The Smart City Podcast

"Not the Terminator" - Ted Connell of Blue Skies AI Explores Benefits, Challenges and Real World Applications of Artificial Intelligence

July 18, 2022 The Smart Cities Team at ARC Advisory Group Season 7 Episode 12
The Smart City Podcast
"Not the Terminator" - Ted Connell of Blue Skies AI Explores Benefits, Challenges and Real World Applications of Artificial Intelligence
Show Notes Transcript

Listen in as Ted Connell of Blue Skies AI Explores Real World Applications of Artificial Intelligence including a general discussion on the evolution and opportunities for industrial / commercial AI as well as a detailed discussion of healthcare inspection applications.

A snippet of the conversation:

Jim Frazer  

What are the challenges of AI? And perhaps even before you answer that, Ted , for those of us who might not be steeped in the technology, what really is AI? 

 Ted Connell  

That's the greatest question. Because most people out there think of AI, they think of the Terminator, and Arnold Schwarzenegger going around blowing everything up. And what they were describing there is consciousness, what AI is, is math. It's not scary. It's just math. And it's like a loop. It's like plastic. And it's like electricity, AI is going to go everywhere, because it makes everything better. So, think about your brakes in your car, your ABS brakes, remember, we got ABS brakes, I can now put a couple dollars’ worth of electronics in there and make your brakes a lot better, because they're measuring how hard they're pushing in there. They're deciding when they want to push depending on what you're trying to do and how hard you're pushing your brake pedal. But we can make your brakes better. There's no consciousness in there. It's just a mathematical equation that's watching a pressure here and a pressure there and a temperature and it's doing a mathematical calculation to say what's the optimal. So that's all AI is just math trying to optimize equations.

Would you like to be a guest on The Smart City Podcast?

If you have an intriguing, thought provoking topic you'd like to discuss on our podcast, please contact our host Jim Frazer

View all The Smart City Podcast episodes here: https://thesmartcitypodcast.buzzsprout.com/

Jim Frazer  

Welcome to another edition of Smart City Podcast. Today, I'm thrilled to be joined by Ted Connell of Blueskies.AI Welcome, Ted,

 

Ted Connell  

thank you very much for having me.

 

Jim Frazer  

It's great to have you here today. Ted, let's get started by me ask you a little bit about your background. And how did you come to the world of AI?

 

Ted Connell  

Great question. So, I started as a process engineer in the stainless steel industry. So, I've lived the life growing up in the industrial sector. But AI really didn't come to being till around 2010 with IoT in such, what happened in around 10 years ago or so is we got enough compute enough storage at low enough cost that people like iPhones and you know, television sets could start pulling AI in their business. And the early people to do it were the consumer electronics players. And my career progressed, I was with IBM in the service business, and I was working with a lot of the consumer electronics players trying to understand how do we leverage these new technologies? What do we do with them? And how do we make money with them? So, 10 years ago, I was working on the Smart TV Alliance when we were connecting TVs on putting platforms together. And it's just progressed towards that. So, I've been in the ecosystem, most of my career and AI for the last 10 years or so. Is really, I think that interest question. Okay,

 

Jim Frazer  

What are the challenges of AI? And perhaps even before you answer that, Ken, for those of us who might not be steeped in the technology, what really is a? That's a start.

 

Ted Connell  

That's the greatest question with that. Because most people out there think of AI, they think of the Terminator, and Arnold Schwarzenegger going around blowing everything up. And what they were describing there is consciousness, what AI is, is math. It's not scary. It's just math. And it's like a loop. It's like plastic. And it's like electricity, AI is going to go everywhere, because it makes everything better. So, think about your brakes in your car, your ABS brakes, remember, we got ABS brakes, I can now put a couple dollars’ worth of electronics in there and make your brakes a lot better, because they're measuring how hard they're pushing in there. They're deciding when they want to push depending on what you're trying to do and how hard you're pushing your, your brake pedal. But we can make your brakes better. There's no consciousness in there. It's just a mathematical equation that's watching a pressure here and a pressure here and a temperature and it's doing a mathematical calculation to say what's the optimal. So that's all AI is just math trying to optimize equations.

 

Jim Frazer  

Okay, So let's move on to you know, there's been great hype about AI in the world that it will be deployed everywhere. Some of that is overhype. Some of it is true. So what are the challenges of AI today?

 

Ted Connell  

Great question. Today, it's very hard in the industrial sector, you want to focus on the focus on our industry in the industrial sector. Today, AI is very hard because most of the data is siloed. Most of the data is not in a system, like the IT systems that you can network it together. And so the first problem is just getting the data you want. And that's very hard. Once you have data, you have to go through a long process of labeling the data, what am I looking at? And that takes a lot of time. And then you have to train models. With all these big data's. That's the State of the State today. But that's not where we're going to be in tomorrow.

 

Jim Frazer  

So just to review siloed data. Yeah, we need. So first we need to aggregate the data. Yep. We need to label the data, perhaps with a common data for data model.

 

Ted Connell  

So not only that, but once you let's say I'm looking at an image I have to label what is in the image. What is this? Is this a dog or a cat? Is this a defect are a good part? So I have to understand what the data is I'm looking

 

Jim Frazer  

and then lastly, we need to create some mathematical model to improve that situation upon which we're going You're

 

Ted Connell  

thinking about and the timeline could be that the mathematical model takes 10% of the time and 90% of time is getting the data and making the data useful. So the training is not necessarily the problem though the training can take weeks, but getting the data can take a lot longer.

 

Jim Frazer  

Certainly, I'm interested because I my initial thought was particularly in the industrial world and models, there, there may be a plethora, there may be innumerable amount of different models for industry specific processes, or even manufacturers Pacific's patented processes, secret processes, secret sauces, whatever. But that that is all just coding, which could happen quite quickly once you have this data repository upon which to act.

 

Ted Connell  

Yeah, so absolutely. So if you're a company, you're going to go through a digital transformation to get the data out of the silos, make the data available on some kind of network so that you can feed these engines. So that's one path. And this conference is talking a lot about that the ARC Forum is talking a lot about that here in 2022. Here in Orlando, that's, that's a big theme. And it should be, there are other ways to get AI to work today. So AI is working on your phone, right now, if you pick up your phone, AI is working on Facebook, it's working on Google, it's working on all those apps that are not necessarily on your phone, but they're in the cloud. So today, you can buy appliances that are standalone solutions that can do AI for you. So because of COVID, a lot of people were doing AI to say are people wearing their masks, you know, personal protective equipment, you can't walk into the school unless you have a mask on or something like that. So a lot of people applied AI to do things like that, and they did it on a PC. So because the technology is advancing, and because compute is becoming so powerful, we can now run AI on literally laptops, and phones. So the technology is changing, there are opportunities. Now, if you're an industrial company, you might not be able to optimize your whole process, because you don't have the data from all of your process. But you can drop an inspection solution in there, you could drop a predictive maintenance solution in there, you could you know, they're always to do it, you can monitor your electricity with tools and reduce your with AI and reduce your energy input. So you know, with smart buildings, you can monitor people coming into the building know when people are in rooms and change temperatures around and things. There are things that you can do that have economic value. But the big win is when you get what we've talked about, under the silo data's open up the data. And when all that's flowing, then you can optimize your whole business. But that's attorney. Yeah,

 

Jim Frazer  

indeed, indeed, if you if you do have that common data repository, the applications you can imagine, well, are ones but cannot imagine today, things will things will be created that we couldn't have imagined, just like the phone created a range of different things that we could never have imagined before. Absolutely. Let's go back. You mentioned digital transformation. Yeah. And let's focus on perhaps the deployment. Issues around AI. Digital transformation has three pillars. The first being the technology itself. Yep. The second is the business processes that change because of that technology deployment. And then the third is the people that must embrace both that technology, maintain it operated service it as well as support those new business processes that may have been unseen. So you're is AI fundamentally different than some new tool set that someone might get at their industrial job? Or is it just another tool?

 

Ted Connell  

It's a tool. So at the end of the day, it is a tool set. And, you know, remember a generation ago, we went through Lean Six Sigma. That's a good bench, good, good benchmark for how you go through digital transformation.

 

Jim Frazer  

Indeed, so you're at your organization, you is blue skies. A is What does blue skies do.

 

Ted Connell  

So we're a spin out of Intel. We've been partnering with Johnson & Johnson for the last couple of years, because J&J had a inspection problem. Quality is everything in pharmaceuticals, and they're using human beings to inspect product. And that's across the industry. So it's not just this company. And human beings cannot concentrate for eight hours straight. We have to have bio brakes. We can't stop the line. It's a job that is thankless, and it's very, very difficult. I started as I said, as a process engineer. I've worked with inspectors. Not only is it a job that you can't do, so you have low self esteem because you can't physically do your job. When you find a defect. You have to go tell the person who made a defect that they did something wrong, so it's not a fun job. And as you know, in the industrial world, there are no aren't enough industrial workers. So we're paying people to do a non value added job of inspection where we really want to apply them to a value added job and move them where they can do not only feel good about themselves, because they're doing a job they can do, but they're adding value to the company. So I built a company to use AI to replace the human to do inspection and pharmaceutical. And we run it on Intel Core i Five commodity off the shelf hardware, the whole model was we had to do something that's effective. It’s got to work. And we're 99.999 plus percent efficient, so far greater accuracy than human beings. And we had to do it to scale. So j&j has plants all over the world, 400 plus plants. Each plant has a different skill set, but each plant will have a process engineer and site services. So all you have to do with our product is install it. Which site services can do show it the gold standard of your inspection, which every company has is how you decide this is good, or this is bad show with those examples. The tool trains itself, you don't need to program you don't need to know AI. So we built a spot solution that solves an immediate problem for our client. And we're just now launching it to the world. So we're really excited to be here at the the ARC Forum. We've had great, great conversations with people. And the good thing about the ARC forum, is you get industries together, and we were talking about digital transformation. Johnson and Johnson, when they started their digital transformation effort went out and benchmark best of breed. And that's how I was introduced to them as they wanted to benchmark the Intel factors in Intel will let you do this. And I advise anybody, if you're thinking about digital transformation, talk to Intel. The reason is the factory in Chandler, Arizona. And there's only a little bit I can tell you about it, but it's building to the manufacturing process is building a wire and all their electronics that are 10 atoms in diameter 10 atoms or 10 gold atoms in the whole wire. How do you do that this is manufacturing at the atomic level. So that factory goes through a petabyte of data every day, with billions of endpoints talking to each other. That's how they do it. So they've gone through this digital transformation because they had to make that integrated circuit. So I would benchmark Intel, they have some best practices. And j&j did that. And they benchmark some other companies like Georgia Pacific, who's got a digital transformation practice. That's it's very provocative and thought that very thought invoke invoking, and work together, don't try to do this alone, because it's new for everybody. And there's some people that are ahead of other people. And if you're not a competitor they're willing to share.

 

Jim Frazer  

So let me ask let's let's drill down into into your application. Okay, you're inspecting a pharmaceutical product of some sort? What, what sensor? What data do you collect and contrast to decide what's good or bad? Is it is a visual? Is it a camera?

 

Ted Connell  

If it's a camera,

 

Jim Frazer  

it's a camera? So what happens? Is it? Is it just a visual matching system? Or for the uninitiated to me, I think to you know, the maybe a visual metric system, I'm thinking about what about, what about other defects that are unseen by by a camera so readily or, or otherwise? Yeah,

 

Ted Connell  

so we can only do what we see. That's the whole purpose. And so their spec was point 0.15 millimeter. So we can see we can see smaller than that with a higher resolution camera. So the size of the defects, we can get very small electronics, we're going a lot smaller than So the size of the defect is dedicated, just tells you what resolution camera you need. And at what distance you are above what you're inspecting. It's just straight physics. And then we use convolutional convolutional neural networks, machine learning, if you will, deep learning to train our algorithms. So there is a lot of math in there. But again, we are using, you know, the math is the hard part for most people. Who, so scaling at Georgia Pacific, and Johnson and Johnson and a lot of these companies. Most that I know have a corporate data science team, but they don't have data science teams in the factory. Because you can't data science are hard to get and they don't want to work in Muskogee, Oklahoma. It's as simple as that. And so how do you scale your data scientists? That's that was one of the problems that was when we went in with j&j. That was one of the problems we had to have. So how do you scale people who work at corporate but need to look at stuff that's out in the factory so I can connect my appliance, and let corporate look at all the pictures? Because I'm networked. So I can now have all their data scientists As remote into my tools and do data science, if they wanted to do, let's say they wanted to do something different. It's a tool that they can use to go do that. And they can stay anywhere in the world. So we did this during COVID. And that's why we have an autonomous tool is because we were not allowed to go into the factory, our engineer at j&j was not allowed into the factory. It was that difficult. So it was literally when we shipped it, it was a moon launch, we'll never touch it again. So we have to do everything autonomously or remotely. So walk me

 

Jim Frazer  

through with, with, I have a very accurate camera. Yep. And I take two pictures, and one is approved, one is disproved. I could imagine a very simple scenario I have I have a pixel, I have an account of pixel by pixel x and y axis. You know, a very, very simple mathematical function, well wouldn't be if if certain number of pixels don't match the good picture, then it's it's a reject. What type of next step mathematical functions might be applied to those two images to define that one is rejected.

 

Ted Connell  

So a reason for the data scientists to go in is they have a different SKU, they have a new product, and they have to retrain a model for the new product. Okay, so that's really, you know, once I've got a model trained, I don't really need to do much, it might migrate a little bit, and I have to look at it occasionally, and maybe retrain it occasionally. Because it's kind of a weird dynamic of AI, as the models can migrate can move over time. Because they're continually learning, you have to make sure they're, they're going in the right direction. So there is a supervisory role that we play as these models work. But we're where you automate that. So you were automating that,

 

Jim Frazer  

but I'm trying to really drill down to the core of what, what's magical about that mathematical AI mathematical function. Because I could, I could, in my vernacular approach to this I, I can see a digital image of, you know, whatever, 10,000 pixels by 10,000 matrix and another picture 10,000 by 10,000, of an almost identical product, but but every one pixel is different. I could create a very simple model that says if one pixel is different average, okay. But I sense it's not as it's not as simple as that, yeah, because

 

Ted Connell  

you're gonna get the images are on a conveyor belt that's moving, so they're not blister packs. So you might have a blister pack here and your image, or you might have a blue speck here and your image, so your model wouldn't work, because it's not perfectly in the middle. So in the old AI world, they did exactly what you did. It's rules based, it's Windows base, and they say, this has to be two millimeters this week, three millimeters. And it's all calculated out that way. And you know, someone has to go in and program it. That's absolutely not what AI is. So that's, that's,

 

Jim Frazer  

that's very valuable for for our listeners. So the mathematical model may or may actually define the legal approved shape of a band aid. So there was something like that the state

 

Ted Connell  

of the state is we are, the whole industry is trying to understand how the mob why the model predicts what it predicts. And the government is pushing this because to us in defense, you have to have that, that that, that rule, whatever, whatever you want that that requirement, right now, these models trained themselves, and we can test them and know how accurate they are. We don't really know we know why they're going through the loops, but they're going through these loops trillions of times. So the trail back of why the decision was actually made, it's still work in progress, we'll probably have that out in a couple of years, Intel has a whole team working on it. Because some industries will require that automotive requires that for certain use cases, but we're not there. See,

 

Jim Frazer  

you're very likely to apply AI to the timestamp changes in the in the output to determine the how that got to its point. There must be incredible amounts of data if you're doing it. If you're if you're also

 

Ted Connell  

to understand why the model decided to do say this is good, or why the model said it's bad. You go through the model and you see all the different loops it goes through to calculate it. And then you have to put vectors on them and understand what were the key factors that made that decision. And that's what the work they're going through now. Because they it's just new work hasn't been

 

Jim Frazer  

done. Because I got an incredible amount of work because every iteration happens. How quickly Oh,

 

Ted Connell  

I don't need pico seconds. Right. Right. And you might be

 

Jim Frazer  

talking about a month's worth of data. Yeah. Times a couple of vectors. Yeah, so so so it's it's not a trivial situation.

 

Ted Connell  

I just had the required when we were building out Ai no one had asked them the question, why is it doing this? Now that someone asked the question and it has economic value, someone's going to answer that question,

 

Jim Frazer  

you can certainly see where it where someone would ask for the documentation. How do we get where we are? Absolutely. That's, that's fascinating. But

 

Ted Connell  

I can for for inspection, I can prove that I'm 99.999%. And I can repeat that proof. Just give me you know, put samples under things that were good and bad. And so we can show you how accurate we are. We just can't tell you why.

 

Jim Frazer  

Okay, so that's, that's so far, this has been been very enlightening. What are the obstacles to AI today,

 

Ted Connell  

getting data, getting data scientists, and a big obstacle, here's a big obstacle, having the disk making the decision for what hardware you want to buy, if you're gonna buy hardware. So I was talking to farmers, for I was up at Ford talking Ford. And they're, you know, 3d printing, machine vision, autonomous driving robots, everything Ford's doing all this kind of work in Deer above Dearborn at the Advanced Manufacturing Center. And I was talking to one of the engineers there and he's like, you know, you got these tensor chips with Google, you know, Facebook's building chips. AMD is better than what I What's Nvidia. I don't know what to buy right now. And this is a serious engineer who has been doing machine vision for 30 years. He's like, there's, it's risky. And so that's, you know, a concern, don't do I want to spend money today and not get it back. That's a big push for the cloud, moving it to the cloud, as opposed to buying it yourself. So that's one of the obstacles is it's confusion? You know, they've never AI hasn't been offered as a solution. It's been piecemeal.

 

Jim Frazer  

Well, it's interesting. I mean, you're, you're gone very quickly through this. But let's, let's back up to simply the data set. Yep. That's not a trivial attack. That's not a trivial piece of work to go and try to assemble all that data, catalog it, label it, organize it and have it to be operated on with any mathematical function,

 

Ted Connell  

you forget if it's even. Yeah, you know, Brownfield? Absolutely.

 

Jim Frazer  

Absolutely. That's a challenge in itself, then we have, particularly the challenging environment we live in. There are only so many data scientists, and it's very title hard to get it.

 

Ted Connell  

I got job openings.

 

Jim Frazer  

There's all there's only so many. And that second, that last one, what chipset do you pick? Is is, and I'm sure each one has its advantages and disadvantages, and what's

 

Ted Connell  

coming out next year? You know, it's, there's so many difficulties in it. It really depends on what your skill set as a company, what do you want to be when you grow up? Because remember, when we first did ERP systems, I have gray hair. And, you know, the cost of the ERP system was irrelevant. I remember when Koch did an ERP system globally, they gave Ernst and Young $300 million to integrate it. And then what everybody realized after that first iteration of ERP is don't customize anything, because you can't want to customize it, you own it, you're now a software company. And so that's the decision you're making? I mean, do you want to be a software company you want to be a hardware company? Or do you want service providers to provide it for you?

 

Jim Frazer  

Indeed, can a end user today work backwards and say, Hey, I've got here's a challenging problem that I would like to solve. And it's arguably maybe unsolvable by other by any solution other than an AI based solution. Do Is that do that? Do that end users go and say, Listen, okay, let me go out pursue an AI powered solution to solve that previously unsolvable problem. Do you see? Yeah, ideal starting to do that?

 

Ted Connell  

Um, well, that's what we're doing with Johnson and Johnson. We're using AI to, to do inspections. So that's it. For the problems that you have, you know, pick your biggest problems, I would absolutely talk through it with industry peers and data scientists say is it something that the AI could help with? So it really depends on what the set of the problem is. But what are computers really good at? They're really good at 100% Focus. It's like, you know, the kids book. Where's Waldo? Yes. So the computer can go through that book of wells wall, where's Waldo before you can open the book. Literally, by time you've opened the cover, it's finished it. It's really good at finding minut little things. So there's some skill sets computers have that we as human beings don't have. So if the problem involves looking at a lot of stuff and coming up with a creative solution to it, don't give it AI. AI won't work there. That's what humans do. If it's a mundane, repeatable tasks that you're doing a lot, AI will solve that. If it's mundane and repeat. If it's a task that you think is mundane and repeatable and you don't like doing it,

 

Jim Frazer  

give it to AI. Like perhaps driving your car. Hey, I like driving my

 

Ted Connell  

car. So I don't want to accept certain times but yes, exactly. When I'm in traffic, I don't drive my car. I have a stick shipping duty now. So yes. So yeah, mundane tasks, give them you know, at the airport, the trains moving back and forth. You know, they're they're things that computers do incredibly well.

 

Jim Frazer  

That's, that's very good. So what do we see for the future?

 

Ted Connell  

So I believe AI is gonna go everywhere. AI is just math. So a lot of people are putting a lot of value in the IP behind the algorithms. I know some of my ex companies are doing that. Good luck. It's just math. If I can solve a math problem one way I can solve it another way. So what's IP? And if you have IP, what if I put it through a Laplace transform? It's now mine, huh? What do you do that? So I don't know. I'm not a lawyer. I don't like paying lawyers. But I don't know how you patent math.

 

Jim Frazer  

That's true. It's maybe most of the copyright issue.

 

Ted Connell  

You maybe you could do a copyright? I don't know. Maybe. But but

 

Jim Frazer  

you changed a little bit. Exactly. I'm

 

Ted Connell  

not copywriting it anyway. It's gonna be there's gonna be a lot of money litigated.

 

Jim Frazer  

And that's that that's that. What else have we not covered? You? We're, we're nearing the end of our of our time together. What? What are the messages might you have for our audience? What calls to action? Might you have?

 

Ted Connell  

It's, I think these are the most exciting times to be in the industrial sector. If you were an engineer, you probably your whole career been asking for a budget and not got one. To do automation or to do any of these kinds of things. Your board of directors is now setting aside money to do sustainability. To do IoT to do AI the board of directors is giving you a pot of money. So this is new in your career. Be smart benchmark with people that are ahead of you. You know, look, look to peers, go outside your industry, obviously don't go to your competition. Look at the people go talk to Intel. I'm not kidding you. Intel will talk to you very few people compete with Intel and very few people actually buy from Intel. And they they've got a digital story because they had to to build those

 

Jim Frazer  

chips. Right. Incredible story. Incredible. So

 

Ted Connell  

it's great times if you're an engineer and industrial, this is the gold times gold standard.

 

Jim Frazer  

Well, I thank you for joining us. conversation again. My guest was Ted Connell of blue skies ai 10. Before we go, can you share your contact information if someone would like to speak with him?

 

Ted Connell  

Yeah, absolutely. It's T Connell co nn ELL at Blue skies.ai. And my phone number is 480-797-3570. And our website is blue skies.ai.

 

Jim Frazer  

Well, once again, thanks, everybody, for joining us on this edition of Smart City podcast. And thanks again. Take care and our audio engineer Tom Gammon. Thank you. Take care. We'll see you guys again. Cheers.

 

ARC Advisory Introduction  

Broadcasting from Boston, Massachusetts. The Smart Cities podcast is the only podcast dedicated to all things smart cities. The podcast is the creation of aarC advisory group Smart City practice. ARCC advises leading companies, municipalities and governments on technology trends and market dynamics that affect their business and quality of life in their cities. To engage further please like and share our podcast or reach out directly on Twitter at Smart City viewpoints or on our website at WWW arcweb.com backslash industries backslash smart dash cities