Structural engineering has always relied on shortcuts that encode accumulated experience.
Empirical formulas. Rules of thumb. Span-to-depth ratios. Tables in the back of textbooks. These weren’t derived purely from first principles - they were patterns extracted from what had been built and tested, conservatism baked in to cover the uncertainty.
In a sense, that’s always been a form of AI. Compressed experience, translated into something a practitioner can apply quickly.
The difference is that a genuinely well-trained AI model - trained on thousands of real designs, real loads, real outcomes - could potentially do the same thing with far less conservatism, because it has far more data. The patterns it learns don’t need to be as blunt.
What I find interesting is that the industry has always had an implicit understanding that the engineer is supposed to provide value above those tables and formulas. To find the 10% that the rule of thumb left on the table. To know when the standard approach doesn’t apply.
If AI is essentially doing what empirical formulas always did but much better, then the question of where human expertise adds value becomes a lot more pointed. I don’t have a clean answer to that. But it’s worth sitting with.