When AI Meets Everything: Google Research on the Transformative Power of AI at the Intersection of Disciplines

The True Potential of AI Lives at the Intersection of Disciplines

At MIT Technology Review's EmTech Future 2026 event, Yossi Matias, Vice President and Head of Google Research, put forward a thought-provoking idea: the most far-reaching impact of artificial intelligence will not come from breakthroughs within any single field, but from its convergence with other disciplines — biology, infrastructure, manufacturing, and science.

This perspective is reshaping how we understand the value of AI. Over the past few years, public attention has largely focused on visible applications like chatbots and image generation. But in Matias's view, these are just the tip of the iceberg. When AI's capacity for reasoning and learning is embedded into the foundational layers of scientific research and industrial processes, it begins to release the energy needed to change how the world operates.

Why Interdisciplinary Intersections Are the Key

AI is fundamentally a general-purpose capability, not a specialized tool for any particular industry. It excels at identifying patterns in massive datasets, generating hypotheses, and accelerating trial-and-error cycles. Once these capabilities are combined with domain-specific expertise, they can produce leap-forward breakthroughs in fields that have traditionally moved slowly.

In other words, the marginal value of AI varies enormously across different contexts. In a domain that is already highly digitized and rule-driven, AI can deliver efficiency gains. But in fields like biology and materials science — where unknowns abound and experimental costs are high — AI has the potential to radically compress the cycle from hypothesis to validation.

How AI Is Reshaping Scientific Research

Matias placed particular emphasis on AI's transformation of scientific research itself. Traditional science relies on researchers' intuition, experience, and large volumes of repeated experiments. AI, by contrast, can simulate and predict at the computational level, dramatically shortening the path of exploration. Protein structure prediction is one of the most compelling examples — it compressed experimental work that once took years into a matter of hours of computation.

This change goes beyond speed. When AI can handle high-dimensional complex relationships that humans struggle to grasp, it effectively expands the boundaries of the questions science can ask. Researchers are able to explore territories that were previously inaccessible due to computational or experimental constraints.

From Biology to Infrastructure

Matias outlined several areas where AI is making inroads:

• Biology: Accelerating drug discovery and deepening our understanding of complex biological systems
• Infrastructure: Optimizing the operational efficiency of large-scale systems like energy and transportation
• Manufacturing: Improving design workflows and elevating the intelligence of production processes
• Scientific Research: Acting as a universal accelerator, boosting discovery rates across disciplines

What these areas share in common is this: AI does not replace the domain experts — it becomes an amplifier in their hands. Domain expertise defines the problem and judges what matters; AI efficiently searches through the space of possibilities.

---

The breakthrough case in protein structure prediction is AlphaFold2, developed by DeepMind. A protein's three-dimensional structure determines its biological function, and resolving that structure through experimental methods — such as X-ray crystallography or cryo-electron microscopy — can take months or even years. AlphaFold2 dramatically outperformed traditional methods at the CASP competition in 2020, and its prediction database has since covered over 200 million protein structures, encompassing nearly every known protein type on Earth. The significance of this achievement lies not only in accelerating the discovery of drug targets, but in enabling researchers to "pre-experiment" computationally — screening candidate directions worth pursuing in wet labs at a fraction of the cost. The AlphaFold paradigm demonstrated that AI in scientific research is not merely a supporting tool, but can directly produce results with both academic and industrial value. It stands as the most compelling illustration of what Matias means when he says AI expands the boundaries of scientific inquiry.

---

Implications for the Industry

For practitioners who follow AI development, this perspective offers an important strategic signal: the real opportunities may not lie in building yet another general-purpose AI tool, but in deeply embedding AI capabilities into a specific, high-value vertical domain.

Whoever combines deep domain expertise with the ability to effectively integrate AI capabilities is best positioned to capture the compounding dividends AI can deliver. This also explains why institutions like Google Research are placing increasing emphasis on cross-disciplinary collaboration and real-world application, rather than simply racing to scale model parameters.

From the perspective of the MIT Technology Review editorial team, this keynote from Google Research leadership reads less like a product pitch and more like a recalibration of coordinates — a reminder to the entire industry to shift attention from "what can AI do" to "what can AI do when combined with something else."

---

What is referred to here as "deep embedding in vertical domains" corresponds in industry practice to the path of Domain Foundation Models or Science AI. Unlike general large language models, domain models are trained on massive volumes of specialized data from specific disciplines — such as genomic sequences, molecular structures, crystal material data, or medical imaging — and incorporate domain constraints and prior knowledge, making their outputs more reliable and interpretable in professional contexts. Notable examples include Meta's ESM series of protein language models, GraphCast (designed for climate and weather prediction), and molecular generation models in the pharmaceutical space. The competitive moat for these models lies not in parameter count, but in the accumulation of high-quality domain data, long-term collaboration with domain experts, and the engineering capability to integrate model outputs into actual research or production workflows. This is precisely why Matias insists that deep domain knowledge and the ability to integrate AI capabilities are both indispensable.

---

Closing Thoughts

Yossi Matias's core argument is clear and compelling: the next major phase of AI is not about how powerful AI becomes on its own, but about how deeply it can fuse with existing human knowledge systems. When AI truly "meets everything," it will evolve from a remarkable technology into a foundational force reshaping the landscape of biology, industry, and science.

This is a trend worth watching closely — the most exciting AI advances over the next few years may well emerge at the disciplinary boundaries we haven't yet paid enough attention to.