Tania Roy, PhD, researcher at the University of Central Florida talks about new brain-like AI and how devices could possibly work without internet connection.
Interview conducted by Ivanhoe Broadcast News in June 2022.
Tell me a little bit about what you and your colleagues here in this lab are working on.
ROY: We are working on developing devices for artificial intelligence using specifically two-dimensional materials. It’s a different category of materials called two-dimensional materials. As the name suggests, they’re extremely thin. So that’s how – I’m generalize it. They’re extremely thin materials, a special family of materials called two-dimensional materials, which are being used to make devices for artificial intelligence applications. So, devices in the sense I mean, nanoscale. Small, tiny devices which will be connected to form a circuit, and then they will behave in a particular manner that will make you feel that something – a robot is moving, or we can understand what humans are saying, those kinds of applications, yeah.
So, as you’re doing this, there will be no need for a connection to the internet?
ROY: So, you are seeing around you a lot of different artificial intelligence applications going on. Google just a few days back showed the LAMDA Artificial Intelligence, right? So those are lines of code. What are they running on? They are running on some hardware. Where is the hardware located? They are located in remote servers. And that’s why you have to connect to the remote servers through the Internet all the time. So, a lot of progress can happen in terms of software. We are in that right direction. Everybody’s thinking only in that perspective. But then what will they run on? If it requires a lot of hardware to run those lines of code, then one day you will have an entire city filled with just hardware, right? And that’s not feasible. That will limit how much we can do in terms of artificial intelligence. That is why what we are trying to do is make small devices which will mimic the neurons and synapses of the brain. And then, we will be able to make the hardware as small as the human brain is. And that will enable us to realize very complex artificial intelligence tasks on a very small footprint. That’s the idea.
You alluded to this a little bit in the beginning, but when you have this just tiny area with incredible capacity, like the human brain…
What are some of the applications? What kind of things can run off of that tiny brain-like device?
ROY: Say if somebody is trapped in a remote location. Let’s say it’s a remote mountain and mountaineers are trapped there. And we send a drone, which has a camera eye, and it can just go and locate those people and rescue them. It can go to remote areas and do that without any prior knowledge, any prior, any internet access to continuously supervise it. It can do it completely unsupervised. So that’s one of the applications. One interesting application could be, let’s say if you have a sunglass which has a camera fitted to it and a microphone fitted to it, and a visually impaired person is wearing it. That person will be just able to see and navigate so easily in just that very, very small, tiny object, tiny pair of sunglasses.
Do you have a third example?
ROY: Yes. So, let’s say we have a lot of – I mean, we have family members at home who are getting injured, let’s say, and they need assistance. So, we are worried about their health conditions, what is going to happen all the time? Let’s say we have an Apple Watch type of monitoring system, but it is more than what the Apple Watch can do. Apple Watch can diagnose the conditions, but it cannot – I mean, monitor the conditions. It can report the blood pressure, heart rate, etc., but it doesn’t really know when things are going wrong. Let’s say we have artificial intelligence to say that something has gone wrong and then call 911 immediately. So, it’ll just put us at peace, right? Address that OK, we know that somebody is there to take care of our elderly relatives. Those kind.
And you had mentioned again in the beginning, that this is what you are working on in the lab, is those microscopic, nanoparticles. Can you describe again what it is that you’re working on? And how do you look at it? How do you manipulate it? How do you know where you’re moving everything?
ROY: Sure. So, these are materials which are atomically thin, as thin as a layer of atoms. And we can grow them very efficiently in our lab and in other labs also in our university. So, we take those materials and then we put them through the standard processes of fabrication of circuits. And that’s how we can manipulate them, get small devices out of them and use them the way we want to.
When you’re talking about small devices, for our viewers, is there something you can compare it to? Like, it’s the size of a quarter. It’s the size of your fingernail.
ROY: Each device that we have is the size of 100th of a human hair, diameter of the human hair. So, it’s micro-scale, and the thickness of those materials are 1000th of that dimension. So, they’re really, thin. And in terms of lateral dimension, they are 100th of the human hair.
And to be able to manipulate those and work with those, can you talk to me about the equipment that you’re using? Just basically, what will your you and your colleagues be looking through in order to work on this?
ROY: Sure. So first, we want the material to be adopted by the industry in whatever several years it takes. So, we want to make it as commercially viable as possible. That’s why we are using standard cleanroom techniques. Whatever technique is used for silicon, the conventional material, we want to use exactly those same processes so that the industry can adopt that technology as soon as possible. So, we grow the material in our lab, but the processing steps that are required are exactly what, say, Intel would like to follow in the distant future.
And the machine that your colleague will be on to kind of demonstrate for us what is that called? How should I describe the process that she’s going through?
ROY: OK. So, we’ve already made the devices. They constructed the devices already, and they are now going to measure the properties in a probe station under vacuum. The probe station – the device, the chip will be kept under vacuum, and we’ll measure the device in a probe station with a semiconductor parameter analyzer. So, they will blast voltage and measure the current through the device to see if the response is just like the brain synapses are.
What will that measure? What will that tell you?
ROY: We pass voltage and measure the current. So, if the device behaves like the human brain synapses, we will see the signatures clearly.
Is it comparable to neuroscientists when they image the brain, and you could see parts of the brain light up or firing? Is that like what you’re measuring here? You’re looking for things to light up?
ROY: So, we have – let’s say one particular – they see the neurons light up like that. We have that neuron. We know that it is the neuron. So now to that, we are applying the voltage because we don’t know if it’ll behave like the neuron or not. So, we apply the kind of stimulus that the neuron in the brain will get, just different magnitudes of voltage. So, let’s see, we apply voltage just like the way the human brain will get the voltage, the neurons will get the voltage. Then we will – the neuron’s supposed to respond in a particular manner. We should see that same response in our device. If it doesn’t, then it’s not a good device. Then we go for another run and make another device like that.
I had asked you this question before, and I know scientists hate when I ask this question, but it’s a consumer-friendly question. How long?
ROY: Yeah. Definitely, the goal is so that it can be commercialized someday. So, if everybody works together on it, we should be able to do it in five years. But realistically, maybe 10 years to actually see a commercial product out of these things.
And I want to ask you a couple of questions just building on the research that you’ve done. You and your colleagues have been able to devise something that mimics the human eye.
ROY: Yes.
Can you describe that for me?
ROY: The broader goal is, let’s say when we are talking about the camera, camera will go and see – so what – if we can start the artificial intelligence process right at the camera node. So, the camera itself is a part of the A.I. Right now, what is happening is the camera takes the image just like our phone camera or anything. And then that image is sent to the cloud for processing to say that, OK, this is a cat, it’s a human being, all of that. We want to do everything within the camera itself. So, one part of our work was the one we described earlier that we are trying to make things much more compact. And the first – before – another part is that the cameras if you just considered – don’t consider the optics. So, the optics of the camera remains the same – lenses, everything is the same. But then we have an imaging system of the camera. That imaging system is the part of the artificial intelligence. So, the kind of devices that we are making, they will be able to take the image and tell us that this is a cat. So that’s the whole idea.
One-stop. And again, how – where are you in this process? And, you know, what are the practical applications that you want to see for this down the road?
ROY: The practical applications, we can say the same thing as before. The camera will be able to look at a stranded mountaineer or be able to provide turn-by-turn navigation because it will scan the environment around us and tell the person that, OK, turn left, turn right, depending on the situation and look at obstacles coming in the way. So autonomous navigation is a very – it’s one of the applications of this technology, for sure. And that’s why it works for space as well. Autonomous navigation in space, let’s say. Yeah. You were saying – sorry, there was a first part of the question. That’s OK.
How close are you to realizing this? About the same?
ROY: Oh, where are we with the work? So, what we are doing is that we are – right now we are doing something equivalent to making each optic nerve, let’s say. Each – how each optic nerve will behave. But to – we have to make the whole eye. So the product that we have made so far is just one small part of the whole. Now, next step is to connect all of them together and make the actual pattern recognition hardware.
So, you have one of the optic nerves, if you’re comparing it to the eye?
ROY: And for the eye, we shine light on the device, and we see the current out of it.
And you had mentioned – I think you mentioned something like silicon that you use? What is the material that you use?
ROY: The materials we are using are – one of the materials molybdenum disulfides.
Again, what is that like that people might be familiar with? Is it like silicon?
ROY: Like silicon. It’s just much, much, much, much thinner.
Is there anything I didn’t ask you, Tania, that you would want to make sure that our viewers know about the work that you and your colleagues are doing?
ROY: The eye work that is supported by Airforce Research Lab, so we are working with them to realize the whole product, and the entire neuromorphic work, which we call it neuromorphic because it is mimicking the neurons. Yeah, it is supported by National Science Foundation.
What kinds of applications?
ROY: Let’s say again to identify any attacks or any – if we are being targeted. So, they want to load the information – right now, what happens is they have only specific information that can be loaded into, let’s say, even a drone or any recognition system. And it’s very limited. So now, with this technology, they want the whole device to be much more able to detect any kind of threat. Yeah.
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
If you would like more information, please contact:
Robert Wells Zenaida Kotala
(352) 213-54 (407) 446-6567
Robert.wells@ucf.edu zenaida.kotala@ucf.edu
Sign up for a free weekly e-mail on Medical Breakthroughs called First to Know by clicking here
