Life Science Labs Can’t Be Automated, But They Can Be Orchestrated
Wet labs at life science companies look and work the same pretty much everywhere. They’re full of incubators, refrigerators, centrifuges, liquid handlers, gene sequencers, DNA and RNA synthesizers, and all sorts of other complex equipment. And a lot of these machines are automated—but the larger workflow in a life sciences R&D lab is very much not automated. And that’s a problem, because if you’re trying to collect evidence for a scientific paper or a regulatory filing or trying to manufacture a product that’s verifiably safe, you need to make sure that the same procedure gets carried out exactly the same way every time. Our guest this week, Artificial CEO David Fuller, believes that life sciences labs will always revolve around manual labor, but thinks there’s a way to orchestrate the process more precisely. Artificial makes software that allows lab managers to create what he calls a digital twin of their entire laboratory, where data structures track what’s happening with each piece of lab equipment and keep them in sync, providing what Fuller calls “a single pane of glass that makes it easier to see the state of the equipment and the science as it’s running in your lab.” Humans will always stay in the loop, but Fuller says the benefit for companies who orchestrate their labs in this way is that the data and the products coming out of the lab will be more consistent—which will be even more important as laboratories start to act more like factories, where a lot of the actual production of biologic drugs or other materials happens.
Wet labs at life science companies look and work the same pretty much everywhere. They’re full of incubators, refrigerators, centrifuges, liquid handlers, gene sequencers, DNA and RNA synthesizers, and all sorts of other complex equipment. And a lot of these machines are automated—but the larger workflow in a life sciences R&D lab is very much not automated. For the most part it’s individual researchers who decide how and when to use each piece of equipment, and individuals who move samples and materials back and forth between the machines. And that’s a problem, because if you’re trying to collect evidence for a scientific paper or a regulatory filing or trying to manufacture a product that’s verifiably safe, you need to make sure that the same procedure gets carried out exactly the same way every time.
Our guest this week, Artificial CEO David Fuller, believes that life sciences labs will always revolve around manual labor, but thinks there’s a way to orchestrate the process more precisely. Artificial makes software that allows lab managers to create what he calls a digital twin of their entire laboratory. Inside this digital twin, data structures track what’s happening with each piece of lab equipment and keep them in sync, even if they’re from different manufacturers. The software provides what Fuller calls “a single pane of glass that makes it easier to see the state of the equipment and the science as it’s running in your lab”…meaning what’s happening, why it’s happening, and what errors may be cropping up. Humans will always stay in the loop, but Fuller says the benefit for companies who orchestrate their labs in this way is that the data and the products coming out of the lab will be more consistent. Which will be even more important as laboratories start to act more like factories, where a lot of the actual production of biologic drugs or other materials happens.
For a full transcript of this episode, please visit our show page at http://www.glorikian.com/podcast
Please rate and review The Harry Glorikian Show on Apple Podcasts! Here’s how to do that from an iPhone, iPad, or iPod touch:
1. Open the Podcasts app on your iPhone, iPad, or Mac.
2. Navigate to The Harry Glorikian Show podcast. You can find it by searching for it or selecting it from your library. Just note that you’ll have to go to the series page which shows all the episodes, not just the page for a single episode.
3. Scroll down to find the subhead titled “Ratings & Reviews.”
4. Under one of the highlighted reviews, select “Write a Review.”
5. Next, select a star rating at the top — you have the option of choosing between one and five stars.
6. Using the text box at the top, write a title for your review. Then, in the lower text box, write your review. Your review can be up to 300 words long.
7. Once you’ve finished, select “Send” or “Save” in the top-right corner.
8. If you’ve never left a podcast review before, enter a nickname. Your nickname will be displayed next to any reviews you leave from here on out.
9. After selecting a nickname, tap OK. Your review may not be immediately visible.
That’s it! Thanks so much.
Harry Glorikian: Hello. I’m Harry Glorikian, and this is The Harry Glorikian Show, where we explore how technology is changing everything we know about healthcare.
Most of the new medicines and treatments that are beating back disease, from antibody therapies to vaccines, were invented and tested in wet labs that look the same and work the same pretty much everywhere.
These labs are full of incubators and refrigerators and centrifuges and liquid handlers and gene sequencers and DNA and RNA synthesizers and all sorts of other complex equipment.
And a lot of this equipment is automated. But the larger workflow in a life sciences R&D lab is very much not automated.
For the most part it’s individual researchers who decide how and when to use each piece of equipment, and individuals who move samples and materials back and forth between the machines.
An investor friend of mine named Kouki Harasaki, at Microsoft’s M12 venture fund, commented that even inside the pharma giants, labs are more like test kitchens than factories.
And why is that a problem? Well, because sometimes, if you’re trying to collect evidence for a scientific paper or a regulatory filing or trying to manufacture a product that’s verifiably safe, you need to make sure that the same procedure gets carried out exactly the same way every time.
This week on the show I have a guest who believes that life sciences labs will always revolve around manual labor, but thinks there’s a way to orchestrate the process more precisely.
His name is David Fuller and he’s the CEO of a company called Artificial.
Artificial makes software that allows lab managers to create what he calls a digital twin of their entire laboratory.
Inside this digital twin, data structures track what’s happening with each piece of lab equipment and keep them in sync, even if they’re from different manufacturers.
The software provides what Fuller calls “a single pane of glass that makes it easier to see the state of the equipment and the science as it’s running in your lab”…meaning what’s happening, why it’s happening, and what errors may be cropping up.
Humans will always stay in the loop, but Fuller says the benefit for companies who orchestrate their labs in this way is that the data and the products coming out of the lab will be more consistent.
Research programs can move to completion faster, more affordably, and more repeatably.
And that will be even more important as laboratories start to act more like factories, where a lot of the actual production of biologic drugs or other materials happens.
I had a wide-ranging and mind-expanding conversation with David about the role of automation and AI in the modern life sciences lab, and I want to play it for you now.
Harry Glorikian: David, welcome to the show.
David Fuller: Yeah. Thanks, Harry. Thanks for having me. I appreciate it.
Harry Glorikian: So, you know, I was so pleased, like, you know, learning about the company, seeing what you guys are doing, digging into it. But but before we get into the company, I’d love to talk to you about your background. You know, I do this with a lot of my guests because it’s like, how did you get here? But, you know, because you don’t come out of the life sciences industry, your experience has been in sort of advanced robotics, industrial robotics, to be specific. I think it was named, a company called Kuka, which was a Chinese owned, Germany based robotic maker for supplying car companies with assembly line robots. Just to put it into perspective for everybody that’s listening, like how did you go from an industrial robot to biopharma and then how did you end up founding Artificial, right? Which, you know, what made you think of sort of software and life sciences.
David Fuller: Yeah, absolutely. I appreciate that. Yeah, it’s an origin story that is fairly indirectly arriving in the life sciences space. That is for, so, maybe a little bit on my background and I’ll try to do a thumbnail and not belabor the points, but I’ve been in love with programming computers since I was a kid. I got a degree here in Texas from Texas A&M in computer engineering and then immediately started at age 18 for an Austin based company called National Instruments, rebranded as NI, where I work from age 18 to 43. So it wasn’t exactly startup entrepreneur oriented. I wasn’t exactly life sciences focused, and that was a great place to learn how to be a solid engineer. They make tools for scientists and engineers to solve super complex problem that are combination of software and hardware. And I focused on the software side. So those were my sort of days of an engineer programming making tools for scientists. One sort of claim to fame for National Instruments, from a commercial point of view, that’s more broadly known because they’re fairly niche in the test and measurement space. They’re still about a $1,000,000,000 company, but it’s still somewhat esoteric. They partnered with Lego Mindstorms to build the robotic software for the Lego robots. To me, this was an amazing sort of realization that here we are working on this fairly complex, fairly market, independent technology that can be applied to things like the CERN Hadron Collider. So they control the beams that shoot back and forth to make sure it doesn’t run into the wall.
David Fuller: It’s the foundation of SpaceX ground control, but at the same time, learning from Lego, we were able to say, okay, can we take this really complex concept, narrow it down and apply it in a domain where lots of people can use it, including children. So this was, for me, sort of an epiphany. It’s not that there isn’t amazing technology that solves a lot of problems. It’s that a technology is not accessible, usable or understandable by the people that really need it as a tool. So this is one key pillar and learning that fed into how I ended up at Artificial, working on tools for scientists in the lab automation space.
David Fuller: So around nine years ago now, I met the CEO, the then CEO of the Kuka Group. And so at the time, Kuka was a fully German owned industrial robotic company. If you see the car commercials showing Detroit or all of these other things, it’s the big giant orange robots throwing around hunks of steel. So I went to work on that for them, really focused on how to apply modern Internet based technology, the best that it has to offer, plus things emerging in the AI space. And could we take that and shake it up and apply it to disrupt industrial robots? So I moved to Germany, became the CTO of their group, looked at logistics and super high throughput automation. But I kept coming into this, this duality of two key sets of problems that again, feed into Artificial.
David Fuller: And so one is what I’ll just call digitization. I hate to use the term, but, you know, it broadly means everything to anybody that’s using a computer. But for me, modern digitization is basically, have you taken the fundamental infrastructure and front end technology of the Internet, which is the most amazing at-scale, fully integrated cyber physical system on the planet, and have you applied it to your problem today? And if you haven’t, you haven’t digitized. I mean, that’s basic. You can have old school desktop stuff and you can have a ’90s-era tech. But if you haven’t done that, you haven’t digitized. So. So digitization, even though it’s a hyped buzzword, is real, which is have you modernized our digital culture? The other one is just automation. And automation is a more niche, less scale, complex domain. It deals with the boundary of the real world and the digital world, and it’s messy. And so you end up with, in robotics, in the industry and automotive and electronics and warehouse logistics, digital interfacing with physical. Complex. So I did that for a bunch of years trying to hyper optimize how fast you can make a car body come off a production line. So I did a lot of screening of AI companies and this ultimately gets to how we ended up starting a company and focusing on life sciences. I love the pillars of digitization and automation. I love the ease of use and democratization of complex capabilities for people. In this case, I was really looking for a market that was both well capitalized, so I need to have something substantial size, but one that I actually could get behind and believe in and motivated me on some fundamental level.
David Fuller: I’m getting older, I want to work on things that motivate me. And so for me, there is no better purpose than what I see in the life sciences space. This was pre-COVID, and it’s never been even more amplified as to the noble mission of the entire market. So it’s profita ble, it’s well capitalized, it has an amazing purpose. So that mix, how can I take a lifetime of interest in digitization and automation, and enter a market that motivated me. And so that was, sort of, I had the luck of running into a venture capital firm that I would say had the same appetite for the same kind of companies as me. And I got to know one of the partners and we thought, post-Kuka, we should do something together. And so I had the fortune and opportunity and responsibility of stepping out of what had basically been climbing the ladder in well-established, mid-sized multinationals and wake up one day in my office going, so it’s just me and a few other people and what the hell are we doing? And that’s been fantastic. A fantastic journey. So that makes a life sciences market, trying to unlock the automation domain for scientists, combined with expertise in digitization and automation, is the mix that defines the purpose of Artificial.
Harry Glorikian: Yeah. No, I mean, it’s funny because the way you described what’s so fascinating about this market is why I’ve never left since I started in this market. Right? It’s like, why would anybody want to work in another market? This is awesome. But. But I’m biased. So. But so. I mean. Now Artificial is sort of helping biopharma companies and maybe more, right, because I don’t know all all the clients, basically automate their lab automation. But you know, before we talk about the technology you’re building, can you give us a picture of the existing state of automation in the life sciences industry? And, you know, at some point I’d like to get into your explanation of automation, because I usually think of, a lot of people may think of automation is like just moving one thing from one place to another. This is way more intense and complex in the number of operations that are happening and what’s happening.
David Fuller: Yeah, absolutely. And I think it’s, you know, words super matter. And so it’s interesting that we started off on this journey presupposing that the problem we wanted to go straight at was what I’ll just call hardcore automation. And given my background and the background, a lot of the folks that’s that’s basically where we started. I think we deserve kudos, not necessarily me, but my other co-founder who’s great, Nhikita Singh, you know, she’s one of the best product management people I’ve ever worked with. We did exactly an exhaustive tour of labs to answer the question you just asked her, which was like, Hey, what’s up with the current state of automation in the lab? Let’s let’s find out where some of the problems are. And what we found is there’s a massively important separation that bifurcates labs, using labs broadly in the loosest term. You know, are you production oriented? That is, you are trying to produce either a diagnostic or good as an output. You bias towards looking like a classic factory, right? What’s your widget? Right. And you like low mix and high volume and it’s all of the automation stuff that you can pretty much formulaically describe. And then you enter into the chaotic, crazy area of research where it’s less structured. If you move on to the other side of the wall where you’re looking at drug discovery and emerging new diagnostics, you’re perhaps looking at new forms of modalities and methodologies related to things like synthetic biology. So from my view, it’s super important to separate production from R&D, which is we obviously separated in most minds of people that know the life sciences market, but they also have very different needs from an automation point of view.
David Fuller: So we’ll put a pin in that in that. We’re focused on the R&D space, which can broadly be described, as a roboticist, as highly to moderately unstructured. So that’s sort of where we’ve been focused now along the way to the forum of looking at like, okay, let’s do automation for R&D. We really uncovered a current, important, what is the state of the current lab problem that moved us a little bit away from what I’ll call the word automation to a couple of other words which I can talk about. One being orchestration and the other one being operations. So when we look at the state of play in the current R&D lab, it is naive to think that it can be fully automated. So let’s have a hands free. All robots moving, all materials, liquid handler classes, abstracted. I say some magic words and I get a magic result, right? I mean, and this is sort of the ideal storytelling phase, is like the scientist just asks for what they want and out pops the thing. That’s just not the case. So what we’re seeing is there’s a huge amount of manual labor still involved. There’s a human in the loop from a central decision making point of view, which we do not think is going away. Right. That’s just this is not going away. There are aspects of the operations of the lab that can be automated, but the central decision maker being the human will remain.
David Fuller: So what we end up with is a hybrid automation use case where a large percentage of the lab processes will remain manual but formulaic, embodied in standard operating procedures. So this all mixed together led us to some following conclusions. We have to integrate automated SOPs, assays, etc. that mix a lot of human interaction with islands of automation that do very bespoke, powerful, amazing things. But they’re still just one little box doing a magic thing as part of the process. It’s a liquid handler, it’s a centrifuge, it’s whatever it is. So that led us to move away from pure automation, because you’re not really automating humans. You’re helping them with digital instructions to capture repeatable workflows that seamlessly flow data back and forth between automation and the scientist or the automation engineer. So it’s under-digitized, lots of paper, SOPs, lots of random human interactions because humans are random with limited data movement in an automated fashion. And so it’s pretty un-digitized and -automated from our point of view. So our goal is to wrap the human in a digital cocoon of really good instructions for what needs to get done with the automation. So it’s approachable. Connect the data of the human actions with the robotic actions, capture all of that in a digital record and put it in a place in the cloud that’s ready for AI or analytics or compliance analysis. So that’s sort of the state of the play and what we found and the gaps were in human interactions and data collection and normalization.
Harry Glorikian: Yeah, I keep hoping or I keep wishing for, you know, the Harry Potter magic wand where everything sort of does what you want it to do, which would be sort of awesome, actually. My son was saying, God, Dad, wouldn’t it be awesome if we had one of those to clean the house? But I think, you know, one of your board members, Kouki [Harasaki], who I also know from from M12, right, said, you know, most labs operate more like test kitchens than factories. Right. So if you’re doing a test kitchen. You’re constantly picking up a different spice. You’re working on a different aspect. You’re trying to make something happen, you know, to see what happens at the end. You’re testing multiple variables. So you got to have enough flexibility built in there to allow them to be able to do that.
David Fuller: That’s right. I think that that dovetails into and cooking is a great metaphor of like artistry versus industrial kitchen and the flexibility in, say, a test kitchen. What what’s interesting for me is like, one, if you are in the analogy of I’m experimenting in a fluid sort of way, whether it’s cooking or it’s drug discovery or whatever, wouldn’t it be nice if, as you were taking actions, all of what you were doing was being observed so that if you actually liked the outcome, you had a mechanism to remember the exact steps, variables, methodologies and things that were done to produce the result in an efficient and easy way without having to go and take your own notes in a constant manner or have it unstructured and paper and things like that. So, so that process of flexibly doing sets of known lab operations, tracking, sample, origin, history, what happened to it, who did it and when, how long was it where? You know, all of the things that go into the art of the initial discovery, we help track that. Absolutely.
Harry Glorikian: So, why would the average consumer care about like the primitive state of lab automation? Is there a way for you to translate it if you’re explaining it to your grandmother, let’s say, in the most basic terms, is the lack of lab automation, a major obstacle holding back, say, faster drug development, for example, or something else.
David Fuller: Yeah. So? So why automate, or why do we care? Right. So let me let me back up and I’ll steal an analogy that I’ve always loved for for tools that unlock efficiency and productivity. Let’s hearken back to the 1980s. In the early 1980s. And we’ll say, okay, what was the state of play of bookkeeping? What did accountants do? So I reconciled receipts for a restaurant in college, and I sat there with my paper tape and I would add things up and I wrote it down in a journal and I, I was it was great, Wite-Out, all sorts of stuff. At the same time, we had just bought a PC and had Lotus 123 in the back office and no one knew how to use it. So if you look at the way that the artist-accountant reconciled books and their efficiency on back office and in finance and in banking, and it was all manual in ledgers and on paper. And you look at what happened when ubiquitous computing combined with a spreadsheet unlocked productivity, it’s like, well, what was the value of the productivity to the world going from paper ledgers and unstructured formats to a regularized spreadsheet that we all know how to use and share information? And it was a phenomenal transformation that we just sort of take for granted is like, well, yeah, everybody’s more efficient by a massive amount. So I think there’s a huge amount of money being spent. I think there’s a huge amount of loss and friction in the system from the tools not providing repeatable, reliable results in a generalized way that are globally shareable.
David Fuller: And so there’s just a massive inefficiency drain on our ability to deliver new drugs, new therapies to the market. And there’s another trend that I think is really important as catalyzing what is happening. Why it is even more important is that the difference for some really amazing new therapies between the R&D lab and the production lab can start to disappear. So if I’m creating a solution that’s unique to you and I’m extracting your blood and growing your cells and modifying your genes and putting them back into your body, that’s a “lot sized one” production methodology. There is no production oriented pure factory for Harry’s genes. Right. And the tooling and and the hardware and software side basically are still primarily R&D oriented. And so you have these convergence of the wall between production and R&D is going away. And it’s pretty profound, both profound from a business model point of view about what happens when there’s no repetitive therapy and it’s a pure cure that’s unrelated to us. More related to us is it’s a lot sized one manufacturing use in R&D methodologies. So we should care because the most important, newest, amazing therapies that are transforming the world from CAR-T to gene editing need automation, but are using R&D oriented solutions that are largely manual.
Harry Glorikian:. I’m not sure that everybody understands that shift that’s happening or what’s happening in that space. But let’s shift here now to, you know, what is Artificial’s approach to solving this problem. Right. You’re not trying to replace existing equipment with new hardware built by Artificial. Right. It’s your technology is about helping companies build digital twins of their laboratories to bring together software representations of their existing equipment. I’m going to ask you if that’s like a correct assessment on my part.
David Fuller: That was a very correct assessment. Clinically written.
Harry Glorikian: I mean, but a digital twin is is a is a term that comes from, I was explaining it to someone, you know, in a conversation the other day. I was giving a lecture. But it’s from the Industrial Internet and the Internet of Things. Right. People who don’t follow that area might not understand the definition of a digital twin. Can you sort of give us an example?
David Fuller: Yeah, absolutely. I find it, in a grand fit of cosmic irony, you know, I now use most of the terms that I really despised for most of my career. And Digital Twin is one of them. And you’re right, its origins are in the industrial space. I had the pleasure of working with Dr. Henning Kagermann, who was the founder of Germany’s view of Industry 4.0, and he was also the CEO and executive vice president of SAP. So a very, very powerful intellect. He wrote a book on Industry 4.0. IoT in the US is based off of the theories in that book and he is one of the ones that focused on digital twinning. So for me as a computer scientist, I think it’s a fancy word for a well defined data structure, which is just like, are you tracking information about the thing in the real world in a reasonable way in the digital world? That’s like the most basic view of it. Now, if you have that data structure and it’s just sitting there, it’s meaningless in a certain way other than being pure data. There’s an implied with digital twins connection to a thing in the real world. So I have a thing in the real world. I have its twin in the digital world. There’s attributes and fields and metadata and structures that describe the real thing. And then I keep them in sync. So if the real thing moves the field that says “Where is it located?” updates in some reasonable time in the cloud, and then the world can globally understand the thing moved.
David Fuller: So this idea of tracking a real world entity with a digital entity and a centralized view on the Internet for me is sort of the most basic definition of a digital twin. There must be a real thing. There’s a corresponding digital thing. They’re connected in sync somehow. Sometimes I can change the digital thing and it affects the real world element. Sometimes a real world element moves and the digital element updates. So that is the underpinning concept. It’s not exactly complex. It’s like, yeah, if I have a real thing, I want to track it in the digital world. Okay, it’s a digital twin. So that is the basics. When we look at Artificial, we are not creating hardware, we’re not trying to go after the deep, rich, amazing domains of automation and instrumentation that are the domain of the liquid handlers and the sequencers and so on. I think these are not areas that we’re interested in going after. What we see is that there is such a breadth of complexity, domain know-how and never mind the scientific complexity and domain know-how that happens in a lab. That’s a really amazingly rich, complex place. So rich that it’s a barrier to having a scientist come in who just wants to get something done with their automated lab. And so if I have five vendors with five different kinds of instrumentation and I have a scientific process that in and of itself is massively complex, and then I have all the software domain of my ELN and my LIMS and my automation software, it’s like which one human being on the planet can understand what needs to happen there? And the answer is really no one. And maybe if you have that person, great, they have their own unicorn billion-dollar company, but there’s not a lot of them.
David Fuller: And so our mission is to provide what we call a single pane of glass that makes it easier and more understandable to see the state of the equipment and the science as it’s running in your lab. And to do that, the normalizing factor, like what makes it homogeneous and approachable, are digital twins. So the digital twin representation of a liquid handler from Tecan, a Thermo robot, a Precise robot, they all end up in one canvas and it’s like, Oh, I can look at my lab and no matter the complexity of the equipment or who it came from, I can understand what’s happening, why it’s happening, what errors are there? How long has my plate of cells been sitting there? What’s the reading on the on the cell count? Et cetera. So this is for us how we start with approachability, for scientists to work with their automation teams to understand and use their lab effectively.
Harry Glorikian: So essentially it’s making it much easier to move the process forward and, and also like reconfigure the whole process.
David Fuller: Yeah. So this is the reconfiguration part is critical and I’ll use just a very practical, basic example of approachability and the interplay between the scientists and say an automation team, typically. Sometimes it converges into one person for sure, but often at scale it doesn’t. You know, I have a liquid handler. It’s a great thing. It’s maybe from Tecan or Hamilton or whomever. I need to run some sort of assay. To do that, I have to configure the deck with the proper layout. These are very practical things, like how many test tubes do I need to have in the tube rack? What kind of plates need to be on these plate nests? You know, this is configuring the physical things that are going to move elements of your samples and the reagents and a process. Knowing how to do that when you have a bunch of things going on, is not easy to keep track of. We have a digital twin that represents the desired state of physical reality, and then we have a set of instructions, tools to create instructions, so that I, as a scientist, could walk up and go, Oh, I need to do a sample prep, here’s the deck layout, let me go and configure it. And we’ll guide them through that and check that it’s done correctly. It’s very basic. But if you are slightly intimidated about using that equipment because you don’t use it all the time or you don’t know how to reconfigure certain elements of it or what you need to use from a quantity point of view is related to the amount of samples and the complexity of the experiment, that starts to, the complexity keeps growing. We can abstract away all of that with the intent that scientists use more equipment more efficiently and work better with their automation teams. So this productivity through ease of use is is a critical, critical point for us.
Harry Glorikian: Let’s pause the conversation for a minute to talk about one small but important thing you can do, to help keep the podcast going. And that’s leave a rating and a review for the show on Apple Podcasts.
All you have to do is open the Apple Podcasts app on your smartphone, search for The Harry Glorikian Show, and scroll down to the Ratings & Reviews section. Tap the stars to rate the show, and then tap the link that says Write a Review to leave your comments.
It’ll only take a minute, but you’ll be doing a lot to help other listeners discover the show.
And one more thing. If you like the interviews we do here on the show I know you’ll like my new book, The Future You: How Artificial Intelligence Can Help You Get Healthier, Stress Less, and Live Longer.
It’s a friendly and accessible tour of all the ways today’s information technologies are helping us diagnose diseases faster, treat them more precisely, and create personalized diet and exercise programs to prevent them in the first place.
The book is now available in print and ebook formats. Just go to Amazon or Barnes & Noble and search for The Future You by Harry Glorikian.
And now, back to the show.
Harry Glorikian: Can you give us, say, a maybe a before and after picture of how life gets easier for an Artificial customer when they start using your software to rationalize their laboratory process? I mean, I think there’s a good case study on your website about Beam Therapeutics and I do like watching his videos on LinkedIn every once in a while, but maybe you could use them as an example.
David Fuller: Sure. Absolutely. So I think it comes back to… The before and after is, before is disconnected, highly paper oriented, lots of information in different places without a single clear view of what work is there to be done, what work is being done and how is it going, and what work has been done. And do I have a clean digital record of the scientific data, the automation data, and what the people did. So before, in the most basic sense, it’s less productive, it’s less integrated and usable, and it’s less observable. So when it’s not as observable in a complete way and it’s not as approachable in a complete way, then we view that as a very inefficient lab. And that’s not to say that labs aren’t… I just want to keep being very careful. Labs have amazing aspects of scientific instrumentation. Absolutely. It’s the connection of the flow holistically so as many people as possible can use it that is our goal. And so the before is largely disconnected, largely unobservable, and not a complete picture of a data record in a centralized place. You know, there’s this just… So that’s that’s what happened at Beam, if you look at the case study and that’s what we’re doing with other customers. And I look at it from a, you know, I have seen and I’ve been on both sides of discussions around AI driven robotics, AI driven factories, AI driven warehouses. And what’s interesting is there’s a lot of like “lab of the future AI driven drug discovery”—we’re just going to close the loop on the lab and magic will happen and all of these things. And I’ve been doing pursuing that goal in one form or another with practical old school techniques and applying the new ones in different domains. This is also related to the Beam before and after. Look, most of the state of the market talks about those things, but if you ask them, do you keep a digital record of what everyone did in an R&D context as they’re flexibly working, which you would need to feed your AI? Do you track the people? No. Are your instruments connected where the process data and the instrument measurement data is correlated in time to what people are doing and what samples are in it? No. Okay, so you don’t have that. Do you normalize across instrument categories so that you have a digestible digital twin with a standard view of the same physical concept so that you can learn on a common piece of data describing a common thing? No. So all of the preconditions that define where we … These are not the wrong places to go. We definitely want to go where we’ve solved these problems. All of those preconditions add value in the immediate case to make the lab more usable, more tractable, more reproducible, and move towards this dream. I don’t like talking about the dream. I like talking about the, let’s solve the problems right in front of us to move the state of the lab towards the dream.
Harry Glorikian: Right. So. You sort of open the door to, you know, artificial intelligence and machine learning. Where do where do those techniques come into the picture? I mean, doesr an automated lab generate cleaner, more interoperable data that can be used as training sets for AI and ML? I mean, what can customers accomplish with AI once they’ve automated their workflows? I’m asking because I love to get a better picture of the world that you’re trying to sort of corral the cats towards and get them to end up in that place. And then I think I’ve heard this term quite a bit, is “closed-loop lab.” Right. And how they’re enabled by artificial intelligence. I mean, maybe you can touch on those pieces just to give everybody a flavor of… because this show is about where data and biology intersect and how we employ all these tools that are, you know, everybody’s bantering around these words in the world.
David Fuller: Right? Absolutely. I think everyone is bantering them around. I just want to say I don’t think many are applying them in a productive way, because let’s imagine the perfect outcome, which is I have some centralized, trained model that’s connected to enough information that it’s learned how to efficiently suggest some form of drug discovery. And that would be, first off, that statement, it would be phenomenal if we’re applying it in practical, useful, amazingly targeted, successful ways. So let’s imagine that we have that. Then you need, to go with it, a closed loop, which is I’m looking at what’s happening in the lab. I have a lot of data coming in. I have a centralized model I’m training. It’s learning and getting better and better at making suggestions as to what to do next. And then it’s affecting the real world by suggesting the next automated operation to run in the lab with a different set of inputs to maybe find a more efficacious drug. Right. So the closed loop is basically, it’s the loop as old as the scientific process. Run an experiment, record the results, infer something about what to do next, run another one. And so that’s the loop. And the difference here is rather than a human in the loop and moving at the pace of human experimentation and human thought, we’re trying to close that loop more efficiently with automated systems.
David Fuller: So do automated systems produce better repeatable results compared to humans? Yes. Do they produce data that is more normalized and structured than an un-automated system? Almost assuredly. So your chances of success of closing the loop go way up if you’re automating it as much as possible. Now I want to back up a little bit to some very specific, tangible example of applied AI. But it also relates, to me, about when is the human going to be the one in the loop versus when is the machine going to be the one in the loop? And so we’re working with companies that are doing a lot of cell culturing, like I was alluding to. It’s basically, hey, I need to look at a plate of cells and figure out what to do. Do I need to grow them some more? Can I move them on to the next step? And you’re using analytical instruments to take measurements to check out how the cells are doing. That’s fine. And then the question is, who makes the decision to move it on? And is that decision a model in the loop or is it a human in the loop? And then you run into some very fascinating things. We have the technology to train models to look at image results on a plate and decide what should happen next. But in a lot of the cases we’re seeing, it’s not trusted by the scientists who are the artful ones who need to make the decision and have made the decision for years.
David Fuller: And so there is a human element to this related to adoption of the technology by the people that are really controlling what’s happening in these labs, which are the scientists. So I believe cell culturing is a fundamental, perfect example of I want to automate more. I want to normalize the data more. I want to feed it to models. I want the models to be expert decision makers. But I want to keep the human in the loop so that together we can grow confidence that the suggestions of the model are correct. And so the envisioned future is yeah, I can efficiently and with as little human interaction as possible, reliably grow whatever kind of cells I need to grow for these new therapies or these new diagnostics and AI and automation can be a huge part of it. But I don’t think we’re going to flip the market if we don’t bring the humans along. And so I think tactical applications where there’s a very clear human decision should be evaluated on a case by case basis to replace with AI. But I think it’s going to take years and years to get there.
David Fuller: I mean, just some statistics that I find useful. So in my role as the CTO of the KUKA group, there was a constant analysis of white space on processes. So if you look at car production, which has been a global hobby for very large companies for a very long time, you know, there’s like 4% of the known processes required to build cars that are not completely automated. It’s like in final assembly of the car itself. So there’s not a whole lot of room of growth in car production. Electronic batteries, throw them a little curveball. So they had to oh, we have to have a new battery production process. That’s flowed through the system already. So that’s gone. And so you’re basically left with not a whole lot left to automate. I was talking to Big Pharma folks and a few in a bunch of other labs in the space. In the context we’re talking about, you know, we’ve got 10% reliably automated. So and it’s always changing at a much more rapid pace than what you see in, say, a production of cars, because the technology really hasn’t shifted much. So I love the lab domain. It’s rich with complexity. It’s a mix of people and machines. The science is always advancing at a rapid pace. I think it’s an amazing, dynamic environment and you need tools designed to deal with that evolution and dynamism and tools that deal with the people and the machines.
Harry Glorikian: Yeah, I remember we used to like, you know, we, I knew people. I was like, that’s the person you want because they will get it to work. Like all these other people. That’s not going to. I mean, it was almost like that artist that you had to pick because you knew that they had, you know, the imagination and the way they could sort of manipulate something. There’s not that many people that are like that. You actually need automation to get this thing to move faster. But, you know, this is the question that really comes up like for every area when we’re talking about automation and artificial intelligence. Right? In a lab equipped with Artificial, are there fewer scientists? O r r do companies employ the same number of scientists but just put them to work doing higher value things and less manual work? I mean, what’s the what’s the dynamic and where is it headed?
David Fuller: Yeah, I, I don’t think that, specifically with Artificial—this was a much more complicated question when I was building huge amounts of industrial robots—for Artificial in the context of the lab, I we do not foresee it as sort of labor replacement. Only in the extent that the labor you have is more efficient in what they’re doing. So it’s a tool. If it makes you more efficient and you only need to get one thing done, and with that tool, now, one person can do it instead of two, it’s like, okay, yes, that’s the case. You had some sort of efficiency gain. There’s a lot of analogies talking about this, but I don’t think the spreadsheet reduced the number of accountants. As an example. It unlocked the potential to go further or faster as a business and share more information in a normalized way. With Artificial, we believe you can capture repeatable digital SOPs so that more people can do more science. We think that you can have protocols and assays that can run against different kinds of equipment so that you have a repeatable process across labs. You know, these are forms of, hey, shouldn’t the digital tools address these common pain points in running a lab and create it so that you can be more efficient in doing it? And more importantly, to me, like really fundamentally important, I can share it with other people with a better chance of repeating it. And so those things don’t replace work. They make companies more efficient in doing work and doing new work and doing more work. And from my view, the appetite of the consumer and of people for, can I live better, longer, in a more enjoyable way, is not going away. So if you solve one problem, it’s not like there’s another problem that humans don’t want you to solve in the domain you’re in. So I just don’t see it as a, it’s the, for me, it’s a common question, but I’ve never viewed productivity-enhancing tools as bad for society. And in the case of the lab specifically, I think it’s just more and better of what we need.
Harry Glorikian: Oh, no. In health care and life sciences. I mean, we’re way behind almost every other industrial sector. So we’ve got a lot we got a lot to do to catch up. And it’s exciting. I mean, I love being on this part of the curve because it’s a very interesting place to be.
David Fuller: So I just the one comment on that because it really touched on one of the things that I think is super important. Like when I dig into the lab, and like you said, I do think in the absolute scorecard relative to other markets, like I was just describing, the penetration of automation to replace processes is as a percentage really low.
Harry Glorikian: Right.
David Fuller: Now, it’s not because brilliant people from the dawn of society on haven’t been looking at these problems. You know, I really, my favorite punch line internally is “Life is more complicated than a car.” And so it’s like there’s justifiable reason for the pace of automation adoption. On the other hand, there is no justifiable reason for the pace of digitization adoption.
Harry Glorikian: Correct.
David Fuller: And so I think there is a laggard mentality in this market, both in tool adoption and software adoption, that is really hurting the global efficiency of the life sciences space. One is excusable by the complexity of the domain. Automation l agging is, I think, inherent to the complexity. The digitization thing I think says a lot about the existing market and things that are opportunities and really should change quickly.
Harry Glorikian: Yeah, and I believe that in the last ten years, we have seen a significant shift for many reasons. I mean, if you’re specifically talking about health care, everything was on paper. Right. And then they got forced to move to EMRs, which, you know, frankly, I’m not sure I would have pushed everybody to go on to a glorified accounting system and called it a patient management system. But that’s a different show. But what does success look like for Artificial? I mean, what would it be possible in a world where every bio lab was managing its workflow using Artificial? Do you think the company can become the behemoth on its own, or do you see a future where you have to pair with an existing behemoth like a Thermo Fisher so that you have access to all these different pieces of equipment. What does it look like for you?
David Fuller: Right. It’s a fine question. I think about that one every morning when I wake up. I was like, What is it that is the end goal? And I think we have a pretty consistent vision technologically and product wise, of the problems we think are the North Star of what we’re going after. And so I’ll take your question in two parts. It’s essentially what does the fruition of the application of the product look like as it matures and grows in our vision? And then business wise, what are the ingredients necessary for that vision to be globally adopted in a really broad way? I think these are these are two important, important questions for us, like what is our end goal technologically and business wise? And I haven’t touched on it as much, but there’s a really important catalyst for why we started focusing on digital SOPs for people, as well as workflows for mixing human and automated operations. So why that? So right when I was first looking at how to apply the seed funding that we got as a startup and we did a lot of this analysis of the current state of the lab, COVID hit. And so I’m not a life scientist, full stop. But I did go to the World Health Organization shortly after some of the national agencies around the world had started looking at how can I test for this thing? And I downloaded 17 white papers from the World Health Organization that came from Germany, from the U.S., from the early people that were publishing their COVID tests. Those papers were incredibly varied, massively heterogeneous, different forms of legalese, different forms of quality of English.
David Fuller: No offense to my German brethren who I’ve lived with and love working with. Of course they’re going to say it worked. The World Health Organization had a great, beautiful, legally perfect 20 page document that did not, initially. And so we had this this sort of like, why in the world would we not have a standard way to describe how to do a known set of activities that we could then share? And when I shared it, it could be run. And when it ran, it did the same thing. Now, that’s a simple statement. And I know under the covers there’s different kinds of equipment, different reagent mixes, different methodologies. Those should be the things we smooth away with the digital tools over time. I don’t think it will ever be perfect, but you could say, to your point, in partnership with Thermo here is an assay that reliably, repeatedly does a diagnostic. And in the end what you had was companies like Thermo who rallied around that quickly to produce a fixed thing that did exactly that. But that was in the latter days post the research and discovery and more of the generalization. So in answer to your question, I believe the world needs a standard way to describe scientific processes using standard lab equipment to capture what the people need to do and what the machines need to do, and be able to run it as port ably as possible across equipment. And in that sense, we would have a standard language for executing scientific experiments, drug discovery, etc. in a lab. That is our mission. So we want…
Harry Glorikian: I mean, you almost want to teach that. You’d want a class or a subset of a class on that so that anybody graduating would be able to go out and be fully functional.
David Fuller: Yeah. I believe that if you have such a thing and you had digital twins and you could run it in the cloud and you could observe how these processes work and interact with the equipment like you can with our twins and follow along a scientific set of instructions and see it in action, simulated or virtualized. And you had the standard thing, Eliza or whatever. And it was just a learning tool. Absolutely. So this ability to describe and repeat people operations and and machine operations, capturing a standard language for workflows, running it efficiently, operating your lab with it, seeing it run, seeing how it happens both virtually and physically. That’s the game. So so then the question becomes the game .
Harry Glorikian: No, I was going to say. So you want to almost, you know, take a page out of Apple’s playbook and have every university, you know, put this in place, right?
David Fuller: I mean, we’re not starting in the academic segment, but it is natural to me both as a learning tool and as a operational tool, that if you had a way to describe what needs to happen in a lab, it’s a great learning system, especially if it can be decoupled from the physical assets and run virtually, which we can. And I do keep coming back to the spreadsheet analogy which I totally borrowed from National Instruments. You learn on a spreadsheet in school, you’re not not learning on a spreadsheet. So if we are the tool that we envision we should be, we should be used everywhere.
Harry Glorikian: Right.
David Fuller: And we think the desire to repeat a lab operation and describe it digitally and then run it automatically is a universal desire for scientists. So it has really, really broad application. And then to your other question, you know, aspirationally, we want to be a hugely valuable key pillar in the life sciences space. I mean, obviously, if we realized that mission, we would be. This would be how scientists would share research that can be repeated, how we would efficiently operate diagnostics execution in both the R&D side and on the production side. So we envisioned that we would be a highly successful software based lab orchestration and operations company. I think there is a reality of the interaction of such software with instrumentation. And so you need good and strong partnerships with companies that are open to creating a modern digital ecosystem. This is really important for the community, like super important. And so you need as few walled garden strategies as possible by the instrumentation vendors. And so if you end up with silos of disconnected equipment and informatic systems that can’t share sequencing results, you’re really, it’s good for the single business, no doubt, no doubt. It’s bad for the lab in the world and the ecosystem to support it. So the question for me is, can this market get to the same spot that other markets I have been in got to, which was the customers demanded in automotive that there were no walled gardens. The customers demanded it.
Harry Glorikian: That’s that’s going to be hard. Right.
David Fuller: It’ll be really hard. And so so your question is one I grapple with a lot, which is, you have to absolutely have beautiful partners that are open to the digital ecosystem like we do. And there are a lot of forward looking companies and you can look at who they are based off their participation in certain standards and standard bodies, and they’re sort of this coalition of the willing company base of the supply chain. And we’re partnering with them. And it really has to be, there’s a technologically advantageous solution who is open and willing to play in the modern digital ecosystem concepts founded, really, by the Internet more or less. And then break down the walled gardens over time. And it really requires unified customers, a good technological solution, and enough critical mass in the instrument and automation supply hardware to pull it off. So that to me says, crystal ball, I have a very, very clear vision for the company on technology. There’s a lot of business alignment on the hardware side. But what’s been great is I have great partners so far with great companies like Thermo, like Tecan and others who are open and view the world the same way. There’s a digital ecosystem. They’re a participant in it. Very few labs are solely supplied by one vendor. Like you walk into a lab, it’s like I’m a pure Thermo house or whatever. It’s like really rare. So I’m optimistic that there’s a strong coalition already existing and the further we go and the higher value we get, the more customers will demand it.
Harry Glorikian: Yeah. Well, let’s hope it moves that way, because the faster we move things along, the faster we benefit all of us.
David Fuller: Absolutely.
Harry Glorikian: So great having you on the show. I look forward to seeing how the company evolves. I’m definitely going to be watching because this whole area is near and dear to my heart.
David Fuller: Absolutely. I appreciate the time today. The questions were great. I really enjoyed it.
Harry Glorikian: Thank you.
Harry Glorikian: That’s it for this week’s episode.
You can find a full transcript of this episode as well as the full archive of episodes of The Harry Glorikian Show and MoneyBall Medicine at our website. G o to glorikian.com and click on the tab Podcasts.
I’d like to thank our listeners for boosting The Harry Glorikian Show into the top three percent of global podcasts.
If you want to be sure to get every new episode of the show automatically, be sure to open Apple Podcasts or your favorite podcast player and hit follow or subscribe.
Don’t forget to leave us a rating and review on Apple Podcasts.
And we always love to hear from listeners on Twitter, where you can find me at hglorikian.
Thanks for listening, stay healthy, and be sure to tune in two weeks from now for our next interview.