one wire, typically a USB cable, and then a computer or perhaps a set of neuromorphic units that unpacks data from that one wire and distributes it again. So that's a workaround to study our systems. You mentioned that we're really doing two things at the same time, algorithms and hardware. And we're really ultimately trying to build hardware that represents those algorithms. The algorithms are encoded in the hardware in a lot of ways. And it's really hard to do two things at the same time. Inventing those new algorithms that are different from what a conventional computer does and at the same time inventing new hardware is insanely complex.

BRUCE MCCABE: That's amazing. But what an amazing problem space to be working in because it's so transformative. And you've only got 40, you need 400 PhDs here. There are so many things to work on, and they all matter. [laughter]

ALEXANDRE MARCIREAU: I absolutely agree. I don't have a single colleague who's like, "Yeah, I'm kind of bored I don't know what to do next."

BRUCE MCCABE: No one's bored here.

ALEXANDRE MARCIREAU: We all have way too many projects, because it's so interesting and because, so...

BRUCE MCCABE: If there are researchers out there that are interested in this space, because a lot of people I intersect with, who are in the science community and everything, you're looking, aren't you? You're looking for more researchers now.

ALEXANDRE MARCIREAU: Absolutely. So we have 10 PhD positions open at the moment, all have to do with neuromorphic engineering. Some of them very fundamental, looking at the mathematics of neural networks and how that applies to neuromorphic hardware. Some of them are far more applied in particular to space, because that's one of the pillars of our lab at the moment. So we're really playing with a lot of ideas in that bubble. We have data from the International Space Station where we have sensors, so a lot of very exciting...

BRUCE MCCABE: Wow. You've got a camera on the ISS now.

ALEXANDRE MARCIREAU: Yes, we do. Two of them, and we're sending two more in a few months, if everything goes well.

BRUCE MCCABE: What are they monitoring? They're monitoring lightning and...

ALEXANDRE MARCIREAU: They are. So the primary mission is to monitor so-called trans luminous events which are sometimes called sprites, and it's essentially lightning that goes up. So it leaves the clouds, but instead of discharging towards the Earth, it discharges towards the upper atmosphere, and it is actually quite hard to see because, well, usually when there's a storm, clouds are in the way. So the best way to look at them is from space, which you cannot really do unless you have a really high-speed camera. Except that they are...

BRUCE MCCABE: It's all about the speed of the camera.

ALEXANDRE MARCIREAU: It's all about the speed.

BRUCE MCCABE: Because it's neuromorphic, you can run it so much faster when it's event-based.

ALEXANDRE MARCIREAU: Exactly. Nobody is crazy enough to send something like a Phantom [high-speed] camera in orbit, because it uses far too much power. And even if you could somehow send it up there, and it's very heavy, very power-hungry, you would have terabytes of data to send back to earth, and even on the ISS, we have pretty good internet connection, it would take weeks to just get one of those recordings. And that's where having a sensor that naturally compresses all that information in the pixels just in the way it works, makes all the difference.

BRUCE MCCABE: That reminds me of another way of saving the planet. Didn't you tell me you had a little data point on the amount of data we're producing on this planet. I think in the last one and a half years, what did you say? [laughter]

ALEXANDRE MARCIREAU: Oh yeah. I know that... That's a number that just baffles me. Every, one and a half or two years, I can't remember the exact number, I'm going to double check...

BRUCE MCCABE: That doesn't matter, but it gives us the idea, yeah.

ALEXANDRE MARCIREAU: We double the amount of data that humanity generates. So to put that into perspective...

BRUCE MCCABE: Yeah. So in every one and a half years, we double all of the data humanity has ever generated. We double it. [laughter]

ALEXANDRE MARCIREAU: Yes. So think of all the data from the first early traces of writing in the clay tablets, all the way to 2023.

BRUCE MCCABE: That's insane.

ALEXANDRE MARCIREAU: By 2025, we'll have generated that again.

BRUCE MCCABE: Again, yeah. Everything. And well, the whole thing about neuromorphic hardware, is it doesn't collect just everything that passes the lens or the microphone. It collects only the events that are of interest.

ALEXANDRE MARCIREAU: Exactly.

BRUCE MCCABE: Just like we do. So you're solving that problem too, that information overload. [laughter]

ALEXANDRE MARCIREAU: At a very fundamental level. And that directly connects to what we were saying about specialized hardware. It's all about detecting just what you need, which typically means that you need to know what you're looking for. And you need to build a system that captures that and just that. The less data we capture, the better.

BRUCE MCCABE: Yeah. Look, that's wonderful. What's the website if people want to come and learn more about the center, particularly researchers want to come and...

ALEXANDRE MARCIREAU: If you search for Western Sydney/ICNS.

BRUCE MCCABE: And ICNS being for International Center for Neuromorphic Systems.

ALEXANDRE MARCIREAU: Mm-hmm. You will find some of the projects we are working on, as well as the open PhD positions. And I should say, if someone is already doing a PhD in something completely different or a researcher working on something else, and they think that what we do could be interesting to their research, we're also always looking for new collaborations, new people with whom to exchange ideas and techniques. So, yeah, our doors are wide open.

BRUCE MCCABE: Oh, wonderful. Well, thank you for sharing with me today in the lab and letting me spend so much time in the lab with you today, Alex, because it's been... I keep coming across neuromorphic principles, little pieces of the puzzle that are now out there in the world. And I really mean it when I think it's the most important field in computing hardware, bar none. I visited William Oliver (sorry William!) all these luminaries in quantum computing, it will change the world in certain ways, but this changes everything and it's critical if we want to keep doing what we're doing in AI, that we make these transitions. So, I think people are going to be hearing a lot more about it and I wish you a great deal of luck. I think you'll have your 400 PhDs before long. [laughter]

ALEXANDRE MARCIREAU: Thank you so much for those kind words and yeah, of course, absolutely agree with the importance of neuromorphic engineering and how exciting it really is. So yeah.

BRUCE MCCABE: All right. I'll be watching with great interest and we will connect again, maybe revisit.

ALEXANDRE MARCIREAU: Thank you so much for visiting. I really appreciate the attention and yeah, a lot of rambling about neuromorphic engineering...

BRUCE MCCABE: No, not at all. Not at all.

ALEXANDRE MARCIREAU: … every time I'm given the opportunity.

BRUCE MCCABE: We want to get people started, and we want to get people into it because they're not hearing about it. And it really is fundamental. It's as fundamental as computing was originally, and it's more. It's fundamental to the future of artificial intelligence, and everyone now understands how important that is. And this is wrapped up in that future. So, yeah, it's a real privilege. Thanks again, Alex.

ALEXANDRE MARCIREAU: Thank you so much.

[music]

 
Previous
Previous

Neuromorphic Computing AND THE FUTURE OF A.I.

Next
Next

Should we worry about Skynet?