AI Enters the Chemical Lab: First, Help Chemists Cut Out Unproductive Experiments

R&D in the specialty chemicals sector has long faced persistent pain points: complex formulations, long experimental cycles, strict regulatory requirements, and developing a functional material in the lab is only the very first step, followed by three core hurdles to clear: production scaling, performance alignment, and cost control. While AI is often billed as a disruptive innovation in this space, its real-world value is far more down-to-earth: it helps researchers run far fewer wasted, invalid experiments by calculating more variables upfront in advance.

In a joint feature produced by MIT Technology Review Insights and Infosys, Girish Rao, Vice President of IT at Lubrizol, shares how this specialty chemical manufacturer puts this approach into practice.

First Lay the Digital Foundation

Rao notes that Lubrizol is building a full digital backbone that connects business workflows, modern data platforms and cloud infrastructure to support its R&D teams. The goal is concrete and targeted: combine decades of accumulated technical data with the hands-on expertise of chemists and engineers, who hold deep understanding of customer demands and real-world physical constraints. Data scientists are tasked with identifying patterns, testing hypotheses, and modeling all feasible possibilities before any physical experiment is run, so researchers can then make informed decisions on which experiments to conduct.

Rao emphasizes that humans retain full ownership of all scientific judgment, while AI’s role is to lay out far more potential possibilities upfront. This positioning is critical. The most valuable resource in a lab is rarely computing power — it is the researchers’ time and labor.

Integrated Teams: Build Trust With Small Pilots

Chemists focus on experiment validity, data scientists prioritize model performance, and IT teams center platform security and scalability. The three groups approach problems from very different perspectives. Lubrizol brings them under shared goals through cross-functional "fusion teams", starting with small pilot projects. As Rao puts it, once team members build trust with each other and align on exactly what value they are meant to deliver, operations run far more smoothly.

Cloud infrastructure also plays a highly practical role here: teams can access on-demand computing power and modern tools without having to first build a capital-heavy on-premise data center. For developer audiences, this aligns closely with platform engineering thinking: you start by building the smallest viable closed-loop workflow, then scale out gradually.

Sustainable Materials: More Pre-Simulation, Fewer Physical Trials

Digital modeling helps researchers evaluate far more formulations, explore starch-based and corn-based alternative materials, and map out tradeoffs across a product’s full lifecycle. Producing a material that works in the lab is just the starting point: scaled production has to meet performance, regulatory and financial constraints all at once. AI can model multiple variables in parallel, helping teams find the sweet spot between these often conflicting objectives. Do not treat AI as an automated chemist: it acts more like a simulation sandbox that exposes all potential constraints upfront.

The Future Lies in Integration, Not Standalone Point Solutions

Rao expects AI, automation, advanced modeling, digital twins and connected data to gradually converge over time. Labs will become smarter, and manufacturing plants will grow more efficient. The core opportunity lies in scaling and combining all these capabilities in a responsible manner. No single technology can deliver meaningful results on its own: real transformative value comes from the massive volumes of data generated once end-to-end workflows are fully interconnected.

This content is produced by MIT Technology Review Insights in partnership with Infosys. It is sponsored custom content, not independent editorial reporting from MIT Technology Review’s newsroom.