Dominic Jainy brings a sophisticated perspective to the intersection of artificial intelligence and the physical sciences. With a career spanning machine learning and blockchain, he understands that the next frontier of innovation isn’t just about digital chatbots but about making the tangible world programmable. Today, we explore how AI is shifting from “reading” biology to “writing” it, and why the real competitive advantage lies not in the algorithms themselves, but in the proprietary data loops that fuel them.
How is the transition from digital chatbots to “physical AI” redefining our approach to biological sciences and laboratory work?
For half a century, we could only observe and read the biological world, but we have finally crossed the threshold where we can actually write it. By treating the genome as a language with its own four-letter alphabet, researchers are using transformer architectures—the same tech behind conversational bots—to autocomplete genetic sequences. It is a profound shift from the days when AlphaFold first made proteins predictable, leading to the 2024 Nobel Prize, to today’s reality where models like Evo 2 are treating DNA as programmable text. However, we must remember that even the most sophisticated DNA model currently returns about 99% useless output, meaning the software is just the starting line, not the finish.
Many organizations believe that acquiring the most advanced AI model is the key to a lasting competitive advantage; why do you argue that this “model moat” is actually quite brittle?
The idea that owning a foundation model provides a lasting defense is a misconception because these tools are rapidly becoming strategic commodities. Training costs are plummeting and open-source alternatives are flooding the market, which effectively evaporates the “moat” that companies once seemed to have. If everyone is renting the same powerful base model from a few providers, then the algorithm itself offers no unique advantage. To truly win, a company has to look beyond the code and focus on the one thing that competitors cannot simply download or replicate: a specialized, proprietary workflow fueled by unique data.
In the context of the “Oracle” prediction model and wet labs, how does a proprietary data flywheel create a superior system compared to one based on public data?
The real magic happens within the loop, where a prediction model proposes candidate genetic switches that are then physically tested in a wet lab. Imagine a single batch where up to 250,000 of these switches are tested, each one carefully tagged with a DNA barcode just like a product you’d see at a grocery store checkout. These rare, successful outliers are then fed back into the model, creating a data flywheel that grows smarter with every rotation of the wheel. This process replaces internet-scraped data with the grit and reality of pipettes and mice, ensuring the “world model” is grounded in physical truth rather than digital echoes.
While AI provides a powerful new set of tools, why remains human intuition and judgment indispensable when tackling complex challenges like curing cancer?
Having AI in your toolkit is like suddenly possessing a bazooka; it is incredibly powerful, but it doesn’t tell you where to aim or when to pull the trigger. If we relied on pure computational brute force to solve biological puzzles, the process would likely still be running a hundred years from now without reaching a conclusion. Instead, the most successful strategies combine human judgment with software to navigate these complex problems responsibly. This human-in-the-loop approach ensures that we aren’t just generating data for the sake of it, but are instead aiming our most powerful tools at the most critical challenges with precision.
As we begin to program physical reality, what lessons can other industries learn from the rigorous regulatory and safety standards inherent in healthcare?
In an era where physical reality is becoming programmable, the strict discipline of the healthcare sector—often viewed as a hurdle—is actually its greatest asset. We are seeing a move toward Phase 1 human trials through rigorous monkey safety studies and FDA reviews because, in this field, “seems right” is a dangerous standard that we simply cannot accept. This culture of safety-first validation and human accountability is a framework that every other industry, from manufacturing to logistics, will eventually need to borrow. The model might be the easiest piece to acquire, but the true game is won through the responsibility of closing the loop and ensuring real-world validation.
What is your forecast for the evolution of physical AI in medicine and beyond?
I predict that we are moving toward a future where the distinction between “digital” and “physical” discovery completely vanishes, leading to a world where we don’t just treat diseases, but reprogram the very logic of biology. As these data flywheels spin faster, we will see the 99% failure rate of current genetic models drop significantly, turning DNA into a truly reliable canvas for innovation. Eventually, the companies that thrive won’t be those with the biggest computers, but those that have mastered the physical lab-to-model loop, transforming how we manufacture everything from life-saving cures to sustainable materials.