[Hoban Science Bridge] Town Hall Meeting Panel Discussion - Connecting Generations in Science
Date: Thursday, March 26, 2026, 2:50–4:00 PM | Venue: Dynasty Hall, Seoul Shilla Hotel
Q1 (MC)
Question
I’d like to ask you, Sunghee — recently, in fields like AI and biotech, the boundary between scientific research and industry is becoming increasingly blurred. From the perspective of the industry floor, what roles are science and technology talents playing — and what opportunities do you see expanding in the future?
Sunghee’s Answer
I’ve worked across quite different industry landscapes — Samsung Semiconductor, Amazon AI, Gauss Labs (SK Group’s first dedicated AI company), and now Erudio Bio. Through all of that, I want to say something that I think is important and often overlooked: industry needs both types of talent — people with depth AND breadth, and people with extraordinary depth alone. Both are essential. Which one you are is really a matter of your own character and DNA, not a hierarchy of value.
Let me explain what I mean. My own path has been one of depth-and-breadth — I did my PhD at Stanford in Convex Optimization, and that same mathematics turned out to be directly applicable at Samsung (DRAM yield optimization), at Amazon (anomaly detection and recommender systems, driving over $200M in revenue), and now at Erudio Bio (searching for energy minima in protein conformation landscapes). On the surface these look like completely different fields — but the mathematical skeleton is identical. I call this “cross-domain inevitabilities” — mathematical and physical principles that recur across seemingly unrelated domains. This breadth has allowed me to play leadership and C-level roles across multiple industries.
But — and this is equally important — the person who goes extraordinarily deep in one domain is just as indispensable. The researcher who spends fifteen years understanding a single protein family and becomes the world’s foremost authority on it — that person is irreplaceable. No generalist can substitute for genuine world-class depth. At Erudio Bio, my co-founders and scientific collaborators bring exactly that kind of depth in molecular biology and clinical medicine. Without them, the breadth I bring has nothing to stand on. The two types don’t compete — they depend on each other.
So what I’d say to the students here is: don’t force yourself into a shape that isn’t yours. If you are someone who loves going deeper and deeper into one problem until you understand it more completely than anyone else on the planet — that is a superpower. Own it. If you are someone who gets restless after mastering one domain and feels the pull toward the next frontier — that is also a superpower. Own that too. The worst outcome is trying to imitate the other type because you think it’s what’s expected.
Now, regardless of which type you are — deep specialist or cross-domain generalist — there is one thing that applies equally to everyone in this room, and I want to say it clearly: you must embrace AI tools, especially LLM-powered multimodal AI, as a power user. And I want to be precise about why, because the reason is often stated incorrectly.
It’s not that you have to use AI because someone told you to, or because it’s trendy. It’s much simpler and more urgent than that: when your colleagues are using these tools and you are not, you fall behind — not gradually, but rapidly. A researcher who uses AI to synthesize literature, generate hypotheses, write and debug code, and iterate on experimental designs is operating at a fundamentally different speed than one who doesn’t. That gap compounds every single day.
I am not saying you need to become an AI expert. I am not saying you need to understand transformer architecture or write your own training loops. What I am saying is: become a power user. Learn to prompt well. Learn which tools amplify your specific type of work. Learn to critically evaluate AI outputs rather than accepting them blindly. Think of it the way a surgeon thinks about a new surgical instrument — you don’t need to have built it, but you absolutely need to know how to wield it with precision.
The scientists who will have the most impact in the next decade — whether they are deep specialists or cross-domain generalists — will be the ones who combine their irreplaceable human expertise with fluent, unselfconscious use of these new tools. The depth you bring is yours and no AI can replicate it. But the leverage you apply to that depth? AI can multiply it enormously — if you let it.
And the blurring you mention — between research and industry — is not just accelerating, it’s becoming structural. The companies at the frontier of AI-biotech are not merely “applying” research results from elsewhere; they are doing the research themselves, in direct contact with clinical reality. This means both the deep specialist and the cross-domain generalist have a home in industry in ways that weren’t true twenty years ago.
Looking ahead, I see three major areas where opportunities will expand dramatically — for both types of talent:
First, AI-biotech convergence. AI accelerating drug discovery, revolutionizing cancer diagnostics, enabling personalized medicine — this is already reality, not future speculation. At Erudio Bio, what we’re building — a cancer biomarker diagnostic platform combining dynamic force spectroscopy with Artificial Intelligence (AI) — is right at that frontier. We have support from the Gates Foundation and a partnership with Seoul National University Bundang Hospital (SNUBH).
Second, AI semiconductors and edge AI. As AI models grow larger, the demand for people who understand both AI and semiconductor hardware — energy efficiency, compute optimization, NPU architecture — is exploding. People who can bridge those two worlds are extraordinarily rare and extraordinarily valuable.
Third, privacy-preserving AI. How do you analyze sensitive medical or financial data with AI without exposing the raw data? Federated learning, homomorphic encryption — these technologies have enormous societal demand, and the regulatory tailwinds are only strengthening.
The opportunities are abundant. What matters is becoming someone who knows how to connect their scientific foundation to adjacent domains — and isn’t afraid to make that leap.
Q2 (MC)
Question
Sunghee, I have one more question. To genuinely create better experiences for science talents — in education and on the industry floor — what kinds of programs or environments do you think are needed? And please also share what you believe the most critical gap is right now.
Sunghee’s Answer
I’ll be direct. The thing I find most lacking is “a space where it’s okay to fail.”
When I co-founded Gauss Labs, I was a semiconductor AI expert — not a founder. I didn’t know how to raise investment, how to build a team from scratch, how to find product-market fit. I learned all of it by colliding with reality. But if I had treated every one of those collisions as a defeat rather than a lesson, I never would have started.
Let me be specific about what I find inadequate in Korea’s current science education environment:
First, the connection between industry and academia is too thin. Internships and joint projects exist, but most are superficial. At Amazon, student interns were handed real projects — real scale, real stakes, real consequences. That’s what genuine experience looks like. A program where a student writes a report that no one acts on is not experience; it’s a checkbox.
Second, multidisciplinary exposure is too narrow. Korea’s academic structure still has high walls between departments. The fact that someone like me — an electrical engineering PhD — moving into biotech looks “unusual” is itself a structural problem. The people who cross those boundaries are precisely the ones who create the most impactful breakthroughs.
Third, global networking opportunities are insufficient. One reason I lead K-PAI (the Silicon Valley Privacy-Preserving AI Forum) is exactly this — Korea needs real, substantive bridges to global innovation ecosystems, not just occasional conference appearances.
And one more thing I want to add: I believe that curiosity about science and genuine interest in people are equally important. I describe myself as “above all, a people person.” Science ultimately exists to serve people — so cultivating scientists who deeply understand and empathize with people is not a soft add-on. It’s a core requirement.
Q3 (Kim Ji-min, Ghent University student)
Question
I believe the definition of “science talent” needs to be broader than how students currently perceive their career options. Relating to the earlier discussion about the scope of “science talent,” I’d like to direct this question to both Sunghee and Professor Kang. I believe research, industry, and entrepreneurship are all areas where science talents can thrive — but many students, myself included, still tend to perceive science talent as limited to research and publication-centered paths. To move beyond this perception and make the paths toward industry and entrepreneurship genuinely visible to students — so they can more actively pursue those routes — what do you think needs to change?
Sunghee’s Answer
This question gets exactly at the right thing. And I want to first affirm what you said at the beginning — the definition of “science talent” absolutely needs to be broader. Research and publications are one expression of scientific talent. But so is building a company that brings a new diagnostic to cancer patients. So is leading an AI team at a technology company. So is advising a government on semiconductor policy. All of these require the same underlying scientific foundation.
The word “visible” in your question is the key. It’s not enough for the paths to exist — they need to be seen.
I have a memory from when I was in 9th grade — I was playing piano and suddenly the structure of the music started to sound different to me. The same music I had always played, but one day it became visible in a new way. Scientific career paths are the same — they need to be seen to become real options.
Three things need to change to make those paths visible:
First, diversify the role models. Right now, the dominant image of a “successful science person” in students’ minds is either a professor or a researcher at a large conglomerate. But paths like mine exist too — starting as a Convex Optimization theorist and mathematician, becoming a semiconductor software developer and optimization engineer, leading a $200M Amazon AI project, co-founding an AI company, then pivoting to biotech with Gates Foundation backing. These non-linear paths need to be in front of students regularly, not as exotic exceptions but as legitimate options.
Second, make entrepreneurship a first choice, not a fallback. In Korea, when an engineering graduate chooses to start a company, they sometimes still face the implicit question: “Couldn’t get a job?” In Silicon Valley, it’s the opposite — if someone from Stanford or MIT chooses to start a company, that signals they’re serious. Culturally and institutionally, entrepreneurship needs to be reframed as a primary destination.
Third, institutionalize the “middle steps.” Between university research labs and large company internships, there is a vast space of experiences that should be structured and accessible: startup internships, VC research roles, open-source project contributions, international conference participation. Students can only discover their right path if they’ve had the chance to walk several different ones.
Q4 (Lee Ga-young, Neungdong High School student)
Question
I’ve had a strong interest in AI-driven drug discovery for a long time and have been following developments in this area closely. I’d like to direct this question to Sunghee. While AI-driven drug discovery — like AlphaFold — is advancing at a remarkable pace, I also think there is a real possibility that AI models make incorrect predictions or produce errors due to data bias. How are errors in AI models actually verified and compensated for in real drug development processes — and what is your outlook for AI drug discovery going forward?
Sunghee’s Answer
First, I want to say — the fact that you as a high school student are following AI drug discovery closely and thinking critically about its limitations is exactly the mindset that the next generation of scientists needs. This is genuinely one of the most important and least candidly discussed questions in the field.
Let me separate what AI does well from what it doesn’t.
What AI is good at: Protein structure prediction (AlphaFold 3), rapidly filtering millions of molecular candidates down to a promising few, applying known patterns to new data. Have you heard of Insilico Medicine’s INS018_055, now called rentosertib? AI identified a completely new therapeutic target for IPF (idiopathic pulmonary fibrosis, a devastating rare lung disease), designed the molecular structure from scratch, and the drug reached Phase II clinical trials in under 30 months. The traditional industry average for that same journey is 10–15 years. The Phase IIa results were published in Nature Medicine in 2025.
What AI is not good at: AI is vulnerable to situations outside its training data. The complex human responses that emerge in clinical trials — individual genetic variation, long-term side effects, drug-drug interactions — are still very difficult for AI to learn adequately. And data bias is a serious structural problem: most clinical data has historically been collected from Western, male, specific age-range populations. An AI model trained on that data may systematically produce errors for underrepresented populations.
How do we verify and compensate? At Erudio Bio, AI predictions must always be validated against real experimental data. AI generates the hypothesis; wet lab experiments either confirm or refute it. This iterative cycle is the core discipline. We also continuously update our models with clinical data from hospital partnerships like SNUBH — not as a one-time calibration, but as an ongoing feedback loop. That is partly why hospital partnerships are so strategically important for an AI-native biotech company.
And here is something concrete that illustrates exactly why verification matters — and why the idea of a single universal AI solution is a myth. At Erudio Bio, we have found that for cyclic peptides specifically, our own in-house AI algorithm significantly outperforms AlphaFold. AlphaFold is a remarkable achievement, and I say this with full respect for what DeepMind built — but it was designed and trained with certain molecular classes in mind, and cyclic peptides are a domain where a purpose-built, domain-specific model beats the general-purpose one. This is not a criticism of AlphaFold. It is simply the reality that one solution does not fit all problems — and anyone who tells you otherwise is either selling something or hasn’t gone deep enough into the actual biology. Which brings me to something I cannot overemphasize: domain expertise is absolutely irreplaceable. You can have the most sophisticated AI model in the world, but if you don’t understand the underlying biology — the structure of the molecules, the mechanisms of disease, the clinical context — you will not know when your model is right, when it is wrong, or even what question to ask it in the first place. AI amplifies expertise. It does not substitute for it.
My outlook: cautiously optimistic. But “AI will solve everything” is a dangerous overestimate. AI is a scientist’s tool — not a scientist’s replacement. The scientists who understand both the power and the limits of their tools will be the ones who actually deliver.
Q5 (Kim Ha-rang, Daejeon Science High School student)
Question
I recently read an article about Erudio Bio, and I was very impressed by the technology that introduced concepts from the semiconductor industry into a drug discovery platform. At a time when the importance of interdisciplinary research is increasingly recognized, I believe your example — founding a biotech startup as a semiconductor expert — will be a tremendous source of inspiration for the students here. What was it that led you, someone who had been deeply immersed in semiconductor research, to become interested in drug development? And what do you think is absolutely essential for conducting interdisciplinary research across fields with completely different ways of thinking?
Sunghee’s Answer
Thank you for reading about Erudio Bio — and yes, the connection between semiconductor concepts and drug discovery is real, not metaphorical. Let me explain both parts of your question.
What led me from semiconductors to drug development?
There was no single moment of decision. I followed mathematics deeper and deeper, and the path led me here. Specifically: my Stanford PhD was about Convex Optimization theory and its application to various electrical engineering (EE) areas; I think here the application part was crucial, not just theory, but my intended efforts to find exciting applications! At Samsung, I applied that mathematics to DRAM manufacturing process optimization — maximizing yield, minimizing defects. Then one day I was reading a paper and realized that protein folding — the way a protein collapses into its three-dimensional structure — is fundamentally an energy minimization problem over a high-dimensional conformational space. The same mathematics I had been using to optimize semiconductor yield was structurally identical to one of biology’s deepest problems. I call this a “cross-domain inevitability” — a mathematical principle operating underneath what looks like completely different fields.
That realization deepened at Gauss Labs, and eventually led to founding Erudio Bio — where we now apply precision measurement concepts from semiconductor inspection to cancer diagnostics. The core technology in our VSA (Versatile Smart Assay) platform, dynamic force spectroscopy, measures forces at the nanoscale — a concept with direct analogues in semiconductor metrology equipment. So when that student said they were impressed by how semiconductor concepts were introduced into drug discovery — that’s literally what happened, and it came from recognizing the underlying mathematical connection.
But I want to go deeper than the mathematics for a moment — because what truly drove me toward biotech was not just a technical insight. It was understanding a durable truth about the human condition itself. I wrote about this extensively in my blog, and I think it’s the most important framing for why the bio-medical industry will be the defining industry of our generation. For hundreds of thousands of years, Homo sapiens evolved under brutal selective pressure — famine, disease, violence, early death. Our cellular machinery was optimized for lifespans of 30 to 40 years. Then, in just over a century — a blink in evolutionary time — we eliminated famine through chemistry, conquered infectious disease through medicine, and built societies where violence became the exception. The result is that humans now routinely live 80, 90, even 100 years — far beyond what our biology was ever designed to support. Every cell in our body carries a probability of developing cancer that was negligible when lifespans were short. Degenerative diseases like Alzheimer’s and Parkinson’s were not significant evolutionary pressures because most humans simply didn’t live long enough to develop them. This is the evolutionary mismatch that defines our era — and it will not resolve itself. The durable truth that follows is this: humans will always need solutions to live healthily throughout lifespans that far exceed what their biology was designed to support. Unlike most industry trends, this need is universal, inevitable, urgent, and permanent.
The second thing that drew me here — and that I believe students considering this field should deeply understand — is the paradigm shift from reactive medicine to life-long health asset management. We readily accept life-long financial planning: from your first job, you think about investment, diversification, compound growth over decades. Yet when it comes to health — our most valuable and irreplaceable asset — we operate almost entirely in reactive mode. We see doctors when sick, address symptoms after they appear, and often only begin serious health optimization after problems have already developed. This reactive model made sense when lifespans were short and health challenges were acute. In an era of 80-plus year lifespans, it is fundamentally insufficient. The companies that recognize this paradigm shift — that build platforms for proactive, continuous, personalized health optimization the way financial institutions built platforms for wealth management — will be the defining companies of the next several decades. And crucially, this must be done with human flourishing, equity, and access at the center — not just commercial return. AI can accelerate biology, but it cannot experience the profound satisfaction of helping a patient recover, or the meaning that comes from dedicating one’s life to reducing human suffering. These uniquely human dimensions must remain at the center of everything we build.
What is absolutely essential for interdisciplinary research across completely different ways of thinking?
Honestly, when I reflect on my own journey, it wasn’t just “two things.” It was a genuinely holistic view — one that spans mathematics, markets, industry dynamics, human history, evolutionary biology, and a deep sense of where humanity is going. Let me try to articulate what that actually means in practice.
Mathematical foundations remain essential — they are the universal language. Optimization, probability theory, linear algebra — these give you a common skeleton that you can recognize in biology, semiconductors, economics, and medicine alike. This foundation gave me the confidence to move into new domains without fear, because however unfamiliar the surface, the underlying structure felt like home.
Curiosity paired with genuine humility is equally non-negotiable. When you enter a new field, you must genuinely acknowledge what you don’t know and be willing to learn from the experts already there. When I started Erudio Bio, I learned enormously from biologists and clinicians. “I know the mathematics well but I don’t know the biology — please teach me” was the beginning of every meaningful collaboration. The arrogance of thinking your existing framework automatically transfers — without learning the new domain’s language and intuitions — is the most common failure mode in interdisciplinary work.
But beyond those two: you need to understand markets, industry evolution, and the long arc of history. The reason I could see the opportunity in AI-biotech convergence is not just that I understood the mathematics of protein folding — it’s that I understood the demographic trajectory of global aging, the durable human need for healthspan optimization, the market readiness, the regulatory environment, and the moment in the AI capability curve when the tools finally became sufficient. That is a holistic view — spanning evolutionary biology, economic history, technology S-curves, and social trends simultaneously. Without that view, a technically brilliant person can build something extraordinary that solves the wrong problem, at the wrong time, for a market that doesn’t yet exist.
And finally: always keep the human at the center. Technology is a means, not an end. The ultimate optimization problem is not maximizing lifespan or even healthspan — it is optimizing for human flourishing in all its dimensions: physical health, mental wellbeing, social connection, creative expression, and the search for meaning. The interdisciplinary researcher who loses sight of this — who gets so absorbed in the technical elegance of the solution that they forget the human being it’s meant to serve — is the one who ultimately builds something the world doesn’t need.
And one thing I’ll add, regardless of which type you are: don’t be afraid of the discomfort of being the least knowledgeable person in the room. Every time I have entered a completely new domain, that is exactly what I was at first. That discomfort turns out to be the fastest growth environment you will ever experience.
Q6 (the town hall meeting organizers)
Question
You framed the challenge of science talent not as a matter of “cultivation” but of “attraction” — can you explain that distinction? And what do you believe is the genuine motivational force that leads students to choose the path of science?
Sunghee’s Answer
This is the central question of today’s session, and I think the distinction between these two words contains a profound insight.
Cultivation is something done to someone from the outside. Build the curriculum, offer the scholarships, develop the infrastructure — all important, but none of it, on its own, causes a student to choose science as their path.
Attraction is something that pulls from the inside. It’s creating the conditions in which a student arrives at the feeling: “this is my path.”
Looking at my own experience: I didn’t choose this path because someone told me to study hard. I was drawn in at the moment I witnessed — directly, viscerally — a mathematical structure explaining something astonishingly complex about the real world. The same thing happened when I was playing piano in 9th grade and suddenly the music revealed itself as something structured and mathematical underneath. That was a moment of attraction, not cultivation.
How do you create that moment of attraction? Based on my experience, three conditions need to align:
First: show science solving real problems. Not formulas in a textbook — but science as the technology keeping a cancer patient alive, as the AI model analyzing climate data, as the robotics system restoring mobility to someone who lost it. Science needs to be shown in contact with the real world.
Second: give students the sense that they can contribute. Not as spectators but as participants. Even the smallest project — “I worked on this problem and my contribution mattered” — plants the seed of attraction. Observation alone is insufficient.
Third: make the scientists themselves appear compelling. Science is ultimately done by people. If the scientists students encounter seem interesting, diverse, and humanly engaging — not just technically formidable but genuinely alive — that is a powerful pull. A gathering like today, where students sit alongside researchers and industry practitioners and speak as near-equals, is already performing this function.
Let me close with the thought I submitted as a comment for this forum’s press release: I believe that the most sustainable and impactful scientists emerge when genuine scientific curiosity and deep care for people come together. Science must ultimately be science for people — not science for its own sake. Students who discover their own role within that larger picture are the ones who will not lose their way later — who will walk the path of a scientist with their own standards, their own compass, and their own reasons.
Prepared by: Sunghee Yun | Erudio Bio, Inc. & Erudio Bio Korea, Inc. & K-PAI | March 2026