01 / Why it matters?
Engineering is the harder frontier.
AI has learned to perceive, to generate, and to reason. Engineering resists all three: data is scarce and expensive, physics is unforgiving, and a wrong answer is not a typo — it is a failed part.
The opening is not only how to effectively apply AI to engineering, but also how to change the way engineering problems are solved.
Distilled from a long tradition of engineering and research practice — including Integrated Computational Materials Engineering (ICME) — and from AI projects across engineering and science whose failures taught us as much as their successes, we set out the concept and framework of AI for Engineering (AI4E) in 2022, and have shared it ever since through courses and lectures grounded in hands-on, real-world cases.
02 / Definition
AI for Engineering (AI4E) is a framework that starts from the engineering problem and — grounded in empirical, theoretical, experimental, and computational science — integrates artificial intelligence to generate effective data, build physical models, and create innovative tools, so as to solve it with high efficiency, high accuracy, and high reliability.
Its next-generation paradigm is AI4AI-Powered AI4E ↓.
Also written AI4E, or AI4Engineering.
03 / Key proposed approaches
Three moves, from the problem outward.
Generate effective data
Engineering rarely comes with internet-scale data. We treat data as part of the method — combining experiment, simulation, synthetic data, and theory so a model can learn from what little ground truth exists.
Build physical models
A model that breaks physics is worse than no model. We build physical common sense into the model itself, so its answers stay inside the laws they describe. And a model must know how far to trust itself — a prediction whose uncertainty is unknown is hard to act on in engineering.
Create innovative tools
A method matters only when an engineer can use it. We turn models into tools that fit the way engineering is actually done.
04 / The next era of AI4E
AI4AI powered AI4E: From one domain to many.
Solving engineering one domain at a time does not scale — every new field costs another round of data, modeling, and tuning.
What if anyone could be an AI4E expert in any domain — without learning AI, without learning AI4E, without even knowing the domain itself?
The next move is one level up: not a model for every problem, but an AI that builds the AI for each problem. Less like training one expert who must know everything, more like training a teacher who knows how to learn.
In an AI4AI-enhanced AI4E world, AI conducts its own research on each engineering problem, training a specialized — often temporary — model for each target, whether initially set by a human or by another AI, rather than expecting one unified model to solve them all.
We see AI4AI as the next structural advance in AI, at least in engineering — a potential path to ever more capable AI, arguably toward “AGI” (depending on the definition of AGI). Rather than only scaling one ever-larger model across more modalities of engineering, or even science, it points to a different route: progress through a reduction in structure.
This is AI4AI — and AI4AI powering AI for Engineering is the direction we build toward.
05 / Our domains
Examples of where the AI4E framework might fit
06 / Get in touch