AI Evolution
Hear from the innovators and visionaries harnessing the power of artificial intelligence to reshape our future. Topics include AI's rich history, the industries it is revolutionizing, and critical conversations about the fears, boundaries, and ethical considerations that come with this powerful technology. Tune in to learn more about how AI is transforming our world and hear the voices of those at the forefront of its evolution.
AI Evolution
The Digital Divide and Energy Efficiency in AI
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
In this episode, host Michelle Gatchell shares two interviews from AI Week Columbus, a conference all about AI put on by the Enterprise Technology Association.
The first interview is with Doug McCullough, Executive Director of the Beta District in Columbus, Ohio, and he talks about the dangers of AI widening the digital divide.
The second interview is with Steve Brunker, one of the founders of Brain CA Technologies. He shares groundbreaking innovations in AI energy efficiency and reveals how mimicking the brain's complex networks can revolutionize processing technology and reduce reliance on power-hungry data centers.
Topics Covered:
• Understanding the Beta District's role in promoting smart mobility
• The responsibility of the tech community to prevent digital divides
• • Strategies for fostering equitable access to technology
• Addressing energy consumption challenges in AI systems
• Innovations in pattern recognition for efficient computing
• Support for responsible technology practices in community initiatives
If you have any topics you'd like us to explore in future episodes, please contact us at our website, evolution.life.
Welcome to AI Evolution, the podcast where we unravel the mysteries of artificial intelligence. Here's your host, Michelle Gatchel.
Speaker 2All right, got a great episode for you today Two different interviews, people that I met at AI Week Columbus, which is put on by Enterprise Technology Association. They actually are doing about seven different conferences AI conferences around the country this year, so go to Enterprise Technology Association's website to check out the dates and join us for one of them, because they are amazing. All about artificial intelligence. So many bright minds, including my two guests today.
Speaker 2So the first interview you're going to hear is with Doug McCullough. He is the new executive director of the Beta District in Ohio, and that is a consortium of businesses and government entities all with the like mind of moving transportation forward using technology. So he's going to give us insight into how artificial intelligence can create a digital divide and what we need to do to make sure that that doesn't happen. And then my second guest is Steve Brunker from Brain CA Technologies, and their company is a startup that is looking to find an answer to the problem of how much energy AI uses, and it's a fascinating concept and I can't wait for you to hear how they're doing it. It's going to blow your mind and change the way you think about artificial intelligence. So let's get started. All right, everybody, we are here at Columbus AI Week and I am here with Doug McCulloughough, who is the newly named executive director of the Beta District here in Columbus, and you know, before we get started, maybe explain what the Beta District is.
Speaker 3Well, sure so. The 33 Smart Mobility Corridor sits is a 35 mile stretch of highway that sits between the city of Dublin, the city of Marysville, union County, which those cities established a council of governments, and placed fiber along that road, fiber optic cable for network connectivity. They placed dedicated short-range communications on poles so that vehicles could have devices inside them that spoke to other vehicles. That whole scenario was rebranded as the Beta District and executive director was named me in order to help to entice automotive mobility innovators and thinkers and thought leaders to our area to help form the innovations that make up the future of mobility. And when we're talking about mobility, we're talking about all sorts of movement, not just individual commuter vehicles back and forth to work but, also farm vehicles, construction vehicles, distribution and delivery, and walkers and pedestrians and wheelchairs.
Speaker 2Drones.
Speaker 3Drones and air vehicles and all of those things. So the future of mobility is fascinating and Central Ohio wants to be a thought leader, if not the thought leader, where that's concerned. And that's what the beta district is and that's my job.
Speaker 2Awesome. All of these businesses not all of them, I should say a lot of those type of businesses also use AI and we're here today to specifically talk about, you know, responsible AI and how that should happen. What's your take on this?
Speaker 3Well, you know, we just listened to a really good talk on responsible AI. Just listened to a really good talk on responsible AI, and I think it's fascinating that we haven't lost the thread of our responsibility as a society, especially as a technical community, to monitoring and carefully considering things like responsibility on any new emerging technology. And so, a while ago, we were having conversations about bias and how bias could enter into large language models and into machine learning algorithms, and they would exacerbate and perpetuate themselves until we were experiencing deeper problems. We have held on to that thread and I'm really proud of us as a technical community to do that. And that's not the only thing that goes into responsible AI, sure, but I think that as a technical community, we are beginning to take seriously our responsibility for being responsible for the impacts of our technology.
Speaker 3You know these technologies are going to run rampant through the world. They are very fast moving and they're not human, so they don't follow human rules and they don't follow human regulations. And so we are responsible for building and inventing them in the first place in scenarios that we eventually can either control or can you know well, can control.
Speaker 2You mentioned a little bit about. You're going to be talking about kind of how AI can create a digital divide. Explain what that means.
Speaker 3Digital divides, like other divides, glom onto existing divides. We have a wealth gap, we have a health gap, we have pay gaps. We have all of these divides that exist and we already have a digital divide, and part of that comes from access to broadband digital divide. Part of it comes from a lack of digital skills in some communities or individuals more than others, and so every new technology evolution or iteration brings with it the risk that it does not benefit everyone equally and as a result of that, we experience divides. Those divides wind up having an economic impact. They wind up impacting our market efficiency, our ability to hire people and get them paid and have them consume goods and have them invest in their future. All of those things are impacted by our divide. So it's in all of our interest to promote a significant level of inclusion and I think, if we think of it more like a market or an economy, rather than just this person doesn't have a computer or access to the Internet or know how to use artificial intelligence.
Speaker 3Those are problems that are easier to solve, but a digital divide is more of a societal, complex issue that requires more thoughtfulness and more unity amongst an economy to address.
Speaker 2I'm just curious. When COVID hit, I think we saw a huge digital divide in the fact that when schools had to all of a sudden close down but teach kids, there were a lot of places that didn't have Internet access and weren't able to do that right away. Do you feel like we kind of helped push that forward and bring that more equal?
Speaker 3Significant progress has been made.
Speaker 2Yeah.
Speaker 3I think one of the questions that we have to ask ourselves is how was it that we always had that problem but did not come to terms with it until we had a global pandemic and it revealed also work from home. It wasn't just a school from home is. Everybody was at home kids, grandma working and we found that some people could work from home and others couldn't.
Speaker 2Yeah.
Speaker 3All sorts of interesting impacts happened out of tragedy, and that's one of my big questions what else are we living under right now? What conditions and characteristics are we assuming away? And as a technologist and as a proponent of using technology throughout communities, I have to concern myself with you know who has what and what are they capable of doing with it. So that we don't have a pandemic is its own problem.
Speaker 2I mean, it was horrible, whatever.
Speaker 3But there can be good things that can happen, like the introduction of artificial intelligence that reveal a digital divide. Why is that? How blind are we that we can't have enough visibility into the lives of our fellow citizens or our customers, even that we allow these things to sneak up on us? That's unacceptable.
Speaker 2Yeah, so are we getting better at seeing the future, so to speak, and making change ahead of time? Did we learn from this?
Speaker 3Yeah, I don't know that we're getting better. I will say that we got better, but the emerging technologies that we're building are moving faster and so we need to actually accelerate what we've learned before the next technological evolution comes around. And I don't know that we are matching pace with the acceleration kind of algorithm that's there, so I'm not disappointed in it, but I do think that we can and should try to do better, and I tell people whenever I'm talking about AI that quantum is right around the corner. When we were talking about cloud, ai was around the corner and we didn't believe it. You know that'll be years, and we think quantum will be years. Quantum is going to. We're going to have these same conversations around quantum and there will be something after quantum.
Speaker 3So I think what we need to do is institutionalize our responsible technology thinking, as we've begun. To institutionalize cybersecurity right, it's not something we leave to anything. To institutionalize cybersecurity right, it's not something we leave to anything. And I think we'll do better today and with whatever's coming next if we bake it in, as opposed to oh, we forgot to add responsibility or how is this sustainable and these kinds of things. Why do we ask these questions every time? That's unnecessary. So I look forward to us institutionalizing and baking in our monitoring and improvement of how we even manage a market and a technical economy.
Speaker 2So who's the institution?
Speaker 3Well, I do think things like the Beta District can help where that's concerned, but I also think that if you look at a place like Columbus that has a mature technical community with AI experts Columbus that has a mature technical community with AI experts, investors, universities, local governments and startups, large companies we are in a position to build the kind of consortiums and coalitions that can build policies and practices, and I do think that we can do that. I'd like to see us do it more, but it is a loose confederation, not necessarily a government or a government agency.
Speaker 2Okay, well, thanks so much for joining us.
Speaker 3You are welcome. Thanks for having me.
Speaker 2All right, that was Doug McCullough, the Executive Director of the Beta District in Columbus, ohio, and now for my next interview at AI Week Columbus. I talked to Steve Brunker from Brain CA Technologies. It's a startup company and they are looking to find a solution to one of the problems AI is causing and that is how much energy it uses in some of its large learning models. There's a lot of electricity used for AI and a lot of people don't realize that, but Brain CA Technologies is working on tackling that problem and finding a solution. So here's my interview with Steve Brunker at the AI Week Columbus put on by Enterprise Technology Association. All right, we are here at Columbus AI Week and joining me now is Steve Brunker. He is with Brain CA Technologies from Cincinnati.
Speaker 4Yes.
Speaker 2And Steve, thanks for joining us.
Speaker 4Thank you, it's a pleasure to be here.
Speaker 2You know, first off, let's just talk about what your company does.
Speaker 4Yeah Well, we're a startup, so a very small company. And my partner was thinking about AI and realizing that, gee, you know, when you think about how the brain compares to what AI processors do today, there's a real mismatch. And he said you know, think about a mosquito. A mosquito's brain is like a fraction of a gram, it uses a fraction of a watt and it's a small fraction of a cubic millimeter, and yet it outperforms even the biggest data centers that we have.
Speaker 4Doing AI things. It can do things like find a mate, hunt for food, do all of its bodily functions, and yet we can't even come close to replicating that with a computer right now. And so, pound for pound, watt for watt, he said there's got to be a better way to do this. And as we looked at the processors that are in all of the AI engines that we have today, we noticed that they all do math problems. And he said, well, ai is not really a math problem, it's a pattern recognition problem. What if we designed a processor that was focused on doing pattern recognition instead? And so he did. And so we've patented and developed a microprocessor that's based on a technology called cellular automata.
Speaker 4That's why the CA in our brain? Ca name.
Speaker 2Yeah.
Speaker 4And it does not use mathematics to do AI, it uses pattern recognition.
Speaker 2Interesting. So, and I'm going to go chat GPT, for instance, right, so, mathematical wise, when someone puts in a prompt, write me an intro to a podcast, right? How is? What it does can differ from yours.
Speaker 4So what happens behind the scenes when you put in that prompt is it converts all those characters into data, into digits binary digits, right and puts that into an engine. And then there's a massive neural network behind there that goes and runs through an algorithm to say, okay, well, based on these characters, what's the next character likely to be? And it runs back and forth. And it runs back and forth. And when you're training those models and that's really the expensive part of these things is the training part.
Visual Memory AI Development and Funding
Speaker 4It runs through all the possibilities and gets to the end and says this is what I came up with. And the trainer says, no, that's not right. And so then it back flushes a whole bunch of other changes and then says well, let's try it this way. And again and again it goes back and forth through the model and unfortunately, it goes back and forth through the model and unfortunately, it takes millions or trillions of times doing that in order to come. Lots of energy. Yeah, and so what? Our approach is? To say what if you had two inputs at the same time and we stimulate them? We noticed that, just like in the brain, when you stimulate two cells at the same time, they develop a relationship together. And so we found a way, using cellular automata and waves, to connect two input points together and recognize with a third point that that's a relationship between those two. And we can do that more and more and more sophisticated on a grid of hexagons, to recognize patterns and make the predictions using the one as an input and one as a predicted.
Speaker 2So what would you use it for?
Speaker 4So it's going to enable us to move compute the AI compute out to the edge, compute out to the edge. So instead of having to have these big behemoth processors, servers back in the data center that are grinding on these huge mathematical algorithms, our idea is to move that compute all the way out to the edge and say you know what, just like a mosquito's brain, I can do that right on site with very little energy and process that and get the job done, and we can have very specialized examples of that. So maybe a good example would be handwriting recognition. Once you've identified a bunch of scripts, you kind of know what that is. I could put that into a little model that we put into a phone and say now your phone recognizes handwriting or various other things like that, handwriting or various other things like that.
Speaker 2So it's not for turning loose onto the Internet and gathering a bunch of data. It's to be very specific to do a certain. I'm going to use the triangle thing you said you know A to B to C, a to C because you know exactly what your information image is.
Speaker 4Yes, we're training it, just like we do the larger models, but we're training smaller, more specific models to start with. But our idea is and we'll scale to this is to take many of those and put them together, kind of like your brain says a center that's focused on vision, a part that's focused on hearing, maybe on taste and all those different senses. Well, there are different parts of the brain. Well, we'll have different processors that focus on different tasks, but they still communicate. You know, you still know that when you look at a strawberry, you can imagine what it tastes like.
Speaker 2What have you tested it with so far?
Speaker 4Well, we're in the testing right now, so we're doing some very simple problems to start with and we're focused on doing some industry standard ones, so we can compare how our models perform to things that are already in existence.
Speaker 2Yeah.
Speaker 4Measure. Well, how much energy did we take to solve that, to train it, and how did that compare to what a GPT model took?
Speaker 2Ah, so those are because ChatGPT is open source out there already for you to do the similar testing Right and also, you know, any of the large companies are publishing benchmarks on how their models and their chips perform.
Speaker 4Okay so our objective is to put our data right out there with it.
Speaker 2So I don't know a lot about our brains and how they work per se, but in my mind, from what you said, you know visually, like when memory wise is where I'm headed with this right, I have to give myself things visually to remember stuff. If I get somewhere once I can get there again. My mom cannot. There's different ways that people think. Right, absolutely, and visualization is a gift for some people. I think right.
Speaker 4Yes.
Speaker 2So if you're using this visually different than the mathematical version, right will people be able to understand it that don't have a visual mind? Does that make sense? What I'm asking?
Speaker 4I think it does.
Speaker 2Yeah.
Speaker 4Let me take a stab at it here yeah. So a good way of thinking about the old way of doing things is that the old processors actually separate memory from the compute. And it's a traditional architecture. There's a big array of memory that has all the instructions and all the data we're going to act on, and then we take a piece over to the compute, and then we we take another piece over and then we tell it here's what you do with it. I want you to add those two things.
Speaker 4And then oh, there's the answer. I'm going to put it back into memory. So it goes back and forth between memory and compute. But if you think about your brain, your brain doesn't have a separate part of a design is that the cells that we have in our architecture have both some very simple processing but also maintain the memory. And so by networking those together in large numbers is how we accomplish intelligence. Yes, exactly that. It self-organizes, based on the observations that we feed it.
Speaker 4And what's nice is that we don't have to presuppose that. Here's our inputs, and then we're working for these particular outputs and we'll see how close we can get to our target. We just make the observations and it says well, you know, when I smell this red thing, it smells like a strawberry. And now it says the red thing smells like a strawberry, and that's the strawberry. And now it says the red thing smells like a strawberry. And that's the relationship. And now if I see a red thing, I could say, well, it might be a strawberry. Or if I smell a strawberry and say it's probably red, you know, it can go either way. I don't have to know which direction I'm trying to Hold on.
Speaker 2Are you saying you want AI to have smell-o vision? Yeah, I want all of it, because that would be would be cool.
Speaker 4yeah, I think all kinds of of inputs and anything we can digitize, whether it's a a sense, an organic ascent or an optical image or anything um, how big of a server do you need to be testing what you're working on, compared to a regular ai test?
Speaker 4well, that's. What's so fun right now is that we're working on really small computers, like PC-sized computers today, to do our simulations, but the models we're playing with are small. As we grow, we expect to be using larger and larger ones. But the interesting thing is we don't think that that trajectory that we're going to be on is nearly as steep as the ones that are needed for the AI that we have.
Speaker 2Have you done a funding round yet?
Speaker 4Yes, we did. I'll call it a friends and family funding round and raised money to get ourselves going. So we've had about $2.2 million to run our business for about 18 months.
Speaker 2Nice.
Speaker 4And we're just starting now to talk with some venture capitalists and private equity firms to look at us about investing and taking us to the next level.
Speaker 2What's the next level?
Speaker 4Well, really, it's advancing the schedule that we will get there. It's just we're not getting there as fast as we want, and so we're interested to get there a little faster with a little more money.
Speaker 2One last thing Is there anyone else in the country or the world doing similar?
Speaker 4We've done a lot of research trying to answer that question and we've patented our technology. So the patent is one indication that no one else is working on it Sure. The patents were really broad, so that's another indication that no one else is working on it. And the universities that we've spoken to and the professors that we've spoken to that are looking into cellular automata have pretty much told us that the only applications they've seen so far are completely theoretical. So as far as we know, we're the first practical implementation of cellular automata.
Speaker 2Okay, anything else we should talk about, steve.
Speaker 4No, I mean, that's my main focus for these days. It's just so much fun. We're having a great time meeting some really smart people and sharing what we're working on. Everybody has pretty much the same process they go through. They kind of look at it and say you know, I don't know about that. And they shake their heads, but by the end of the conversation they're like, wow, this is big and we think it's really big, yeah, great.
Speaker 2Well, good luck. I can't wait to hear your journey.
Speaker 4Thank you. Yes, I'm excited about it and I will look forward to keeping you posted.
Speaker 2You know I neglected to ask you kind of your origin story. What made you so interested in AI to begin with?
Speaker 4Well, it's interesting, I actually studied AI in college back in the 80s.
Speaker 2Really yes, what?
Speaker 4college Northwestern.
Speaker 2University.
Speaker 4And I studied under the head of linguistics, who was also an AI professor as well, and at that time, you know, it was largely centered on linguistic studies. But so I've always had an interest in AI. But really I have to give credit to my business partner, Jerry Felix. Jerry was the one who came up with the idea.
Speaker 4He's been working on this since 1991, on and off and he finally retired from a software business that he successfully was able to sell and focused on this and came up with the breakthroughs that really have made the difference. And he called me and said hey, I think I've got something here and I was skeptical at the beginning, but the more we talked about it, the more exciting it got and we've pulled more and more people in and the excitement is growing.
Speaker 2Great, all right Super exciting, yeah, very fun. Okay, well, thanks again for joining us.
Speaker 1No, my pleasure really. We hope you enjoyed this episode of AI Evolution. If you're as fascinated with the capabilities and possibilities of AI as we are, don't forget to subscribe on our podcast on your favorite streaming site to hear more conversations with the brightest minds in the field. If you have any topics you'd like us to explore in future episodes, please reach out to us at our website, aievolutionlife. We'd love to hear from you. Share your thoughts on this episode and more on our Discord channel. You can find the link on our website. Until next time, keep your curiosity alive and remember the future of AI is just a podcast away. Thank you.