On this episode of the AI For All Podcast, Jim Chappell, Global Head of Artificial Intelligence at AVEVA, joins Ryan Chacon and Neil Sahota to discuss industrial AI and scaling AI globally. They talk about AI policies and guidelines, how governments approach AI policy, AI copyright law, AI and blockchain, the impact of regulation on AI adoption, enterprise AI trends, combining AI with other emerging technology, industrial AI use cases, robots and the impact on jobs, figuring out how to use AI, generative design, and implementing AI on a global scale.

About Jim Chappell

Jim Chappell, Global Head of Artificial Intelligence at AVEVA, is a seasoned leader in industrial software with over 25 years of success growing and improving businesses by leveraging cutting-edge capabilities focusing on AI, Advanced Analytics, and IoT / Big Data. Prior to his current position at AVEVA, Jim led the Asset Performance Management (APM) suite of products and related engineering/analytics services for Schneider Electric. He was also a founding partner and managing officer of InStep Software, a global leader in industrial AI-driven Predictive Analytics and Big Data software, which was acquired by Schneider Electric in 2014.

Interested in connecting with Jim? Reach out on LinkedIn!

About AVEVA

AVEVA is a global leader in industrial software, sparking ingenuity to drive responsible use of the world’s resources. The company’s secure industrial cloud platform and applications enable businesses to harness the power of their information and improve collaboration with customers, suppliers, and partners.

Key Questions and Topics from This Episode:

(00:42) Introduction to Jim Chappell and AVEVA
(00:59) AI policies and guidelines
(04:04) How do governments approach AI policy?
(06:33) AI copyright
(09:12) AI and blockchain
(11:17) Impact of regulation on AI adoption
(13:30) Enterprise AI trends
(17:18) Combining AI with other emerging technology
(22:05) Industrial AI use cases
(24:57) Robots and impact on jobs
(31:30) Figuring out how to use AI
(34:55) Generative design
(40:05) AI on a global scale
(43:29) Learn more about AVEVA

Transcript

- [Ryan] Welcome everybody to another episode of the AI For All Podcast. I'm Ryan Chacon. With me today is my co-host, Neil Sahota, Godfather of AI for Good, AI Advisor to the UN, go on and on, talk about his time at Watson or IBM Watson, all this kind of fun stuff that he's done. Neil, it's great to have you.

- [Neil] Hey, yeah, always a pleasure, everybody.

- [Ryan] Also, we have Nikolai, our producer here to ask questions and chime in as needed.

- [Nikolai] Hello.

- [Ryan] So today's episode, we're actually going to cover a topic we have not really touched on yet around industrial AI, considerations for scaling AI globally, government policies, leading use cases, benefits, potential impact AI is having. All important things to consider when really bringing AI into different industries.

So to discuss this, we have Jim Chappell, the Global Head of AI at AVEVA. They are a global leader in smart industrial software. He has over 25 years of success growing and improving businesses by leveraging cutting edge capabilities in AI and advanced analytics. Jim, it's great to have you on the podcast.

- [Jim] Thanks, great to be here.

- [Ryan] So let's get into it.

Let's talk about or start talking about government policy and guidelines around AI. What are you seeing from your perspective?

- [Jim] That's a good question. There's a lot of a lot of buzz these days, certainly around generative AI is the latest thing, and that's probably the one area of most concern. With government policies and things potentially getting out of hand. Overall, AI is a good thing.

And it's been around decades. It's been around since the 1950s. Generative AI has been around since the 1960s. The big change recently is really the amount of data and the large language models that have been trained on it and with ChatGPT and now Google Bard and various things like that.

The hardware is so such that you have the ability to to train that large amount of data. And it's able to do things it never could do before because it has access to everything that's, or most everything that's ever been written and in continues to be that way. You have to balance that with what that, it can do a lot of good and it can not do a lot of bad if in the hands of the wrong people.

And it's like in automobiles. An automobile can do a lot of good or a lot of bad, but there needs to be regulation. There needs to be guardrails around it. And AI is no different. There's concern that especially the latest versions of AI can take jobs from humans or provide fake information, deepfake videos.

It has what's called hallucinations. It will very enthusiastically and with a lot of expertise give you the wrong answer and make you think it's the right answer potentially. It could, you know, there's copyright infringement. It can give you things that potentially could infringe on other copyrights, or it could use your own intellectual property if you use it in asking your questions. And it's all the way to will it eventually be able to annihilate the human race?

I've heard all, if you watch YouTube videos and things like that, it turns into science fiction very quickly. However, as it evolves, it's going to move more from task based to objective driven where it uses, can use anything at its disposal to achieve a goal. So what can it do that's good? Reduce my greenhouse gas emissions. Make my plant more efficient. Burn less fuel. That's certainly good, and it can use multi component solution to come up with a sophisticated way of doing that. But you could also say, do something evil, track this person and eliminate them or bad things.

And you need, just like anything, you need guardrails, you need guidelines, and you need regulations, and this is what is starting to happen. You see it from the White House. You see it from European Union. You see it all around the world where these things are starting to be addressed.

- [Ryan] Neil, let me pass this over to you since with the work that you do with the UN, I know a lot of what you talk to them about is how to structure their initiatives, policy, things like that. How, if you were to explain to somebody at a high level how organizations like the UN, how other governments approach something like this when it comes to policy regulation, how would you explain that process? Like how are they picking who they're bringing in? Who's advising on this because a lot of the people and the decision makers in the government might not be technical individuals. How are we, how can we be confident or feel more comfortable that the right people are there to help guide us with things that can sometimes be quite uncertain?

- [Neil] That's an interesting question because historically, that's not how regulators actually worked. One of the big things we've done at the UN is actually try and build essentially an ecosystem. So rather than just have the traditional government regulators react to things that happen, actually build essentially a partnership between the regulators, industry professionals, academics, other non profits, basically try and bring in the diverse perspective, understand how some of these technology could be used or misused, and proactively predict what could actually happen.

It's a huge change, like a organizational people type of change that is slowly starting to manifest. And I think we're starting to see some of that as you see especially in the United States with the Senate and President Biden now pulling in we'll call it non traditional people to help formulate regulations or at least be part of that conversation.

But I will call out something that's really key is that while this is all happening, you do still have countries moving forward with their own paths. The U.S. until recently said that if an AI system created IP, like 20% if it was a patent, the AI does not get credit as an inventor, whereas in the European Union, they said, no, the system is credited as an inventor, but there's a reciprocal agreement between the U.S. Patent Office and the European Patent Office. So now that's in conflict. That ultimately wound up being resolved. Now an AI can be credited inventor, but I would be curious, Jim, because you actually brought some of the stuff up with kind of the copyright and all that and that Japan just recently stated, is now put into law, that the right of use for training, you can use copyright material, people's paintings for training purposes. The Japanese government said that's perfectly legal for AI. Any kind of IP is fair game for training purposes. I would just love to hear if you have any thoughts about that.

- [Jim] That's going to be a challenging one. And I've been working with legal internally as well. There, there's a lot of certainly lawsuits and new legal policy that's going to be made in the upcoming months and years. Ongoing litigation that's going to, I think, evolve both nation by nation, as you said, as well as internationally.

And the positive side of what Japan did, it makes it easy. It gives a blanket statement. The negative side is anything that can be used for training then could be used with a competitor to potentially create things that have been patented and compete and violate a patent legally.

There's two sides to every coin and it's not an easy thing to answer. And so I think we're going to have to see where the litigation goes and where the judgments come out with this because there's so many variables. I do think when you start putting guardrails in place and start making policy, like you said, some of this will fall out automatically.

One of those things, and I think one of the most important guardrails, is traceability. Because things are getting more complex. It used to be 20, 30 years ago, one of the big things in the industrial world, and it evolved, especially over the last decade, is predictive analytics, machine learning.

And that's one thing that's a one trick pony that's aged, artificial narrow intelligence does one thing, does it incredibly well. And it's relatively easy to trace. You can see how your model came up with an answer, but as you become more sophisticated, what's happening is things are blending together.

You're using multiple types of AI. And of course there are many and with generative, with the amount of vast data that it's used, you need to be able to trace how did it come up with the answer? Another good example is the self driving car. When a car makes a decision and potentially hurts someone, how did it come up with that decision?

And technologies like blockchain. Technologies that are traceable and non modifiable are going to be important because things are not very easy to figure out. It's not, the human mind's not gonna be able to easily determine how I came up with an answer and a solution. And all of that needs to be factored into these government regulations and legal decisions.

- [Neil] Yeah, it's actually a really fascinating point because I've actually been advocating that AI will only build up the trust if it actually taps into blockchain, that there's kind of proof in the training data that was indelible. So if something does go wrong, there's a way to trace back to that.

So I'm fascinated if you think that's going to happen in the near future, or do we have to wait for a few bad things to happen before that's going to be implemented?

- [Jim] I think you're gonna have to wait for legal judgments to come down and enforce it. That's where the guardrails come in. Some, for example, at AVEVA, we're starting to put traceability into our applications already because it does improve the confidence level in it, but of course the other aspect of for confidence in AI is the proofs in the pudding.

Are you getting value out of it? And things like predictive analytics, predictive maintenance for several couple of decades now, they've been getting value out of that. They've been making their operations more efficient. So that's one aspect of trust. The other is when it gets more complicated and potentially there's other liabilities and data from around the world and everywhere it's being used.

That's where you need the more detailed traceability. But I don't know that we can rely on companies just to do it on their own. I think some will because it will help their situation, help their customers. But I think in general, it's going to require guardrails. We have to be pragmatic.

We can't overdo it, but I think there needs to be some, just like with the automobile, there needs to be some level of accountability when using it.

- [Ryan] Let me ask you, when we think about how all this impacts obviously the businesses that are building AI tools, what about the other side of it? The businesses that are adopting tools? How do you see all of this playing into that? Or how do you see the government regulations potentially affecting adoption with these tools? And because I feel like a lot of our discussions in the past have been around the benefits that AI can provide for businesses and trying to figure out how do we not prohibit them from adopting, but I'm sure a lot of companies are waiting to see what's happening with regulations and these guardrails before they decide to potentially adopt AI into their company.

- [Jim] I think that's a multi part thing. One is, they've been adopting AI for a long time now. It's specifically generative AI with the large language model, the GPT or the PaLM 2 with Bard and things like that, leveraging that in their businesses. I do agree that right now it's a bit of a wait and see because of all the legal things.

AI in general, that's not the case because what can AI do for them? Can it help their business, make them more efficient, increase their margins, make them more sustainable. I think those are all the types of objectives that they have in their fiduciary responsibility to their boards, and they're looking at tools, and they've been using tools like AI for quite some time now, especially in the industrial space.

And it's becoming more and more pervasive. But some of the newer aspects, like you said, with LLM, large language models, absolutely. It's going to be a wait and see. And I think that's been the case. If you look at the trend of AI, you get a few early adopters for the last couple of decades since the late nineties, early two thousands, and they've slowly been adopting it since.

And then once the technology becomes mainstream, like machine learning, predictive analytics, then it starts rolling. And then for a lot of industries, if you're not using predictive analytics, you're behind the curve. We see the same thing with, eventually with large language models, generative AI, but guardrails do need to be in place.

So I think it's an evolutionary thing.

- [Ryan] Besides ChatGPT, generative AI, some of the stuff that I've more mainstream, where do you, what other trends are you seeing from an AI perspective when it comes to those tools transforming businesses. So not so much the company's building, but more the from an adoption standpoint, what have you seen AI or where have you seen AI have the biggest impact on businesses recently outside of the stuff that's in the news around ChatGPT and generative.

- [Jim] There's really two schools. One is do it yourself, subscribe to a platform that provides AI and do it yourself. And there's many. Or go with a company that provides a solution and then infuses AI into their software to make it better. You know the companies that go with do it yourself, then they need to have data scientists, and they need to support the application rolled out to their organization and some have been successful doing that.

I think the biggest success has come where companies are infusing AI into their software. They're already a big player in a, in the industrial market, and they're putting AI to make it better. It's a, AI is a science made of lots of technologies, so they'll use things like machine learning, deep learning, reinforcement learning, various types of neural nets into the software to make it better.

And specifically, one of the early big successes was predictive maintenance. It used to be that you would do maintenance either one of two ways. Run it to failure if it was a low value asset and then replace it. You had plenty on the shelf, and it was just easy to do it. The other way was calendar base.

I'm going to lube the bearings and change the filter every three months, whether it needs it or not. I have no idea if it needs it or not. And these could be small assets or multimillion dollar assets. Now with you, there's two things that have evolved. One is they start doing trigger based. The pressure's dropping and the temperature's rising, therefore I need to do maintenance. That's more calculational, and that's a valid way. And then predictive maintenance where you actually use machine learning to find early detection of issues well before any type of control system alert or alarm. And that's what started a couple of decades ago in a big way in the industrial world.

It started in heavy industry, power generation, oil and gas, and now it's become more pervasive across other industries with mining and chemicals and even into CPG food and bev and life sciences. And it's starting to become more and more pervasive. So that's infusing it. Now, a couple of decades later, the bigger trend is more integration of AI. Outside of, like you said, generative and LLMs, integrated AI, where you're taking machine learning and deep, various flavors of AI, putting them together along with other advanced analytics and making it more sophisticated. Because ultimately we're taking incremental steps from artificial narrow intelligence, which are the one trick ponies, computer chess or predictive maintenance and moving more toward human level of thought, AGI, general intelligence.

And we're not there yet. Even with generative AI and LLMs, we're not there yet. But as you integrate multiple types of AI together, you're taking incremental steps to get there. And that's another trend with what we're seeing. Simulation plus AI is another big trend. So a lot of things like that are happening in the industrial space.

- [Neil] You're actually alluding to what we tend to call convergence, Jim. It's not just the different forms of AI being combined together, but all this science and emerging technology, that each element of their own is contributing to exponential growth. But you can actually combine some of these things together to get exponential growth. Like the combination now, you mentioned simulation, the combination of essentially the metaverse, IoT, and AI together.

To actually be able to leverage digital twins and do design, do simulations, actually do testing in things we can't replicate in the real world, like what would happen to this mine equipment and electrical store, or we can simulate a category seven hurricane like NIST is doing and seeing like what kind of building materials and codes would be the best and most hurricane resistant.

There's some serious firepower here when you talk about the good side of AI. What is the holdback, though, right? We're on this weird cusp of this happening, but we haven't quite gone past that inflection point. I'd be curious to hear what you think, Jim, is why there's a little bit of a holdup right now.

- [Jim] I agree with you that this convergence where you bring in things beyond AI and analytics such as the metaverse and the 3D visualization and all of these types of things is a huge area. When I refer to integration, I was referring to more the brains, the intellect, by pulling all this together.

And then when you bring in the metaverse, it started, a lot of prototypes have happened. I think it's slowed down a little bit because it requires customers to have the infrastructure in place and a lot of moving parts all at once. And sometimes those moving parts come from multiple pieces of software that they may not have the budget to subscribe to or pay for at present. It's not one thing. It is a convergence, but it's a convergence with multiple pieces of software, and I think that's part of why there's a holdup. The lower hanging fruit is AI integrated with multiple pieces because that's the brains. That's a, that's an easier thing to fit in one package or connect together.

Whereas if you do the visual, definitely the next step is the metaverse, and that will happen, but it is a little bit slower just because there's more moving pieces, but absolutely agree that's, we're on that path, and we certainly we do a lot of that at AVEVA because we have such a broad portfolio with things like 3D visualization and AR, VR, the whole XR trend, combined with AI and simulation.

- [Neil] Really powerful stuff. Like I said, these businesses slowly moving there, but I agree with you that the cost to build some of these assets, the infrastructure needed, I can see that's a hold up, and I feel like sometimes there's a bit of a believability issue, right? That maybe again, it's the kind of that trust factor and that can we really simulate to some degree doing all these things?

And is that as realistic? We're still hung up a bit in our conceptions of what computers can do from the second generation. AI is the third generation, and I think that's also where a lot of us struggle with understanding that we've got this new toolbox, new capabilities, we don't fully know how to tap into that.

- [Jim] One of the things that's being done, and we certainly do to improve the believability from a business perspective, there's two things. There's consumer believability, and there's business believability. From a business believability and trust, one of the low hanging fruits is calculate my return on investment. You know, I don't want to do AI or some of these technologies as a science project and hope they do and maybe they will but if it's not, if I can't calculate, if I can't quantify my return on my investment, how did I improve my bottom line or my margins?

Then it's not going to go much very quickly, especially the early adopter technologies. And one of the things we've done and put a lot of effort in is coming up with algorithms and ability to calculate the return on investment of an AI endeavor, especially newer technologies. With predictive analytics, predictive maintenance, that's old hat.

They've been calculating that for 20 years, and it's many dozens, if not hundreds of success stories, you know that we alone have but you move into some of these newer areas, it's fuzzier and this goes back to the traceability. I think because there's more than one type of AI to get working together, it's more complicated to determine the return on investment. What is the specific return on investment due to AI versus adding staff or doing something else? It becomes more complex and by having traceability, now you have the data that will feed the return on investment calculations, so that it all plays together a little bit, I think.

- [Ryan] Let me ask you since we've been touching on a lot of different things on the industrial side, what are some of the leading use cases that our audience can be wrapping their heads around or understanding how AI is playing a role in the industrial space?

- [Jim] In the industrial world, one of the low hanging fruits, or you have large, very expensive assets. Large pumps, steam turbines, combustion turbines, condensers, things like that that break. And you want to make sure, when they break, they can cause an outage. And your entire factory, your entire plant shuts down unexpectedly.

So there's damage to the equipment in the sometimes in the millions of dollars. There's extra costs of labor that you have to bring in people often on overtime. Sometimes depending on the industry, the most expensive loss of revenue is the downtime itself. Loss of production capability. You're not producing the goods, or you're not generating power or whatever it might be.

And that's the biggest thing. So you factor all these things in, and you can calculate your return on investment by preventing the downtime. So from an AI perspective, you want early indicators that you're having a problem. And not just from an alarm condition, you want it well before any type of alarm goes off, before anything like that happens, so that you have time to schedule the maintenance and make sure that you don't have this unplanned downtime.

So what happens, you have, it starts with IoT, the Internet of Things, all these sensors, and you're detecting. If you take a pump, a large pump is a simple example, you have flow rate, you have inlet pressure, outlet pressure, you have temperatures, and you might have vibrations, you might have RPMs, spinning, you might have amps if it's an electric pump.

Things like that are all related, and they're used in a multidimensional model, a form of digital twin, that's used in a machine learning model. And what it does is when the real time data comes in, it figures out, are all these things operating the way they should when you've trained it on past data, and it can quickly alert you to, wow, your outlet pressure is lower than normal for not by itself, but for the operating conditions. Or your temperature is too high, and this can give you early detection of a problem. So that's one of the first things that people do, but then they start moving into other things. Vision AI, they start doing scheduling based AI and other dimensions of their industrial operations to improve it, and they can improve production capacity, they can improve, make their labor more efficient. They can make their operations more efficient, so they have less downtime, and they become more sustainable in doing so because sustainability and profitability also go hand in hand.

- [Ryan] Let me ask when we, you're talking about different use cases. I know there's a lot of automations that come out of the work that's done in the industrial space. Obviously, our other brand IoT For All focuses a lot on industrial IoT, talks about how it does work well with collecting the data to bring it into AI which results in companies making changes and automating certain things and now we've talked about robotics coming into the industrial space. How do you think that or those two things together, those industrial automations and robotics might impact jobs, might impact things that like humans are thinking about when it comes to industrial, the industrial industry?

- [Jim] Yeah, I think it's going to shift things. We're in industry 4.0 now, we're moving into industry 5.0 where AI and computers, robots work side by side with humans. A sharing of the well. What it's gonna do is it's going to improve safety because you can put the robots in the less safe roles that humans are often now doing for whether it be inspection or various types of maintenance and things like that.

It's gonna shift human's jobs to more supervisory of an expert roles, looking over the shoulder and managing the robots and moving into new world. Just like the industrial revolution of a century and a half ago. It shifted jobs. I see the same thing happening with robots and AI.

It's going to shift things. And typically what happens, can't speak with 20/20 clarity, but typically what happens is there's a net increase in jobs because of new opportunities and new, it'll be different types of jobs, but new opportunities, new roles for business.

- [Neil] It's interesting you call that out because there was a study done and at the end of 2022, if you take all the tasks human workers do, 30% of it is now actually done by AI or robots. And they actually are showing that by 2025, that number will hit probably about 50%. There's obviously a lot of automation going on, it's not just the blue collar type of jobs, you see a lot of white collar automation like legal research, for example, police report taking. The interesting thing though is that I think they're predicting that 97 million new jobs will be created from AI and robotics technology. The challenge is how do we get the workforce ready for that? How do we retrain those people in the workforce that are retrainable to be ready for that? I think that's the big challenge that we're seeing, Jim.

- [Jim] I do agree. I do agree. And it's like any shift in anything. 30 years ago, 40 years ago, nobody was really going into computers and in a large scale, and now that's shifted dramatically. And I do agree with you, it's gonna, it's always a challenge. You got to, you have to stay current.

You have to, you have to learn. One thing a little bit ironic is the, this whole generative AI, LLMs can help with learning because it makes knowledge and information accessible to anybody with a computer. And now with the low cost of computers, it's easy to learn. Assuming you can read, interact with a computer.

So that potentially could help, but it's the age old problem of a shift in workforce needs to get retrained. I definitely agree.

- [Neil] That's a really interesting idea. I love your idea about people should embrace ChatGPT as like an educational tool, as a way to help you retrain for those, the jobs of tomorrow, so to speak. It's also ironic in that, that's the thing that most people are most afraid of is gonna take their job.

You look at the writer strike going on in Hollywood right now. It happened a few months earlier than expected. In large part is when the writers saw the explosion of ChatGPT, they're thinking, okay, the studios will probably look at AI to help write TV shows and movie scripts.

That's going to happen 10 to 15 years down the road. They saw ChatGPT go from 10,000 to 100 million users in a month, that was one of the triggers for oh my god, this could happen in a couple of months, the rapid pace of change and the volume of impact is so great I think that's what people are unprepared for, and I hate to say it to all the writers, I get what they're talking about here. At the same time, ChatGPT could be a powerful tool they could actually use to do some of their work. So it's a weird dichotomy going on here.

- [Jim] Absolutely. Yeah, there's, ChatGPT is, it's just like a lot of things, can be used for good or not good. And you do have to be careful what you put into it. But if you're asking generic questions, it becomes the new form of Google or the new form of search engine and a lot of the challenges that people are concerned about with ChatGPT have existed for years with search engines because whatever you put into a search engine has the potential for getting trained in anyway. So it's just amplified, it's not necessarily created new problems in some cases.

But yeah, it's a powerful tool and a powerful engine that can be used for a lot of different uses, and we're just barely scratching the surface today because I mentioned objective driven. We're looking at training, that's one thing, or doing this specific job is one thing. But if you give it a grayer area like reduce the emissions out of my plant, there's, it could involve emailing.

It could involve accessing data. It could involve changing set points on a control system. It can involve rescheduling things and scheduling maintenance and so forth. Lots of different dimensions if it has it all and it's all software driven of what I mentioned. If it has access to all of that, it could really take control. And if it does it well, everyone's going to win because we become more, the company becomes more profitable.

They are more sustainable. They're reducing less carbon to the atmosphere. But if it's not done well, or it can shift, it could cause disastrous effects, and you need to be able to know A, how it came to its conclusion and exactly what it did. And B, you need those guardrails in place within a pragmatic, you can't go overboard, but there needs to be some level.

- [Neil] I was thinking about it because it's a tough, it's a tough question, right? Because Jim, I think has given a lot of great advice and at least a lot of the companies that I talked to, their big struggle is they get the power, they feel like they should be doing something, they just don't know what it is or how to figure out what it is.

You've given some fantastic examples, Jim, in the industrial sector. Maybe that's the right question to ask is how did people, how did the businesses think of these ideas?

- [Jim] In many areas, they've been doing it for a long time, and it's well accepted. Predictive maintenance being a good example. But in the industrial world, AI is going into engineering design. It's moving, we're not at generative design yet, but we're moving there and pieces of it, automated pipe routing and white space optimization and piping and instrumentation diagrams and things like that are happening now.

They're doing laser scans and then we're, systems are being created to identify what's what in a laser scan, not only the type of asset, but the specific asset. And mapping it against other data. Then you move into operations and smart control systems. And then you move into maintenance, predictive maintenance and really end to end, AI is becoming pervasive in the industrial world.

It's just at different levels of maturity. But then you'd look at new technologies, and this really applies to what you said, Neil, is with generative AI and LLMs are very hesitant because how do you start with that? Maybe they started with machine learning, but how do they move to the next level with that?

I think what's going to happen is they're going to use it against their own data because the big concern with generative AI and large language models is it uses public data and anything you ask it goes on to this public database. But what if you go after a private large language model and then used additional AI to instead of ask it general questions, use it as a means of contextualization of your data, so you can automatically set up your data hierarchies and data relationships.

So it becomes more of an inferencing engine instead of a general Q&A, question and answer type engine, so it can be used in a lot of different ways that becomes a lot with those, these are some of the guardrails I'm talking about that make it private. The other problem with large language models today is they're only as good as the date that they were trained.

And with the case of GPT, it's September 2021. It knows nothing about that beyond it. However, if you leverage AI, and companies are starting to do this, Google, Microsoft, you can extend it to new data. With the industrial space, you can use a neural net, use the large language model as a means of inferencing and contextualization, and then blend in your own data, the company data, on a private way, method.

And so things like that, I think, are going to get more adoption because they're not posting company secret type information or data to the large language model at large and keeping things private. And so leveraging it for what it's good at but not wholesale is I think the next step in adoption for that particular technology.

- [Nikolai] Something you mentioned was like generative design. So it makes me think about government policies. Because if you imagine AI starts designing airplanes, for instance, well now the FAA has to become kind of an AI sort of regulation board too. Or you might have a bigger AI board that sort of gets involved in all these other government departments when they're needed.

- [Jim] From a design perspective, and you mentioned what if you had an AI you know designing an aircraft, this is an example where new human jobs will be created because there'll be more requirements for supervision of the design. There's certainly design supervision today, but it's based on professional engineers, PEs doing the design of the aircraft and doing it and but guess what? They're leveraging past designs and improving on them. That's exactly what AI does. It splits apart old designs using patterns and other knowledge to create new designs. So it works much like a human, but it's going to require human supervision. I don't see a closed loop where AI designs the aircraft, and then robots build it using AI as their brain, and then it gets QA'd and tested using AI, and there's no human in the loop.

So this is where new jobs are going to be created. These new, the human in the loop is still going to be essential, but it'd be a different type of role, supervisory. Whether there's new government regulations or not, I see the process, there are already government regulations and those government regulations are still going to be imposed on the humans signing off on the design.

The AI is not going to sign off, they're not a professional engineer stamping that they approved the design, it's going to be a human doing that. It's just the laborious work, and a lot of it is laborious, will be done by the computer. So that you can do things faster, more consistently, but that's where you go back to that traceability that I mentioned earlier.

You need to know if there's any questions, I want to know where AI came up with this design decision or this operational decision, for that matter.

- [Neil] It's an interesting example because seven years ago, I actually worked with Local Motors on using AI and generative design for urban transportation and recognizing that we as humans have some biases about like how a bus might look and that kind of stuff. We didn't want to impose those types of parameters on the AI.

We basically just said, look, we're looking for an urban transportation vehicle that can carry 12 people and uses renewable energy, right? And we taught it things about transportation, that kind of stuff, and the AI generated millions upon millions of design options based on patterns, other stuff.

But one of the interesting things is it was fabricating materials we actually had never made before but actually could make trying to figure this out. And then it whittled it down to its top seven design choices, which then human engineers actually reviewed to see are these viable and if they are which would be the most viable option.

And there was one that was chosen that became Olli, and it was actually a solar powered 3D printable self driving bus. If you look at it, it doesn't really look like what we think of as a traditional bus. It can hold up to 12 people, works well in urban transportation. And it was actually built. It was very cheap.

It was actually in use for a little bit until the COVID pandemic started and everything shut down. But that was a really interesting kind of experiment to see this kind of mesh of hybrid intelligence of pairing up some of these AI machine capabilities with human capabilities and take the best of both to come up with essentially a much better design than we had seen before.

- [Jim] Completely agree. And one of the things that AI is really good at is thinking outside the box and the more parameters you give it to use, sometimes you'll come up with wild designs that you never thought about and wow, maybe that makes sense. And it's interesting you mentioned provide the top seven or so designs and let the human decide and it becomes hybrid.

This really plays into industry 5.0 and matter of fact, we're doing that also with scheduling. We have a schedule AI, for example, and what it does is it optimize the, when you have periods of downtime, there's a lot of activity, maintenance activity and various things and logistics and supply and a lot of coordination.

And these are very complex schedules. Our system comes up with the top several schedules and a human makes the final call as to which schedule would be optimized. And yeah, so more and more working alongside, like you brought up before, working alongside AI solutions, whether they be robots or just a computer is going to happen and it's going to turn human role differently, but new. And so it's definitely an exciting time.

- [Ryan] Before we wrap up here, Jim, I do want to ask something that we did mention in our intro, and I want to be sure we touch on it at least a little bit here is around implementing AI kind of at a global scale, as opposed to just bringing in a tool, adopting it for an organization. We're talking about more mass adoption, implementing at large in across potentially different countries, different industries. What's important to know about how to approach implementing AI at on a global scale? What are things that we should be thinking about considering or what things stand out to you as important when it comes to that global approach to AI?

- [Jim] One is you need the trust, right? You need to prove the return on investment. Once the company is comfortable with that it is the right tool for them, then they can roll it out. Some of the challenges when you roll out AI on a global scale especially, multi country, multi continent, is starting with the data because AI lives and breathes by data, and there's a lot of data domicile rules already in place today that data cannot leave European Union or this and that. All kind of data rules.

And so now the question is can the model that was trained on AI leave the Union, or do you need to train it separately? Oftentimes, you have to train the models separately based on the region that they're in, even though you might be able to use the same system. Now software as a service in the cloud, this helps with global rollouts.

But even if you look at Microsoft, Azure, Amazon, different ones, they have different data centers in different regions, in Australia, in Asia, in Europe and North America and so forth with obviously redundancy. So you pick a macro region and you roll out one system there, and you pick another region, and you roll it out there because of these data.

It's more data driven than AI driven, I think. And then, but yeah, AI systems, once the data is there, once it has access to the data, you need a system that has processes because AI algorithms are great, but it becomes free for all and the wild west, if you will, if you just have an algorithm without the processes around it to take advantage of it in a consistent manner.

And you also need the ability to communicate to get to the right people because you could be a data scientist, and if you know the answer, but the maintenance people out in the field a thousand miles away don't know it, it doesn't do you any good. So you need to get the communication to the right people at the right time.

So starts with data, moves to a global system, software system that has the processes and the communication necessary to get the information to the right people to take advantage of it, to fix that pump or to turn something on or off or to whatever it might be, and then roll up the results to not only managers but executive management on a global scale so that they can see the combined results and then the ability to drill down to the details. So all of those pieces working together create a successful AI global system.

- [Ryan] Neil, anything to add to that from your perspective?

- [Neil] No, I think that's actually probably a good summation and a good way to unfortunately end this conversation. I feel like we could talk for a couple more hours, but all good things must come to an end.

- [Ryan] Yeah, absolutely. Jim, thank you so much for your time. It's been a great conversation. We talked about some topics we haven't covered yet so appreciate it. And just to give our audience some information from your side, what's the best way that if they want to follow up on this conversation, learn more about what's going on at AVEVA, just and touch base in any way, but how do you recommend they do that?

- [Jim] Surely go to www.aveva.com and there's plenty of information, contact information, and I would love to we'll get connected with the right people. And if they have any questions or want more information, it's all on the website, and I look forward to further conversations.