Johnson & Johnson's Senior Director of Data Science, Curren Katz, explains the parallels between neuroscience and AI development, including when we should strive to mirror human cognition in technology, and when copying human learning actually may be a hindrance.
Curren is a curious, driven, and creative leader with vast experience in data science and AI. Her original background was in neuroscience and cognitive neuroscience but entered the industry when she realized how much she enjoyed programming, maths, and statistics. Additionally, her biology background gave her an advantage, making her a perfect fit for managing the neuroscience portfolio for Johnson & Johnson. In our conversation with Curren, we learn about her professional background, how her biology background is an advantage, and what she enjoys most about data science, as well as the important work she does at Johnson & Johnson. We then talk about AI in the pharmaceutical industry, how it is used, what it is used for, the benefits of AI both to the company and patients, and her approach to tackling data science problems. She also tells us what it was like moving into a leadership role and shares some advice for people wanting to take the plunge into leadership.
Key Points From This Episode:
Tweetables:
“Finding new ways to use data to drive diagnosis is a big focus for us.” — @CurrenKatz [0:11:56]
“In data science, it can be challenging to define success. But choosing the right problem to solve can make that a lot easier.” — @CurrenKatz [0:15:27]
“I want the best data scientists in the world and to have those people on my team or the best managers in the world. I just need to give them the space to be successful.” — @CurrenKatz [0:23:55]
Links Mentioned in Today’s Episode:
[INTRO]
[00:00:04] RS: Welcome to How AI Happens, a podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists, and machine learning engineers, as they get technical about the most exciting developments in their field and the challenges they're facing along the way. I'm your host, Rob Stevenson, and we're about to learn how AI happens.
[INTERVIEW]
[00:00:32] RS: Here with me today on How AI Happens is the Senior Director of Data Science over at Johnson & Johnson, Curren Katz. Curren, welcome to the podcast. How are you today?
[00:00:40] CK: I'm great. Great. Thank you so much for having me.
[00:00:43] RS: I'm really pleased to have you. How has your week been? What is the data science you've been senior directing this week?
[00:00:48] CK: Yeah. It's been a good week. I've been working on a few different projects, clinical trials, and then just helping my team look at our portfolio overall, and plan for the future.
[00:01:00] RS: Got it. Johnson & Johnson, I assume, is a company my listeners will know well. It may even be coursing through their veins at this very moment. It is mine. That's for sure. So let's just start with you. Can we hear about your background and how you wound up at Johnson & Johnson?
[00:01:13] CK: Sure. So my background is actually in neuroscience and cognitive neuroscience. That may seem like an odd fit for data science, but it is not. It's actually a really good fit for my role right now, which I focus on the neuroscience portfolio for Johnson & Johnson. So that's kind of nice, full circle. But neuroscience was actually a great preparation for data science in a few ways. One, just the math and statistics and programming was a great skill set, right? I loved that part of it. I really enjoyed it.
Another way that people – I mean, it may be obvious, but things like neural networks. Those were things we're thinking about in neuroscience and the brain or learning. My focus was cognitive neuroscience and how people think and learn and how the brain does complex tasks. Especially in a connected way, how the neural networks drive this.
From that standpoint and where the field went, it was a great background. But I primarily worked in the payer provider setting most of my career. I did a PhD in neuroscience and did some research and fMRI research specifically, but also just worked in hospital settings. I developed and led the data science department at Highmark Health, which is a large integrated payer provider system.
From there, Johnson & Johnson reached out and said, “You have this neuroscience background, this data science background, and would you consider coming to Johnson & Johnson?” Of course, as you said, there's a pretty big impact on people's lives. So I was intrigued and honored.
[00:02:51] RS: Yeah, yeah. Of course. It’s clear where the name for neural network came from, right? I guess I always had it in my mind that it was more poetic in a way, more like a metaphor, sort of like, “Oh, this is analogous to the way that the brain works.” But it sounds like that's not the case and maybe the naive way of describing it. But could you maybe explain a little bit, when you think of traditional neuroscience and neurology of the biochemical variety that we think about, what is the connection between that and then its role in neural networks in AI?
[00:03:25] CK: Well, I think neural networks in AI, there was some inspiration and maybe poetic inspiration, but also practical inspiration with concepts like feedback and learning and different ways that can happen and vice versa. As a neuroscientist, I saw like ideas and things we had learned with artificial neural networks kind of being applied back to biological networks, which we don't fully understand. So there's this back and forth. Can we use the way the cerebellum connects and interacts with the cerebral cortex? Is there some model there that could work with artificial neural networks and the data that we're working with another scenario that make sense?
Then sometimes, we would look at things that worked in artificial neural networks and say, “I wonder if that is similar to the brain and the way the brain works?” The brain is very, very complicated. So I'm not sure, we have a lot of work to do and a lot to learn. But, yeah, I think it's very interesting that conversation over the years.
[00:04:32] RS: Yeah, of course. Of course, the most foundational definition of AI that's at least been explained to me is that we're trying to mimic or replicate human cognition, of course, taking place in the brain. So it makes all the sense in the world why we don’t have more brain experts participating in the field.
I know we wanted to talk about other stuff, and we will. But now that I have this neuroscientist here, I feel like it'd be wasteful not to get into it. Could you maybe like – what are some other examples of ways that we can sort of use a traditional scientific education in neurology to inform not just neural networks but AI at large?
[00:05:06] CK: Well, I will say even more broadly than neurology. It’s just to keep it simple learning or how children learn and how algorithms learn. I know and I've explained AI to people who are not in the field, to business leaders and different areas, and trying to explain what we're doing and why it's helpful to them. Sometimes, I go back to how children learn and what funny things you notice. Children have a limited set of inputs like this is a cat. Okay. All cats are this color. So maybe they see a cat that's a different color, and they say, “Oh, that's a lion.” It takes time to learn all of these distinctions, and you see these funny sorts of mistakes that a kid makes as they're learning.
Similarly, you have to think about that when an algorithm is learning, right? It’s the same consideration. Sometimes, it's such a mystery to people like how the data turns into the prediction and how the algorithm’s learning and what things they, as an expert in their business area and maybe the data generates, how they need to think about it. But that it's not very technical or brain-specific. But it's just like a practical example that I think studying people and the human brain and how it works on a very practical level, like anyone can use to help people understand the types of things they need to think about when training an algorithm. That's probably a less interesting technical example than you were looking for.
[00:06:39] RS: No, it's fantastic because I understand it. When you put it that way, it's such an obvious dot to connect. How much do we understand about human learning? Are we really ready to replicate it? Like what is the scientific measure of how we understand the way that human brains grow and learn?
[00:06:53] CK: I would ask if we want to replicate it. There are a lot of ways that AI might be better suited for what we're trying to do than human learning exactly or the way the human brain works. I don't know, but I've always questioned. It's a good inspiration. What makes it more fit for purpose for what we're trying to do? We're trying to predict something or make a next best action recommendation. Maybe the way the brain does it is the most efficient way. Or maybe the way people learn is the best way to get this algorithm to learn. But maybe we can make it better or more efficient.
It’s certainly not going to be as complex and comprehensive, but it could be better in certain ways. We might have more control over bias and fairness and things that are hard in people because the brain is complex, and people's learning and experiences are complex. We have an opportunity to keep it a little more simple and maybe optimize the outcomes with an algorithm.
[00:08:00] RS: Yeah. It's a great point that maybe trying to replicate the human brain is thinking a little too small because human brain isn't perfect. So we have this opportunity. If you were inventing the human brain, which I guess kind of we are, you would change some things, right? You would change this tendency for bias maybe? Or can we remove the bug where I go into a room to get something, and I forgot why I walked into that room. Then I have to leave and come back later. It’s not a perfect tool, right? So why strive for sub perfection?
[00:08:31] CK: That is a great bug to remove. I personally would very much like my brain not to do that. That sounds great.
[00:08:39] RS: If someone out there within the reach of my voice can get working on that, it would really, really make my life better. But, yeah, it's a great call out that the brain is inspiration as a starting point. But let's think a little bigger, shall we? Let's get into Johnson & Johnson, though. Can you share a little bit about your role and how you kind of characterize what your job is at J&J?
[00:08:57] CK: Well, it's interesting. My role is to lead the portfolio of data science products for the neuroscience therapeutic area. We cover discovery through development and neuro psychiatry, neuro degeneration, the whole range of things that can affect the brain and people's lives.
So interestingly, as we say, start with the brain but think bigger. We're also trying to use AI, and not necessarily a one-to-one matching, but use AI to look through data and understand that the brain and these diseases better, so we can better treat patients. There are a lot of unmet needs in psychiatry. If you think about it, we have all of this data, all different types of data. We are taking a precision approach to psychiatry, for example, to understanding the disease, to understanding the best treatment for a given person.
You don't get the same recommendation shopping. So maybe, well, we should bring as much precision to treating these serious things in people's lives, as we do in other areas.
[00:10:17] RS: Yeah. I'm glad you brought up psychiatry because particularly if you were to diagnose a mental disorder, for example, it's sort of on a finite list of self-offered symptoms, right? It’s like interview-based, and there's all these challenges I'm sure you know with two psychiatrists probably don't agree on a diagnosis, which is not the case in medical diagnosis. So can you go a little bit into what precision means in that sort of example?
[00:10:44] CK: Well, there’s precision diagnosis and precision understanding of a psychiatric disease. Maybe there are different subtypes of patients that respond in a different way. There's precision understanding of the disease itself. What are the subtypes? How is the brain different in this particular group of patients versus another, maybe even with the same diagnosis? So really rethinking or getting a little more precise on what the diagnostic categories are, so we can better treat them and then making the diagnosis, as you mentioned.
There’s a lot of kind of self-report and symptoms and maybe caregiver report, and that can be hard. It can be subjective. It can be really challenging to reflect on yourself. Then there's this subjective nature with how psychiatrists may interpret that, especially in real life clinical practice. If we can find some ways to measure and diagnose these diseases in the same way we diagnose any other disease, it'll be very helpful to patients. So finding new ways to use data to drive diagnosis is a big focus for us.
Then it's not always the first thing that comes to people's minds in psychiatry or anything else. But we want to improve the operations of what we do every day. We want our clinical trials to be efficient. We want them to use technology that's patient-friendly. We want to go to the right places where the patients live that could benefit from our future therapies. So it's all of those things or opportunities to use AI. It's a really endless list and a huge opportunity for any therapeutic area. But I think it's particularly exciting in psychiatry.
[00:12:40] RS: Yeah, yeah. Of course. Then your role sort of overseeing this portfolio, how would you say you kind of involve yourself and spend your time. Are you just sort of looking over, okay, this is the correct data science approach on a case-by-case basis? Or what is the high-level strategy you are putting over the operation?
[00:12:59] CK: I think from time to time, I do get involved. I just can't help myself in the best data science approach. I've learned to try not to dive down in the details, moving from being a data scientist to leading data scientists. I have found myself doing that, and then I have to say like my team is like probably not all that interested.
Back in the day, when I was a data scientist, we used to do this when it’s a field that's growing all the time, and no one likes to be micromanaged by their boss who has some memory of the field when they used to do it. I try not to do that, but I'm sure I do it from time to time. But really, I'm thinking about the strategy in the future, how we can make an impact.
Of course, if anything doesn't work out, I want to make sure I'm the person who's out there. So my team has the freedom to keep going and when I just want to set them up for success and what will work. So finding new opportunities, and partnerships, thinking long term about what we do with our portfolio. What is the mix of things? How do we measure success? Those sorts of things.
But there are projects that I get involved in from time to time more deeply, if it's something that's very new, or we're really working something out, just to make sure everyone is set up for success, and I can step away.
[00:14:25] RS: Right. To scratch that itch, you have being a data scientist yourself, right?
[00:14:29] CK: Right. The data science, it really is something I think you have to find a productive way to scratch the itch. I had a former boss who I really liked his approach. He would dive into the details from time to time, but he would always say at the beginning of these meetings, “I'd have all the data scientists there.” But he would tell everyone, “Look, this is for me and like energizing me. I'm not asking you these questions because I don't trust you. This is just over time I've learned this is what I really, really loved to get to do every once in a while.”
So like you have to find a productive way to spend time diving in and having a bit of fun, but make sure that your team doesn't think you're doing it because you don't trust them or you're second guessing their work. Make sure you leave time for those things you really need to do as a leader that your team needs you to do to be successful.
[00:15:23] RS: Yeah, yeah. Of course. How do you measure success incidentally?
[00:15:27] CK: In data science, it can be challenging to define success. But choosing the right problem to solve can make that a lot easier. Sometimes, it's a financial metric, and that's always great to be able to pull out and show. But there are so many other things that you can measure. I think the important thing is being able to find something, even if it's not super precise. It doesn't have to be but something that can show you've made a difference.
At Highmark Health, we worked on scheduling for patients receiving chemotherapy and being able to show that their wait time was shorter, they were more satisfied with their visit, and that was reflected and in the patient satisfaction feedback that we were getting. It’s amazing to know you've made a difference in that way. So really thinking about how is this thing I'm doing in data science going to make a difference, and then how can we best measure that and make sure that we're doing that.
[00:16:31] RS: Yeah, yeah. It sounds like efficiency is sort of a big driver because particularly with something as complicated as clinical trials with as many steps, any chance to shorten that process has huge downstream effects, right?
[00:16:45] CK: Absolutely. I mean, medicines get to patients faster. It’s quite serious. That day, week, month, year is life or death, in many cases. So, yes, that's a big one. It's also the experience during the clinical trial. If we're more efficient, the experience for the patient, for the investigators, and the research sites that are working with us, it's a better experience for them as well. This isn't pure AI or data science. But actually, I think it is. You have to be able to deliver a product that people can use and that has an impact, and part of that is understanding the workflow, how your data science product is going to fit into that, and how people are going to use it effectively.
We want to avoid in the middle of a clinical trial a coordinator who has so many things to do having to open multiple screens or coordinate multiple devices. We want to avoid anything complicated for the patients who are volunteering to participate. So those things, efficiency, operational, simplicity, I think go a long way and are quite important.
[00:17:59] RS: Yeah, yeah. Definitely. So I want to crack open the hood of Johnson & Johnson a little bit. What are some ways AI is being deployed on your team anyway?
[00:18:09] CK: There are quite a few ways. Measuring treatment response is a big one as well. If we can better identify someone who's responded to a treatment, and we can use AI and AI-driven technologies to do that, then we have a better chance of getting that therapeutics which could help them get to the market.
Also, as we talked about planning for clinical trials, finding the right places to do our clinical trials. Where do our patients live? Who are great patients for this clinical trial? All of those sorts of things. How do we maybe optimize? These are all broad definitions of AI, and we can talk more specifically. But these are all things that you think about.
People are very familiar with COVID and COVID-19. There's a lot out there about how we found sites for the vaccine trials, right? We have predictive algorithms to say where's the disease, like where's COVID spreading? Because it takes some time to plan a clinical trial and get it going. So we needed to know ahead of time where COVID was likely going to be, so we could plan clinical trials in those places. That's a way that affected a lot of people's lives that was AI, like a great classic use case for AI.
[00:19:30] RS: So million different use cases, obviously. We did kind of skip over this earlier, though, and I wanted to make sure we spent some time speaking about your career a little bit, Curren, because you have this background as a data scientist. I think anyone, if you do your job well enough, eventually you get tapped on the shoulder and asked like, “Hey, are you interested in management and leadership?” You made that shift recently.
Would you mind sharing a little bit about that journey, how you knew it was time to move from the individual contributor role into leadership, and then what the experience has been like since?
[00:20:00] CK: Sure. For me, the shift was somewhat gradual. I had been managing people in the past, and I knew I liked that. But it was more of a working manager role, where I was also hands on doing some work, doing some coding, but had a team that I was kind of breaking that up and having them work on different things, different parts of that. That's kind of one thing. People need to decide, first of all, do you like managing people. There are maybe other leadership tracks at the company, where you really don't need to do that day-to-day people management.
But that's one consideration. Then that level of kind of being in that mix, working managers, doing people management. You're not too far away from the work. That can be something that is a good stepping stone. I kind of went in my career in multiple directions. I’ve done the people management part, and then I moved out of my postdoc to go to Highmark Health and as a data scientist, when they were starting up their data science function and did that for years. Then moved directly in a few levels up to a director role, where the expectation was I was going to be leading people who were leading people.
It’s said that whatever that role is in your company, that I think you need to think about what you enjoy. Not about career progression because there should be career progression on a technical track, where you don't need to step away from the work. But in that role, you really – many companies, it's that level where you're expected not to be hands on doing the work and just the timing of everything. I kind of went from hands on developing the models as a senior data scientist, to working like with others and doing a lot of like meeting with executives.
So it was a mix of both. But needing to immediately stop that and make sure I was building a team of people who could do that and, hopefully, do it better than I ever could and have this mix of skills and building a team of people who could lead and manage the people doing that effectively. So one thing I did, and I think it was helpful, was immediately cut myself off from all data and tools, just did not renew my access, didn't –
[00:22:29] RS: Yeah. Don’t give yourself the option to even like get in the weeds, right?
[00:22:33] CK: Exactly, exactly. I was a little worried that if I did, when things got a little tough, I just might be tempted to go in and do it. That would be both discouraging to my team and leave a gap in this strategic role that I needed to be stepping into. That’s sort of some considerations and some suggestions of how to make that transition successful.
[00:22:55] RS: That is a fantastic piece of advice, and I can – most leaders, I feel like, would be reluctant to sort of relinquish control. But I think that as a team member, I would feel much better about the trust inherent in that if my boss was like, “I don't need to spot check your work. I don't need to be in this tool and make sure that this says what you say it says.”
[00:23:18] CK: Right. Exactly, exactly. The trust is very important, and I have been so lucky to have such incredible team members. Honestly, I don't even know if there would be any value of me going in and spot checking their work, right, especially at some point. Even immediately, like I saw that I was very good at being a data scientist. But I thought, hey, I can really like scale this by helping others be successful and creating opportunities. So the last thing I want to do is spend all my time second guessing the people that I trusted.
[00:23:52] RS: The people who you're trying to give opportunities, right?
[00:23:55] CK: Yeah. I want the best data scientists in the world and have those people on my team or the best managers in the world. I just need to give them the space to be successful.
[00:24:05] RS: Definitely. Thank you for sharing, Curren, because I just feel like, well, that is not specific to like the technical data science and AI role. Like I said, everyone, given a long enough career, I think it was going to be asked. Do they want this? They have to ask themselves that. So that was a really great framework for approaching the question.
This has been a fantastic conversation, Curren. I'm just so pleased you joined me. So at this point, I would just say thank you so much for being here and sharing your expertise on the show.
[00:24:29] CK: No, thank you for having me. It's been a great conversation. I really enjoyed it.
[END OF INTERVIEW]
[00:24:35] RS: How AI Happens is brought to you by Sama. Sama provides accurate data for ambitious AI, specializing in image, video, and sensor data annotation and validation for machine learning algorithms in industries such as transportation, retail, e-commerce, media, medtech, robotics, and agriculture. For more information, head to sama.com.
[END]