“AI is a tool. It’s obviously a very powerful tool… And so I think there is value in having some kind of framework for helping evaluate these things and looking at allowing people to make wise choices.” – Ethan Mirsky
In this episode of The Chemical Show, we explore how NobleAI is revolutionizing the chemical industry by using science-based AI. Ethan Mirsky, President of NobleAI joins host Victoria Meyer to discuss how traditional AI approaches aren’t working and how NobleAI’s team is focused on incorporating scientific laws, design constraints, and materials properties into a model that could provide relevant insights and understanding. Ethan emphasizes the importance of customer data privacy, the benefits and potential negative effects of AI and algorithms, and the importance of creating a framework to evaluate their use.
Topics discussed this week:
- The problem with traditional AI approaches
- Importance of customer data privacy
- Role of AI in science and engineering
- Concerns and need for frameworks with AI
- Customer data and expertise in the modeling process
- NonleAI updates and goals
Please subscribe to The Chemical Show on your favorite podcast player. And, visit www.thechemicalshow.com to subscribe to our email list and get additional insights.
Watch the episode here
Listen to the podcast here
Using AI to Accelerate Product Innovations & Sustainability with Ethan Mirsky
Welcome to The Chemical Show. This is Victoria Meyer. Today, I am speaking with Ethan Mirsky, who is the president of NobleAI, a company that focuses on accelerating product development with science-based AI. Ethan is a lifelong entrepreneur with deep roots in technology and has built several successful businesses that tackle complex, high-impact problems. Ethan is passionate about using innovation and technology to improve environmental sustainability and reduce human impacts on the natural world. So I think we are going to learn a little bit more about that and we’re also going to talk about AI and all things great like that. So, Ethan, welcome to The Chemical Show.
Victoria, great to be here. Thank you.
Absolutely! What is your origin story? What got you interested in AI and applying AI to chemical and chemistry problems?
It was certainly a long and winding road to get here. Although, there’s a bit for me of coming back to my roots. I actually did my master’s degree at the MIT AI Lab quite a long time ago. At the time, it was sort of towards the end of one of the AI winters, so things felt a little struggling but some promising seeds. I was actually on the hardware side. It’s the building chips that could accelerate computation that led me to my first diversion, which was telecommunications chips. It’s the idea of taking these chips that were designed to process information and using them to handle the voice-over IP.
So the conversion of the old telephone lines into digital back in the first dot-com era. So that was my first company. And then in the middle of the dot-com crash, everything kind of fell off a cliff there. I ended up getting really interested in the field called synthetic biology, which is all about engineering biology. For me, I think it’s that transition of applying one set of disciplines to another one. So applying engineering to biology was really exciting. Because of all the things you could do particularly around things like sustainability and making materials better. I ended up starting a company. I got my Ph.D. in that field and then started a company doing materials for textiles.
So using biology to make a variety of new textiles in a sustainable way. That was super exciting. I worked on that for about 10 years, and then time to move on time to try something different. I really wanted to get more into the sustainability side of things. I joined a fund run by a group of future ventures. It’s a family office of investors in my first company. He started funding sustainability. They’re exploring the field and really ran through this problem of there are so many different approaches.
There are lots of materials and lots of chemical processes, and all trying to address these problems that we see in the world of how we reuse materials. How do we more efficiently use materials? They’re all excited. I just couldn’t figure out which one I wanted to do. And then he introduced me to NobleAI. He was an early investor. I realized it was a tool that anyone could use to attack these kinds of problems. And that was really exciting. So selling the picks and shovels to the gold miners approach. This ended up getting me excited and joining NobleAI.
Awesome! So tell us a little bit more about NobleAI. What it is and what are your ambitions for the company?
Normally, the mission here is to use what we call science-based AI. I’m sure we’ll talk about what that is, and our reactor platform to create commercial solutions that allow chemical and materials, product development companies to develop their products and make these innovations faster, more quickly, less expensively than ever before and really be able to address the problems in the world. So creating those kinds of commercially available solutions that AI can do.
Alright, so we’re gonna get more into what NobleAI does, but let’s take a step back maybe a bit. Can you break this down for our listeners? What is AI? We’re hearing a lot about it. Who hasn’t heard about Chat GPT? Actually, I’ve talked to a few people who apparently have not, although my mom, who listens to the podcast, knows all about Chat GPT, although she hasn’t used it. But there’s a lot of AI around us., I think we don’t really understand or appreciate it. So what is AI? And where did it originate?
AI is really a catch-all name for any kind of machine that is capable of mimicking aspects of human intelligence. So that could be vision, language, prediction, and anything like that. And in some sense, it’s been around for millennia. The ancients had an automaton that would, that would do things. Technically, that’s artificial intelligence. Obviously, now, we generally refer to AI to meaning computer-based artificial intelligence, which really got it started in the 1950s, as computers became starting to become a little bit more powerful, and people started imagining what they could do with this universal machine.
How is it different from just computation? I think about it, certainly, I’m an engineer at my roots, and we always talk about how it needs an LP or linear programming model to solve some of these problems. Is it different than that? Is it just extending upon that? How is this different than just computation?
It is computation, in a way. It is a different way of doing computation with the aim to get into attacking or approaching the kinds of problems that we think of as intelligence. So at its core, it is still computation. There’s usually an element nowadays of what we call machine learning. So an approach whereby the system, or the machine is capable of taking in new information and improving its own computation based on that new information, changing the way it is doing that computation based on its own information. So there is a kind of positive feedback loop going on in there. And there are many, many different ways of doing that.
Product Innovations & Sustainability: We need to create a regulatory
process for AI that's similar to the biotech industry. builds models
that scientists and engineers use to accelerate experimentation, while
the feedback loop helps maintain integrity in the results.
Let’s talk about that. What are the different forms of AI? And how are we interacting with them today? Because I’m sure that we’re interacting with them more often than we think.
Absolutely! It’s everywhere. It comes down to what things you perceive as not intelligent. It really kind of has some intelligence to it. Obviously, people think of computer vision in self-driving cars here. In San Francisco, you’ve seen people driving cars without drivers, or people in the driver’s seats.
Do people really do that?
They really do that.
All right. I’m not there yet, but it’s okay.
With face recognition, you point and shoot a camera that happens to know where the target faces. That’s a vision. That’s an AI-type problem. Obviously, it’s natural language processing. You talk to Siri or Alexa. That is AI. Expert systems are very frequently used. It’s been used for many, many decades to help make recommendations for medicine or troubleshooting, or processes, where you’re just trying to encode. What is the thought process of how you diagnose a problem? That’s a form of AI. Recommendation systems that you might use to recommend music and movies. And then obviously, the more recent thing of the large language models like Chat GPT, that generate texts or images based on prompts. Those are all different forms of AI. Frankly, there’s fun that NobleAI. We call it science-based AI.
Tell us about it. It’s a science-based AI. What is that? How does it differ from other forms of AI?
You are in the chemical industry for a long time, and you know how difficult and time-consuming it can be to develop a new product and to develop new material. These are very complex problems. They’re very difficult simulation technology, which really sort of made a huge difference. I came from the electronics world where you don’t do much in physical experiments because you can simulate everything. That doesn’t work so well in chemistry. There are just too many layers going on. For instance, like a simple problem of even creating an environmentally friendly detergent, you need to actually know about what are all the ingredients. What is the evaporation rate?
What’s it going to smell like? How long is it going to last? All of these things require knowing what’s going on at the molecular level, and what’s going on at the macro level. What the environmental conditions are? And trying all of those possibilities. And that’s just a very difficult thing to simulate. And so the idea was to apply AI to that. The challenge there are most forms of AI, as we talked about, require training. And so you need a lot of data. Chat GPT is famous for having basically scoured the entire internet for its training data. You don’t have that when you’re making a product. You’ve done a few experiments.
Some of which are actually proprietary, probably.
Exactly! Even if there are publicly published experiments, they may have used a different method. So the data is not really as relevant. Traditional sort of AI approaches didn’t work. The idea here was to approach the problem the way the scientist or engineer would do it, which is you’re going to look at the data. But you’re also going to look at what are the scientific laws and principles that apply here. What are the design constraints? What are the properties of the materials? And try and incorporate all of that into a model that can understand this problem, not all problems, but the problem you’re dealing with, and deliver the kind of insights and understanding that allow the developer to make smart decisions, and really accelerate instead of doing thousands of experiments on your computer before you do the handful of experiments in the lab to confirm that it works.
Interesting! Do you have any specific examples of product or development areas that you or your clients are working on at the moment to utilize NobleAI?
We’re actually working in a variety of areas. We work on batteries, predicting the lifespan of a battery based on its structure and chemical properties. I mentioned the sort of laundry consumer products. It is the sort of thing that we would do. We’ve also worked in generating clean energy, and hydrogen, and there are actually quite a few. They continue to grow. We have a large number of customers growing.
That’s really awesome. I think you’ve maybe touched on this already, but where do you actually find this data? You’re starting with a certain data set, and it sounds like scientific principles. And then if the machine creates more, I suppose, and then companies are adding data into that. Is that how that works?
Yes! The process of working with a sort of science-based AI approach is to take the data that the customer has. You’re in your lab. You’ve done some experiments, and you have some old data from older products. We take that data and combine it. Sometimes, we have publicly available data, if appropriate. We talk to the customer subject matter experts to make sure the system understands what the relevant constraints are. What are the insights that you’ve already generated? So we don’t have to repeat what you already know. We have our own internal science experts that can look at a problem and say that this is diffusion. So we need to include this law. We also work with outside experts and university professors who advise us that in this field, these are the things that you need to model and those are the things that we incorporate into our models that the customer can then use.Noble AI helps products get to market faster and reduces reliance on risky supply chains, making it an important tool for achieving sustainability goals. Click To Tweet
How do you keep proprietary information? I know this is a hot topic at the moment in the public domain because I think there’s been some government-related documents that have been input to Chat GPT to improve and now all of a sudden, where did it go? From a Noble perspective and with you and your clients, how do you maintain firewalls or keep proprietary information proprietary?
That’s a really good question. It’s something our customers care a lot about, and I think everyone should care a lot about it. We’re actually in a pretty unique position. Because we don’t work the way that Chat GPT does. We’re not building a single model that everybody gets to use, and therefore, we need everyone’s data to train it. We’re building models specifically for a customer. When we take some customers’ data, that data exists in an isolated cloud instance, whether it’s our cloud or even bare cloud. The model is trained there and it is that customer’s model exclusively. The only thing that we carry forward is that we learned how to create that model quickly. So we can create more models more quickly. But we do not use any data from any customer to ply to another one.
So it’s all isolated. You’re effectively creating multiple AI models for each customer.
Exactly! In some ways, it’s both a challenge, as I said. There’s a limited amount of data for a given product. But it’s also an advantage in that sense because, with only that limited amount of data, there’s no point in trying to apply it to someone else. Because there’s a different product over there. It’s a different system. So we get to keep those things separate.
One of the things that I know people are concerned about, and quite honestly, when thinking about algorithms, social media, Facebook, Instagram, even some of these new sources, Apple News, and what have you, gets a bit of a bad rap because it’s a spiraling effect. It feeds into each other. I think there’s a concern along the way that AI and the algorithms are eventually really skew results of whatever variety, whether it’s news pieces that are being fed to you, whether it’s scientific results. What do you think about that? Because I think people want to keep integrity in it. Algorithms obviously speed things up, but how do you prevent the skew?
Let’s answer two ways. I think there’s a general problem. And then I’ll talk more about what we do in Noble. In general, AI is used as a tool. It’s obviously a very powerful tool. A lot of its implications are poorly understood because it’s very new. As with any powerful tool, it can be used for good or bad and has consequences that we may not have anticipated. And by itself, these tools and algorithms can’t take away choice or information but people can use them and choose to use them in a way that does. And so I think there’s value in having some kind of framework for helping evaluate these things and looking at people to make wise choices. For instance, I came from the biotechnology industry, where there are some incredibly powerful tools that manipulate DNA and can sound really scary.
At the same time, there are systems, regulations, and processes in place to ensure that or at least to help ensure that it’s used appropriately. And obviously, those systems are not perfect. They need to be continually updated as technology advances. Still, over the last several decades, it has gotten to the point where we can make COVID vaccines and other kinds of life-saving treatments really quickly with these tools and mostly avoided most of the major risks that we’ve all might be concerned about. I’m sort of imagining there’s a value to a similar kind of process that needs to be in place for AI, in general.
Speaking more specifically about Noble, one of the reasons why I’m attracted to it, is that I feel like it’s easier to guide and direct because by nature, we’re aiming at problems and we’re aiming at systems and embedding science into the eye so that it’s not capable of producing answers that are wrong. They’re always verifiable. If you’re going to say, hey, this is the product you should make, you can go and build it. You can test it, and you can see that actually does work. The scientific laws do apply to that problem, too. We’re building in a way that the scientists or engineers do. This isn’t a replacement for them. This is their tool. This is like your Excel spreadsheet for the accountant. It’s something that allows you to do what you would naturally do just faster. You can explore more things. But you can always go back to check something and go back to the lab and verify if that really is the case. I feel like that kind of feedback loop is what helps keeps us honest. That’s what we really want to aim for.
When we talk about Noble and science-based AI, how do you see this changing how material products are developed? Where does this fit in the process, maybe product development?
Obviously, the traditional approach is that you have an idea, make your design, do some experiments, you collect data, and you repeat the cycle. There are definitely tools out there to speed that up and help do the data analysis more quickly. Science-based AI and what Noble’s platform is really intended to do is allow you to do that process, but do a lot of that kind of testing. So instead of trying to design, doing your experiments over weeks or months, and trying again, you do some experiments. You may even have data you’ve already created from previous products. Train our science-infused machine learning models, and then use that to run a bunch more experiments. There are thousands of them.
That helps you explore the space. And so you get a lot of information that you can look at and say that now you understand where you should be doing more experiments, or even say that this is where the product needs to live. One of the benefits of having the science in the network is like Chat GPT, which might make up answers. Our tools can say that this is our confidence level in this answer. We know this is very likely to be true, and this is less likely to be true. So you can use that to say that now you can understand that. So it’s really going to accelerate your process by again, may allow you to re-explore the design space, and just very quickly develop that understanding of here’s all the things you can do.
Product Innovations & Sustainability: Applying AI to product development
in the chemical industry can help by simulating various possibilities
and conditions, allowing designers to quickly explore the design space
and better understand what works and what doesn't.
It’s good. I actually read an article recently that referenced, I want to say it was Unilever, using AI. So it’s a similar process, I suppose to NobleAI, to actually accelerate some of their personal care development and some of the cosmetics. So the example they use that stood out to me was lipstick, trying to get to the right lipstick colors that they were able to instead of having to test it physically a hundred times, they did it all in the background, and they got very quickly to the optimal answer in terms of pigments and chemistry and what have you that goes into that. So we’re starting to see it. It takes time to get to that.
That’s absolutely right. What Noble wants to do is to make that available both to companies like Unilever, which have their data science teams, and just might need some extra support from you. There are so many problems that rather than having multiple teams tackling so many problems, we have a platform. It’s robust. It’s cloud-based, and it works very reliably. They can use that to support their work, or for many companies that don’t have access to that kind of internal resource. They can come to us and get that benefit.
That’s what I was gonna say because I suspect that many companies don’t have data science teams, where they think that data science is about IT and how the computer runs, as opposed to how the application and how the products are developed and those resources. I think there are still a lot of companies, most companies probably don’t have that level of resources.
Absolutely! We work with all of them.
Where is it Noble in its development lifecycle? I know you guys are still pretty early. So when are we going to start seeing you in primetime or are you primetime yet?
We’re beginning primetime. We are in commercial use. As I said, we have many customers. They’re continuing to grow. We recently closed series a of funding led by Microsoft, Chevron, and Ge PVC. And we’re out there in the market and working to grow rapidly. You might see us at trade shows. Hopefully, you hear the results of companies that have used this. Just anyone can come to our website noble.ai and ask for a demo.Science-based AI provides a confidence level in the answers it provides, allowing designers to make more informed decisions rather than relying on potentially unreliable information. Click To Tweet
That’s really cool. So what’s next for you guys? In terms of a platform of where you are and where you’re going, what do you see as being next? What’s your next focus?
We’ve just completed a brand refresh that we put. You’ll see sort of the new logo, new materials out there to really help our customers and our potential customers really understand what it is we do and how we work with us. And really, the focus in the next year is to get this out there. It’s a tool and it works. The more you use it, the more impact you can have. And to me, that’s what’s so motivating. This is actually going to make a real difference. Products are gonna get out there faster. People are gonna be able to make a more sustainable product, or product that doesn’t rely on supply chains that are risky. You can switch your materials and still know that your product is going to work. That’s really want to do. Obviously, we’ve got many new features coming, new capabilities, and we can talk more about that. But really, for now, the focus is that let’s get more people using it.
Let’s use it and make it happen. Ethan, this has been great. I really appreciate you taking the time to talk about AI because I think we need to demystify it a little bit more. And also to share the story of NobleAI and what you guys are doing. So thank you for joining us today.
It’s my pleasure. I’m happy to be here.
Absolutely! Thanks, everyone for listening to The Chemical Show. Keep listening, liking, and sharing, and we will be back next week with another episode.
About Ethan Mirsky:
Ethan Mirsky is the President of NobleAI, a company that focuses on accelerating product development with Science-Based AI. He is a lifelong entrepreneur with deep roots in technology and a history of building successful businesses that tackle complex, high-impact problems.
He has worked in a wide variety of fields including semiconductor chip design, software engineering, synthetic biology, and sustainable textiles. He co-founded and built two successful companies while leading efforts in every aspect of their operations. Ethan is particularly passionate about using innovation and technology to improve environmental sustainability and reduce human impacts on the natural world.
Love the show? Subscribe, rate, review, and share! https://thechemicalshow.com/