In The Arena by TechArena - Real-Time Supply Chain Control with Oii

Episode Date: January 19, 2023

TechArena host Allyson Klein chats with Oii CEO Bob Rogers about his team’s AI-based solution for real-time modeling and management of supply chains and the opportunity for AI to drive actionable solutions for business and society.

Welcome to the Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome to the Tech Arena. My name is Alison Klein, and I am so delighted to have a friend and former colleague, Bob Rogers, in the studio with me. Bob, you are CEO of OII, which is the topic today, but also a published author and astrophysicist by training. Incredible background. Welcome to the program.

Thank you. I'm really excited to be here. So let's just get started. You started as an astrophysicist developing digital twins of black holes in a galaxy far, far away a long time ago. Tell me about how this part of your trajectory first introduced AI to you and how that has been influential in shaping the trajectory of your career. So the big features of working on digital twins of supermassive black holes in other galaxies. Really, the most important thing was it introduced me to the power of technical computing and simulation starting in the late 80s, early 90s, when our access to hardware was limited, but the potential was starting to grow and our ability to simulate was really there. That led to, when I was a postdoc, a related interest in this really cool new thing that was computing the way brains compute.

In other words, artificial neural networks. And so I got interested in artificial neural networks, did some academic research with a colleague at Davis, and he and I ended up writing a book called Artificial Neural Networks Forecasting Time Series. And so it would be easy to imagine that that led to me sort of launching into a serial AI specific career. But actually what happened was people started calling me and asking me, Hey, I saw your book. Can you forecast the stock market? And so I was a physics kid, so I didn't really know that much about the stock market, but being a researcher, I researched it

and it turned out that I could forecast the stock market. So I ended up having a 12-year, not exactly detour, but formative foundational period co-founding and running a quantitative futures hedge fund, which was an adventure in its own right. But what that did was it allowed me to continue to develop advanced models for forecasting, for modeling. Fast forward a couple more years, I co-founded a company called Apixio, which is a healthcare AI company. My experience with modeling and simulation played a role there. And then I was lucky enough to come to Intel, where I was chief data scientist for analytics

and AI and worked with you. And so that kind of got me there. And of course, now the simulation and the digital twin is central to the OII story. So it really does come all the way back to the beginning for me. Now, as chief data scientist at Intel, you had an opportunity to work with companies all over the world who were trying to grapple with a transition of looking at data as something to manage and store to something to abstract value and monetize. And it was at the dawn of the AI era. That kind of led you to writing a book about demystification of AI. And I think that one of the questions that I have for you is, when you look at where we're at with AI, and we're going to get to OII in a second, where do you think enterprises are in terms of their adoption of the technology?

And then more broadly, where are we in terms of the core capabilities of where you think AI will go? Yeah, it's a great pair of questions. In terms of where we are, I think if you had asked me even two months ago, I would have said, you know, there's a continuing desire to put data first and to build, you know, a lot of people call it digital transformation, right? This idea that you're making data-driven decisions, that data is a first class part of your overall operations. But I think the reality is most organizations were still struggling with that and trying to find point solutions that would solve very specific problems. But they were still a little bit isolated. I can give you an example at UCSF where I was expert in residence for AI. We had a challenge that 1.4 million faxes

come into UCSF each year. There are referrals, there are requests for information. There's all this stuff that sort of centers on these faxes. Well, faxes are, of course, horrible to deal with digitally. And so we built a document processing pipeline using AI that combines computer vision and natural language understanding to interpret what's on those faxes. And remember, a fax is not a digital document. It's a photograph of a digital document. So you have to extract the digital parts from it and then do all this inference to interpret it. And that was a huge success and really opened up UCSF's eyes to what they could do with AI. And that it's not this sort of thing

that sits over all your processes and tells the doctors what to do. It's more like this very practical operational tool that can come in and automate things that are really, really painful for people. But even with that huge success, UCSF is still struggling to then turn that into a sort of eyes wide deployment. So

the challenges are there still in terms of enterprises really, really focusing on data and harnessing AI the way it can be used. The thing that's changed in the last two months is some of this new natural language understanding capability, GGPT, I think is at this very moment catalyzing a transformation in people's thinking about how you would use AI. And it's suddenly going from this place where there's a tool or a module that we've shoved in somewhere to a workflow to AI is going to be like electricity. One of my favorite anecdotes is that around the turn of the previous century, in the late 1800s, early 1900s, most large organizations had a chief electricity officer.

And that sounds ridiculous today, but back then it was this, okay, we know this is going to be foundational, but we don't know exactly how it's going to fit in. And so I think we're right at that moment right now. That's a fascinating little piece of trivia, Bob, that they had a chief electricity officer. I just had a conversation with Brian David Johnson, futurist, and he was talking about how it's not like we haven't seen transformational technologies before but we forget sometimes let's get to the how you're disrupting supply chain in 2019 i would have told you that supply chain was yonder of a topic but my 2023 version of me says, oh my gosh, of course, this is a prioritized area where we need to use AI.

Tell me about OII and tell me about how you've looked at that application of assistance to help supply chain managers make better decisions and have better forecasting of their supply. You know, and it's funny, supply chain is a little bit like electricity in the sense that we don't think about it that much unless it goes off or if it's not working. So the pandemic version of supply chain definitely was a rolling blackouts kind of scenario. And so what we realized, in fact, it's interesting, we were working on the foundations of OII before the pandemic hit, actually. And it arose out of a conversation when I wrote my book, Demystifying AI for the Enterprise, I was elected by my co-authors to write the chapter on supply chain. And so one of the people I reached out was an expert from England named David Evans. And he and I had a fantastic conversation.

I used a lot of his content for the book, but we went beyond that. And he said, you know, Bob, the way supply chain is managed, the tooling really assumes that the world is static. The data that describes the way product flows through distribution networks, the tooling for using a single forecast and then applying that to figure out where inventory should be all of that really assumes that the world is static and of course what happened in the pandemic is we saw in vivid color that the world is not static and i like to equip that even variability is variable and we we know that now. So the core realization that we had is that supply chains need to be able to be operated in the way turn-by-turn GPS works.

That is, as the conditions change, you need to be alerted and told, what are your options? Where do you go with your supply chain? And what I'm talking about is really the specific parameters that determine the design of a supply chain. So where is inventory held? How does it move from place to place? How is it replenished? What are the other characteristics of that distribution network that moves products from

manufacturing out to customers. And what is prevalent today, and I mean like 99.99% of cases, is that supply chains are set up with a few parameters that control those attributes of the distribution network. And they're set by humans. So they're not particularly optimized when they start, but then they're just left for years at a time. I mean, it's typical for companies to not review their supply chain designs more frequently than once every two years. So it really leads to, and especially in a time like this, where the lead times and the availability of product and the characteristics of demand, as all of those things are varying all over the place, this fixed set of parameters is failing miserably. And so our approach has been

to build a digital twin. We have a way to automatically build a digital twin from the data that's in the ERP, in the supply chain management software. We build the digital twin that allows us to then find out how the supply chain is going to operate under any set of circumstances. Then we use AI to predict what are the circumstances that the supply chain needs to be prepared for. Is it major disruption in certain channels? Is it just drift in various lead times and network performances? Or is it something else? And we can simulate, we can basically create that set of future scenarios, run it through the digital twin, and then see how the supply chain is going to act. That points out the risks in the supply chain. So risk of running out of stock, risk of having

inventory in the wrong place, too much, too little, having product expire while it's in the warehouses, all these factors. And actually, interestingly, also the impact of the supply chain on the carbon output of the entire supply chain, That is choices around transportation and storage that impact how much carbon is generated by the supply chain. Anyway, all these things can be computed and then we can optimize for the objectives of the organization so that they can have the supply chain that achieves their goals in the most cost-effective performant way and ensure that that supply chain is operating that way all the time. So the critical thing about the automation is that

this is not a one-off calculation where you set some parameters. It is every time information changes in a meaningful way, you recompute, you re-optimize, and you make recommendations about where to go next. So it really is like turn-by-turn navigation. So to take that into an example, if an auto manufacturer was using this in 2021, your model might have already identified processors, silicon, as a key thing to be looking for. And then as the supply chain changed in real time, it would be optimizing and making recommendations about how to navigate that acute situation.

Exactly. And responding very, very quickly, right? Because with global awareness of how the whole set of supply chains interconnect, because you don't have just one product. You have many or tens of thousands of products that are all sharing and fighting for resources. So in crisis, it's almost impossible for humans to react quickly enough. And a supply chain leader told me not too long ago, whoever responds fastest wins. So when these crises happen, you've got to already have some resiliency built in, but you've got to have the ability to reconfigure your supply chain quickly. And by gaming it out, you've already come to those decision points, right? You've come to the decision points in simulation. You've already faced that decision and know how you should be deciding. That's what's really thrilling about this is that it just gives a company the tools to be very nimble and to be savvy to the market.

Tell me more about the carbon footprint part and how do you look at that and how did you model carbon utilization across all of those elements? Because that's a very complex model. Yeah, the starting point is transportation and how are you transporting goods? How far are you transporting goods? So one of the questions that has come up for our customers has been, well, okay, if I source a product more locally, but the reliability of that supplier is different. So maybe you have a long lead time for a supplier that's far away, but they're like, just absolutely like clockwork. You've got a supplier that's far away, but they're just absolutely like clockwork.

You've got another supplier who, when you're not thinking about sustainability, you might not use simply because their performance characteristics are potentially not as good. Well, the trade-offs in your supply chain can calculate how much carbon that each of those replenishment strategies would be. There's longer transportation, but maybe you can go by ship instead of by truck. And the shape of the supply chain changes because you're bringing something in from closer with maybe a little less reliability, but a shorter lead time. So you can calculate exactly what the best scenario is for both and find out exactly what are the trade-offs in terms of the carbon output. And so it can go down to as granular

as vendors, although typically it's choices between different kinds of truck, rail, air, ship, and those have big impacts on how much carbon you output the next horizon in that is tying that a little bit more to the factory operations because a product can be replenished every week it could be replenished every 12 months and the impacts on what pressure you put on the factory also impacts carbon. But right now, the primary thing that we're calculating is the trade-offs with different types of transportation because that's the dominant and most manageable piece of the question. And we actually had a customer who said, OK, I care about my impact on the planet or my company's impact. I want you to double the carbon costs for everything in my supply chain so that when

I optimize, I'm actually finding the solution that is really taking carbon into account. It's my lowest cost, highest efficiency solution, given that I'm doubling the impact of carbon. So it's really nice because you have this knob that you can control about how you allocate the various prioritizations. Now, you and I have talked before about where companies are about utilizing their data to actually drive their business decisions. And we've had those conversations in broader context. But what is your perspective now that you've been talking to folks about what you're delivering with OII?

What is your broader perspective on where the state of the supply chain management is today, how it's changed over the last two years, and how folks are ready to adopt what you're delivering? Yeah, so it's a great series of questions. Today, the most salient data point that I can give you is that the most common supply chain software in the world is Excel. And I'm not disparaging Excel, but it's not an enterprise-wide, highly automatable, robust platform for doing operational activities, right? It's more like handcrafted macros and updating things when you get around to it. And so there's still a long way to go, but I'm hearing every day now customers talking about how they are wanting to digitize their supply chain, how they're aware of the fact that they need to be ensuring that their data is good enough to use for making decisions.

And, you know, again, that first class citizen role of data comes into play. And actually, one of the big learnings that I had when I was at Intel is that quite often it's good advice for even large enterprises to start looking at data on a use case by use case basis to begin with. These gigantic data transformation projects quite often end up with very high costs and no immediate ROI. So you finish a big project and the CEO says, well, so the project costs $10 million. What's my ROI? And the answer is, well, negative $10 million. And that doesn't sound right, even if it's setting a foundation for something bigger in the future. You may run out of runway by the time you've done that. So we're definitely seeing folks tackling data transformation at the same time that they're bringing in OII. And so one of

the nice aspects of that is as we ingest the data, first of all, we just check to make sure that all the data is consistent, that there's no missing data, that there's nothing that doesn't look right to us from a sort of analytics point of view. And this is a series of, actually, we have some AI running on that as well, just to look at the data and ensure that it's coming in properly and that it makes sense. That is really effectively shining a light on the data and pointing out the areas where they may have some challenges internally that they weren't even aware of. And so we point out where the data might have some challenges coming in. And then as we're monitoring and updating the design of the supply chain, we then start

to see where the underlying network capabilities and also demand behavior are evolving. And that point, that can really, again, focus the energy of companies on cleaning up the processes that are most critical. And so it's not exactly a data thing, but it's based on the data and what we're seeing. We can then point them to, okay, here's a part of your supply chain where variability is really high. And what you didn't realize is variability is going to cost you a fortune in terms of being able to deliver product in a timely way to your customer. And what happens is if you want to overcome that, you have to have huge inventories. And if your product can expire, then your huge inventories might mean that you've got almost no shelf life left when the product hits the customer. So the optimization

and the monitoring of that data actually is very helpful to sort of create that core data transformation that then can nucleate a much broader process, but with an initially a very high ROI so they can get some good value right away. So there's a lot of interest right now in just getting supply chain more manageable because as you said, supply chain has shown up in the headlines and supply chain leader said at a conference the other day, I guess we should be careful what we wish for. Everyone cares about what we do now, but it's not a good thing. Well, it sounds like a fascinating solution and one that can really

change how supply chain is managed to very good effect. I want to go back to what you were talking about with chat GPT. You know, one of the things that I've read about is this concept of narrow applications of AI and then artificial general intelligence. And what you're describing for OII, while complex, seems like a narrow application of AI. Absolutely. Absolutely. I think that what got everybody excited about chatPT is maybe this is the first steps towards artificial general intelligence. And I wanted to ask you about that. And if that's something that you've thought about, do you think that's true?

Or is there something else in play that is the reason why people are so excited about what's been done here? Yeah, this is the beginning. This is the first time an AI system has actually been good enough to make us think, wow, this feels like what an AGI would be like, right? You ask chat GPT a question and it gives you a very sensible answer that is both generally pretty correct and also the response is structured in a fairly sophisticated way. And it's like, wow, if this was on the other end of a phone, I'm not a hundred percent sure I'd know that this wasn't a human. So I think it's really opened up people's eyes about this question of going from, as you said,

point solutions to this more general capability. And it's like, oh, wow, what does that mean? For myself, having written a couple of books on AI, I thought, well, so is my role as an author kind of over with now? AI can just write the books. And so I actually set out to have ChatGPT write a book about the history of creativity and communication and how's it been influenced by technology as we've gone from cave paintings all the way up to AI. So I actually, without putting in any of my own content, prompted ChatGPT to create this book. And then I used text mostly from the document and a little bit of my own input to generate illustrations using a variety of text to image algorithms and illustrate the book. And then at the end, I talk about the process I did and where things worked well, where

things didn't work well. And it sort of ended up being this really nice journey about the story about the history of creativity and communication was interesting. Actually, chat GPT taught me a couple of things i didn't know about so that was fun but the process the meta for the whole thing was the technology still has some shortcomings i think allison you commented this before we we got on the air that you know it's not quite ready to replace us as writers or as content generators. And it lacks perspective. As you said, it, it lacks sort of flavor in some ways, but it's actual surprisingly close

to being able to generate significant amounts of coherent content. So anyway, I added some content to just talk about my observations about the process. It also becomes a sort of how-to for people who want to use these kinds of technologies to generate visual and textual content. And so it ended up being a great process. And so my wife, Teresa Hart, and I actually put the whole thing together as a book. She's an attorney. So she talked about and asked ChatGPT about the ethical considerations and intellectual

property protections around these kinds of technologies. And we're actually going to publish it on Amazon. So it's an experiment that turned out at least interesting. And so I'm curious what people will learn as they kind of go through that journey. I can't wait to read it. That's fascinating. I just had it write a blog for me, which you can read about on the tech arena, but I didn't go quite as far as you did.

I want to read your book and see what, what it's come up with. And I'm sure the blend of chat GPT and you. When you look at that and your experience, where are you on, is this the dawn of artificial general intelligence? Yeah, I think it is. There's probably some pieces missing. There's sort of a circling back to check information and to sort of be able to create a perspective. One of the things that I found while I was writing it was I really wanted

it to make some inferences about what it was saying. We've gone through this journey in this chapter. We've talked about how we've gone from the telegraph all the way up through TV, and we're close to the dawn of the internet. And I want it to not just summarize the facts, but to give me some inferences and some takeaways. And that is still missing. I think AGI probably needs to have some of that. But again, given how much of a leap ChatGPT is from GPT-3, just by doing sort of additional training. So I think what they did is they did reinforcement learning to give it some guidelines on how, what's good, sort of non-harmful, but also coherent and interesting, longer text. And they trained it with reinforcement learning. It's not hard to imagine that next leap

being there pretty quickly. So I think it's an important step. It's certainly opening up our imaginations to what living with AGI might look like. Exciting times. Well, Bob, thank you for being on the program today. I know that we're going to have you on again to talk about your

other endeavors in AI, but this was a wonderful introduction to OII. One final question for you. If folks want to engage with you and your team and talk about supply chain and how they can leverage OII, where should you send them to engage and learn more? Yeah. So we have a website, oii.ai. That's a great place to start. And there's a form you can connect. They can connect directly with me, bob at oii.ai. We also have a LinkedIn page and I'm also on Twitter at scientist Bob. So would love to hear from people and help them run their supply chains with full control

and variable speed rather than just puttering along at one speed the entire time as the world throws lots of obstacles at you. Well, I can't wait to see the book. I can't wait to see the trajectory of OII. And we will catch up in the months ahead about the broader field of artificial intelligence. Thanks for being on the show. Wonderful. I really, really enjoyed being here.

Thanks, Alison. Thanks for joining the Tech Arena. Subscribe and engage at our website, thetecharena.net. All content is copyright by the Tech arena.