Physics-Informed AI Can Run Industrial Plants Where Chatbots Fail
Chatbots are great at writing emails. But they cannot safely run a chemical plant. That is the simple idea behind a new push for physics-informed AI. This kind of AI learns the real laws of science, not just the patterns of human text. Engineering firm KBR and AI company Applied Computing say their new platform proves the point. It is built to optimise plants, lift output, and cut harmful gases — jobs a normal chatbot was never made to do.
Here is the problem in plain words. A large language model (LLM) is the tech behind tools like ChatGPT. It guesses the next word based on huge amounts of text. That works for chatting. It does not work for safety-critical choices in a refinery or an ammonia factory. One wrong number there can cause a fire or a leak. So the makers built a different kind of system.
What is physics-informed AI?
Physics-informed AI is AI that is taught the rules of physics and chemistry. Think of pressure, heat, and how gases react. These rules are baked into the model. So the AI can reason about new, unseen situations the way a trained engineer does. A plain chatbot only knows what it read in text. It has no built-in sense of how a real machine behaves.
Dan Jeavons, President of Applied Computing, summed up the gap. He said AI can be about 80% useful in everyday life, but that level does not work in heavy industry. In a plant, “mostly right” is not safe enough. Andy Webster, a Senior Director at KBR, was even blunter. He said plain language prediction is useless for engineers making safety-critical decisions.
The platform: INSITE 3.0 and Orbital AI
The new platform is called INSITE 3.0. It runs on Applied Computing’s “Orbital AI” model. The model holds core energy and process science. KBR adds decades of real engineering know-how on top. Together they aim to fine-tune how a plant runs, squeeze more product from each batch, and lower emissions.
One smart trick solves a data problem. A well-run plant produces very steady, repetitive data. That is good for the plant but poor for training AI, because the AI rarely sees rare events. So the team mixed real plant data with engineering simulations. Simulations create safe “what if” cases — like odd faults — that almost never happen in real life. This fills the gaps in the training data.
Why small gains matter so much
In big industry, tiny improvements are a huge deal. The team noted that even a 1%, 2%, or 3% jump in efficiency can translate into very large cuts in carbon dioxide (CO2), the main gas linked to climate change. At the scale of a refinery, a 1% saving repeated all year adds up fast. The first paying customer is in the ammonia sector. Ammonia is a key chemical used to make fertiliser.
Key facts
| Item | Detail |
|---|---|
| Platform | INSITE 3.0, powered by Orbital AI |
| Companies | KBR (engineering) and Applied Computing (AI) |
| Goal | Optimise plants, raise yields, cut emissions |
| Efficiency gains cited | 1%, 2%, 3% — large CO2 cuts at scale |
| First customer sector | Ammonia (used for fertiliser) |
| Other target plants | Refineries, petrochemical complexes |
| Timeline | Met Aug 2025, first contract Nov 2025 |
FAQ
Why can’t a normal chatbot run a plant?
A chatbot guesses words from text. It has no built-in grip on real physics or chemistry. In a plant, choices must be exact and safe. Being “mostly right” can be dangerous, so plain chatbots are the wrong tool.
Does this AI replace engineers?
No. The makers say it is built to support engineers, not replace them. It gives them better data and faster reasoning. Humans still make the final safety calls and keep trust in the loop.
What does “yield” mean here?
Yield is how much useful product a plant makes from its inputs. Higher yield means less waste and more output for the same raw materials. That saves money and energy.
Why it matters (especially for India and founders)
India runs huge refineries, fertiliser plants, and chemical hubs. Cutting energy use by even a few percent across them would save crores of rupees and a lot of carbon. So the same idea matters a great deal here. It also sends a clear message to Indian AI founders. The biggest value may not be in another chatbot. It may be in deep, expert AI that solves hard, real-world jobs.
This fits a wider trend of practical AI being built into Indian industry. We are seeing it in finance, where one startup is turning India’s financial advisors into AI-powered wealth coaches. We are also seeing it in sales, with the massive agentic AI deployment inside IndiaMART. The common thread is clear. Useful AI is moving from clever demos to real work that changes the bottom line.
The takeaway
The lesson is simple. Match the AI to the job. Chatbots are for words. Physics-informed AI is for machines, plants, and safety. As this approach spreads, expect cleaner, leaner, and safer factories. For India’s heavy industry, that could mean big savings and lower emissions — without putting safety at risk.
