The intersection of artificial intelligence and biotechnology holds immense promise for revolutionizing healthcare and pharmaceutical industries. Head of Data Science Ivan Drokin has been working in the AI space for over 15 years now, resolving tasks in both research and engineering. Delving into the nuanced world of AI in biotech and medtech, our conversation illuminated the complexities faced by researchers and practitioners alike. Challenges abound from sourcing high-quality data to navigating regulatory frameworks, these obstacles also pave the way to opportunities for innovation and advancement. When and how did you start studying machine learning, and how did you transition to working in the field of AI? My background is in applied mathematics, and during my studies, I focused mainly on control theory and related areas. However, we had a few elective courses on the current machine learning algorithms at that time, around 2008. I became interested in this topic because it struck a great balance between theory and real-world applications. Specifically, it allowed me to do something tangible in the physical world by implementing the principles of applied mathematics. It was the perfect kind of activity for me, as I've always been oriented towards practical applications. And so began my journey into the world of machine learning. Much of it was self-education. There was neither hype nor "ready-to-go" educational courses, at least not where I lived. So, it involved a lot of self-learning and experiments with various algorithms. My very first job was at a small fintech startup, where we dealt with analyzing financial assets to make decisions on which assets to sell and which to buy. It was my first significant practical experience in applying machine learning. After that, I moved into biotech and joined a company, where they developed pharmaceutical drugs from scratch. There, I mainly focused on drug discovery, drug design, and related areas. For example, how to find a drug that would treat a disease without harming the patient faster than the disease itself. Among other things, I was responsible for implementing deep learning models for small molecules properties prediction and building mathematical models for cell lines optimization. Have you had any "wow" moments while working with AI models? From time to time, yes. I have a good example here related to track research. It was a long track research project of several years, and from time to time, articles would come out that radically redefined the task. Initially, we were all working with CT scans, which are three-dimensional images of the human body, so we started working with it as a three-dimensional picture. It seemed logical, but then the guys came in and said - look, doctors work with this slice by slice, let's try to reproduce the same process. So we started talking more with doctors, learning how they work with it and drawing ideas from there. Sometimes the models themselves exhibit unexpected behavior. Well, starting from extra fingers in the Midjourney output and ending with strange amusing responses from ChatGPT. Sometimes you think, maybe there really is some kind of “consciousness”. Here, of course, not everything is so simple and one should not perceive modern models as something that has consciousness in the human sense. In this regard, I am generally a techno-optimist. I believe that technology, especially based on artificial intelligence, is something that can help humanity survive and become better. It can truly help us solve many problems. Unemployment is not a scary thing because if artificial intelligence replaces us, although there are nuances here, then OK - let's have basic income, let's live, learn, make everything around us better. In the medtech sphere, thanks to AI, we can really scale it globally because it's easier to train lab technicians than doctors. And doctor expertise is such a complex thing in terms of scaling with decades of experience, training, certifications, and so on. We constantly need more doctors. And here AI can help solve this problem. What are some of your accomplishments you are most proud of? How have you implemented them? I am proud to have been a part of innovative companies that have achieved great accomplishments. For example, botkin.ai, where I worked for over five years as Chief Research Officer, received numerous awards in recognition of the work that we have done. , We have developed and implemented over 50 complex projects: from lung cancer screening to assisting in the analysis of clinical trial data. For example, we implemented several retrospective lung cancer screenings on CT scans collected during the COVID-19 pandemic, which allowed us to identify hundreds of new cases suspicious for lung cancer. Without the use of AI, this would have been impossible due to the enormous volume of accumulated images. We also had an impactful case regarding data analysis for clinical trials - the use of our products significantly simplified the data analysis process. And of course, I am proud of braingarden.ai which I co-founded and where I served as Chief Science Officer. The company was dedicated to applying our experience in artificial intelligence to various domain fields, including computer vision manufacturing, smart system assembly, medical image analysis, robotics, and Creative AI. Our cutting-edge projects, which utilized state-of-the-art approaches and technologies such as deep learning with intensive synthetic data usage and domain adaptation, were trusted by industry leaders such as LG and Arrival. Equally important to me is that my published scientific papers have been recognized with Best Paper Awards at international conferences AIST. It is very gratifying to know that your contribution to science is appreciated by the community. I have also had the honor to serve on the program committee of the AIST conference. I personally believe that as intelligent beings, we have a global goal - to spread intelligence throughout the Universe. Therefore, any spread of knowledge is extremely important, from popularizing science to teaching in schools and universities. I am certainly glad that I have had and still have the opportunity to speak at conferences, give popular science lectures, deliver speeches, and teach at some of the best universities. You have invented several patents. Could you talk a bit about them and the path that led you to create those inventions? During my tenure at Botkin.ai, our team of researchers created and patented numerous solutions. One such breakthrough was a novel methodology for analyzing lung CT scans, wherein data is represented not as an image but as a point cloud. This approach boasts several advantages, including heightened accuracy and diminished computational expenses both in training and during CT scan processing in industrial settings. Additionally, a cluster of patents is dedicated to various techniques for modeling and processing patient data, with a primary focus on medical imaging. Our patented technologies have become the cornerstone of the company's product portfolio. AI has been advancing rapidly in recent years, with some public figures like Elon Musk raising concerns about the need to slow down this development. Do you believe any restraints are necessary for the development of AI? This should be considered from multiple perspectives. Let's talk about medicine as one example. To create a good quality model in medicine, you need to find a well-annotated dataset, train the model, and then deploy it. It may sound straightforward, but in reality, it's a complex process that must account for a vast number of corner cases, understand how doctors will interact with the model, and how it will be integrated into existing systems. Regulation is necessary here because we're dealing with matters related to patients' lives. Incorrect diagnosis can lead to serious consequences. Therefore, certification processes and other regulatory measures are crucial. However, I don’t think we should halt our research efforts - AI could revolutionize the entire healthcare market, making it more accessible worldwide. We can't advance medtech as quickly as with AI. I would venture to suggest that these same arguments can be extended to any areas of our lives. Certainly, governments will likely attempt to regulate AI development, especially large language models, as they pose certain risks, ranging from privacy concerns to mass unemployment. However, I'm not among those who believe that the AI apocalypse is imminent. I think we have much further to go before we reach such a scenario. If anything, the real threat to humanity might come from environmental issues resulting from our attempts to train AI models excessively. Can you name three key areas in biotech that AI could directly impact right now, or in the near future? Certainly, let's break down the question into two parts: one concerning pharmaceuticals, which encompasses everything related to drug design – finding, preparing, and delivering pharmaceuticals to patients. This constitutes a significant portion of the budget for pharmaceutical companies, so AI advancements in this area could lead to faster drug development and reduced costs, which would be highly beneficial. Now, moving on to medicine, there's even more potential for AI to make an impact. The emergence of ChatGPT and subsequent language-based AI solutions opens up a crucial avenue. If we examine how medicine operates not in terms of drugs or research, but rather at the operational level of clinics, all interactions occur in natural language. This is critical because when a doctor reads a report on an image, they are reading the report, not analyzing the image itself, as it's deciphered by a radiologist. The fact that we now have powerful language models capable of processing these data flows means we genuinely have the potential to transform industries. This ranges from mundane tasks like quality control in filling out reports or patient medical histories to genuinely robust diagnostic systems that assist doctors in their work. I anticipate that within the next five years, we will see, if not full-scale implementation, at least the beginning of certification processes for systems built around large language models. Something akin to "large language model operations systems for clinics" could yield tremendous benefits. Additionally, there are fascinating developments in other fields of medicine. For instance, generative models that translate text into images, such as Stable Diffusion or Midjourney, could be used to create synthetic datasets for research purposes. This addresses a significant challenge in medicine because real-world data is not abundantly available in public repositories due to stringent laws governing medical data usage and sharing practices in many countries. As a result, synthetic datasets have the potential to democratize research, allowing researchers worldwide to propose and test their models on synthetic data before scaling them to the real world. This represents another significant and positive development. The intersection of artificial intelligence and biotechnology holds immense promise for revolutionizing healthcare and pharmaceutical industries. Head of Data Science Ivan Drokin has been working in the AI space for over 15 years now, resolving tasks in both research and engineering. The intersection of artificial intelligence and biotechnology holds immense promise for revolutionizing healthcare and pharmaceutical industries. Head of Data Science Ivan Drokin has been working in the AI space for over 15 years now, resolving tasks in both research and engineering. Delving into the nuanced world of AI in biotech and medtech, our conversation illuminated the complexities faced by researchers and practitioners alike. Challenges abound from sourcing high-quality data to navigating regulatory frameworks, these obstacles also pave the way to opportunities for innovation and advancement. When and how did you start studying machine learning, and how did you transition to working in the field of AI? When and how did you start studying machine learning, and how did you transition to working in the field of AI? My background is in applied mathematics, and during my studies, I focused mainly on control theory and related areas. However, we had a few elective courses on the current machine learning algorithms at that time, around 2008. I became interested in this topic because it struck a great balance between theory and real-world applications. Specifically, it allowed me to do something tangible in the physical world by implementing the principles of applied mathematics. It was the perfect kind of activity for me, as I've always been oriented towards practical applications. And so began my journey into the world of machine learning. Much of it was self-education. There was neither hype nor "ready-to-go" educational courses, at least not where I lived. So, it involved a lot of self-learning and experiments with various algorithms. My very first job was at a small fintech startup, where we dealt with analyzing financial assets to make decisions on which assets to sell and which to buy. It was my first significant practical experience in applying machine learning. After that, I moved into biotech and joined a company, where they developed pharmaceutical drugs from scratch. There, I mainly focused on drug discovery, drug design, and related areas. For example, how to find a drug that would treat a disease without harming the patient faster than the disease itself. Among other things, I was responsible for implementing deep learning models for small molecules properties prediction and building mathematical models for cell lines optimization. Have you had any "wow" moments while working with AI models? Have you had any "wow" moments while working with AI models? From time to time, yes. I have a good example here related to track research. It was a long track research project of several years, and from time to time, articles would come out that radically redefined the task. Initially, we were all working with CT scans, which are three-dimensional images of the human body, so we started working with it as a three-dimensional picture. It seemed logical, but then the guys came in and said - look, doctors work with this slice by slice, let's try to reproduce the same process. So we started talking more with doctors, learning how they work with it and drawing ideas from there. Sometimes the models themselves exhibit unexpected behavior. Well, starting from extra fingers in the Midjourney output and ending with strange amusing responses from ChatGPT. Sometimes you think, maybe there really is some kind of “consciousness”. Here, of course, not everything is so simple and one should not perceive modern models as something that has consciousness in the human sense. In this regard, I am generally a techno-optimist. I believe that technology, especially based on artificial intelligence, is something that can help humanity survive and become better. It can truly help us solve many problems. Unemployment is not a scary thing because if artificial intelligence replaces us, although there are nuances here, then OK - let's have basic income, let's live, learn, make everything around us better. In the medtech sphere, thanks to AI, we can really scale it globally because it's easier to train lab technicians than doctors. And doctor expertise is such a complex thing in terms of scaling with decades of experience, training, certifications, and so on. We constantly need more doctors. And here AI can help solve this problem. What are some of your accomplishments you are most proud of? How have you implemented them? What are some of your accomplishments you are most proud of? How have you implemented them? I am proud to have been a part of innovative companies that have achieved great accomplishments. For example, botkin.ai, where I worked for over five years as Chief Research Officer, received numerous awards in recognition of the work that we have done. , We have developed and implemented over 50 complex projects: from lung cancer screening to assisting in the analysis of clinical trial data. For example, we implemented several retrospective lung cancer screenings on CT scans collected during the COVID-19 pandemic, which allowed us to identify hundreds of new cases suspicious for lung cancer. Without the use of AI, this would have been impossible due to the enormous volume of accumulated images. We also had an impactful case regarding data analysis for clinical trials - the use of our products significantly simplified the data analysis process. And of course, I am proud of braingarden.ai which I co-founded and where I served as Chief Science Officer. The company was dedicated to applying our experience in artificial intelligence to various domain fields, including computer vision manufacturing, smart system assembly, medical image analysis, robotics, and Creative AI. Our cutting-edge projects, which utilized state-of-the-art approaches and technologies such as deep learning with intensive synthetic data usage and domain adaptation, were trusted by industry leaders such as LG and Arrival. Equally important to me is that my published scientific papers have been recognized with Best Paper Awards at international conferences AIST. It is very gratifying to know that your contribution to science is appreciated by the community. I have also had the honor to serve on the program committee of the AIST conference. I personally believe that as intelligent beings, we have a global goal - to spread intelligence throughout the Universe. Therefore, any spread of knowledge is extremely important, from popularizing science to teaching in schools and universities. I am certainly glad that I have had and still have the opportunity to speak at conferences, give popular science lectures, deliver speeches, and teach at some of the best universities. You have invented several patents. Could you talk a bit about them and the path that led you to create those inventions? You have invented several patents. Could you talk a bit about them and the path that led you to create those inventions? During my tenure at Botkin.ai, our team of researchers created and patented numerous solutions. One such breakthrough was a novel methodology for analyzing lung CT scans, wherein data is represented not as an image but as a point cloud. This approach boasts several advantages, including heightened accuracy and diminished computational expenses both in training and during CT scan processing in industrial settings. Additionally, a cluster of patents is dedicated to various techniques for modeling and processing patient data, with a primary focus on medical imaging. Our patented technologies have become the cornerstone of the company's product portfolio. AI has been advancing rapidly in recent years, with some public figures like Elon Musk raising concerns about the need to slow down this development. Do you believe any restraints are necessary for the development of AI? AI has been advancing rapidly in recent years, with some public figures like Elon Musk raising concerns about the need to slow down this development. Do you believe any restraints are necessary for the development of AI? This should be considered from multiple perspectives. Let's talk about medicine as one example. To create a good quality model in medicine, you need to find a well-annotated dataset, train the model, and then deploy it. It may sound straightforward, but in reality, it's a complex process that must account for a vast number of corner cases, understand how doctors will interact with the model, and how it will be integrated into existing systems. Regulation is necessary here because we're dealing with matters related to patients' lives. Incorrect diagnosis can lead to serious consequences. Therefore, certification processes and other regulatory measures are crucial. However, I don’t think we should halt our research efforts - AI could revolutionize the entire healthcare market, making it more accessible worldwide. We can't advance medtech as quickly as with AI. I would venture to suggest that these same arguments can be extended to any areas of our lives. Certainly, governments will likely attempt to regulate AI development, especially large language models, as they pose certain risks, ranging from privacy concerns to mass unemployment. However, I'm not among those who believe that the AI apocalypse is imminent. I think we have much further to go before we reach such a scenario. If anything, the real threat to humanity might come from environmental issues resulting from our attempts to train AI models excessively. Can you name three key areas in biotech that AI could directly impact right now, or in the near future? Can you name three key areas in biotech that AI could directly impact right now, or in the near future? Certainly, let's break down the question into two parts: one concerning pharmaceuticals, which encompasses everything related to drug design – finding, preparing, and delivering pharmaceuticals to patients. This constitutes a significant portion of the budget for pharmaceutical companies, so AI advancements in this area could lead to faster drug development and reduced costs, which would be highly beneficial. Now, moving on to medicine, there's even more potential for AI to make an impact. The emergence of ChatGPT and subsequent language-based AI solutions opens up a crucial avenue. If we examine how medicine operates not in terms of drugs or research, but rather at the operational level of clinics, all interactions occur in natural language. This is critical because when a doctor reads a report on an image, they are reading the report, not analyzing the image itself, as it's deciphered by a radiologist. The fact that we now have powerful language models capable of processing these data flows means we genuinely have the potential to transform industries. This ranges from mundane tasks like quality control in filling out reports or patient medical histories to genuinely robust diagnostic systems that assist doctors in their work. I anticipate that within the next five years, we will see, if not full-scale implementation, at least the beginning of certification processes for systems built around large language models. Something akin to "large language model operations systems for clinics" could yield tremendous benefits. Additionally, there are fascinating developments in other fields of medicine. For instance, generative models that translate text into images, such as Stable Diffusion or Midjourney, could be used to create synthetic datasets for research purposes. This addresses a significant challenge in medicine because real-world data is not abundantly available in public repositories due to stringent laws governing medical data usage and sharing practices in many countries. As a result, synthetic datasets have the potential to democratize research, allowing researchers worldwide to propose and test their models on synthetic data before scaling them to the real world. This represents another significant and positive development.
