First Principles Thinking in the Age of AI

When facing novel and challenging problems on behalf of our clients, we, as design engineers, can be tempted to jump straight into rapid prototyping and low-cost iterations in pursuit of “speed”. In a previous article, Tom Merle, General Manager at Re:Build Fikst explained “Why first principles thinking should come before your first prototype.” As Tom noted, in order to generate “velocity” toward manufacturing-ready solutions, we use first principles to ensure each prototype de-risks an aspect of the design and furthers our understanding of the problem. Given that and the ever improving landscape of AI assistants, how can we then use AI as a tool to develop first principles thinking without falling into the trap of “guess and check” engineering?

With enough uninformed iterations, we might creep up on a prototype solution that meets many of the performance requirements. However, when we pivot to higher volume manufacturing, we can find ourselves with a system that no longer works (or doesn’t work well). Worse than that, those many rounds of prototypes might circumvent the first principles understanding required to quickly determine why the product isn’t working and how to fix it. When we strengthen our first principles understanding of the design drivers, we can prototype efficiently and increase our “velocity” towards manufacturing-ready designs.

To generate this velocity, we must avoid the trap of “guess-and-check” or “build-and-test” engineering and instead develop a deep understanding of the first principles governing the specific problem.

This is where AI can lend a big hand if we can avoid slipping into using AI as an answer generator. When used for research, AI assistants quickly provide overviews of the basic physics behind complex technologies. This can help to close the knowledge and experience gap between entry level and more senior engineers. Beyond research, today’s AI assistants are powerful enough to solve engineering word problems and provide a detailed explanation of the mathematics behind the solution. Obtaining these solutions is so quick and easy that this can feel like expediency. That’s because it is.

If we take the AI assistant’s output at face value, we are very likely to develop a prototype that works, but we shortchange ourselves of the opportunity to develop first principles understanding of how and why the system works. What happens when we need to change an input as the product matures? For example: the bracket was strong enough when machined from 6061 aluminum as a prototype, but will it be strong enough when cast for higher volume production? The power supply didn’t overheat when everything was on a bread board, but will it stay cool when we package it into a control panel?

With first principles understanding, we develop the intuition to evaluate how these changing parameters affect the performance of the system.

To ensure we are generating project “velocity” toward manufacturing ready solutions and build our first principles understanding, we can use AI not just as a research assistant, but as a tool to develop engineering calculators. By vetting these tools with simple and known cases, we can gain confidence that the outputs behave as expected, and leverage the tool as the design matures. This vetting provides us with some protection against AI hallucination or misunderstanding of a problem.

For example, one of our engineers recently ensured the stability of a relatively tall and narrow component rolling horizontally on casters. They needed to determine whether a rapid stop caused by catching the caster on a rock or some similar debris would pose a tipping risk while someone moved the component at a walking pace. The engineer started with hand calculations based on first principles, then turned to an AI assistant to validate their result. The AI assistant hallucinated, making an incorrect assumption about the direction of motion and reported a tipping risk at a much lower velocity. The engineer investigated the discrepancy by performing their own hand calculation based on the AI’s assumed direction of motion, and these calculations matched the AI model. The engineer then had the AI model explicitly solve for the hallucinated case and the results matched yet again. In this example, the design exercise helped to strengthen the engineer’s first principle understanding and resulted in an accurate calculation and a safe system.

Case study:

Problem: On a recent project, our team developed a custom clock spring mechanism to pass electrical signals through a rotary joint, and needed to determine the number of coils and the geometry of the coils to achieve a specific radial range of motion and expected service life.

Solution: The engineers turned to an AI assistant to help research the first principles and develop a solution which they then physically prototyped to validate the AI assisted calculations. This first prototype “worked”.

Solution Process: When project requirements shifted and required the design to fit into a smaller form factor, the lead engineer didn’t restart the design from scratch. Instead, they used an AI assistant to develop a Python based GUI to allow for quick modifications of input variables, trading physical prototypes for mathematical prototypes and a deep understanding of the tradeoffs involved in clock spring design.

When our clients trust us to address problems we don’t immediately know how to solve, our first step cannot be to jump blindly into CAD and race to the 3D printer. Project “velocity” requires that we start by leveraging and expanding our first principles thinking to understand the problem before we begin designing, modeling, building, and testing solutions. Modern AI tools should accelerate this process, not circumvent it. By using AI not as an answer generator, but as a research tool and an assistant for creating engineering models, we protect our designs from hallucinated errors and strengthen our first principles understanding. There are so many tools and services today to allow anyone to iterate quickly, but generating true project “velocity” requires first principles thinking so that every prototype moves the product towards a scalable, manufacturing-ready design.