Not All Physics Problems Are Physics Problems: Software, Science, and the Limits of AI
ORAL · Invited
Abstract
Some of the hardest problems slowing physics aren't physics problems. They're software problems. Our systematic failure to recognize this has cost the scientific community hundreds of millions of dollars (with receipts).
Why do brilliant people keep underestimating software? Software engineering grapples with irreducible, mathematically-grounded complexity that can grow exponentially, punishing anyone who underestimates it. Understanding what software engineering truly is and what makes it intrinsically challenging reveals something unexpected: the core of software engineering is perceiving and modeling messy and uncertain reality. This is the same epistemic work that defines science itself.
AI makes this vivid. Current models can write code fluently but cannot do the deeper work that makes software engineering genuinely difficult at scale. What happens when they can?
Why do brilliant people keep underestimating software? Software engineering grapples with irreducible, mathematically-grounded complexity that can grow exponentially, punishing anyone who underestimates it. Understanding what software engineering truly is and what makes it intrinsically challenging reveals something unexpected: the core of software engineering is perceiving and modeling messy and uncertain reality. This is the same epistemic work that defines science itself.
AI makes this vivid. Current models can write code fluently but cannot do the deeper work that makes software engineering genuinely difficult at scale. What happens when they can?
*We are proud to be supported by the National Science Foundation.
–
Presenters
-
Calvin Li