Can machines design like humans?

Determine whether artificial intelligence systems can autonomously design complex engineered artifacts at a human level by constructing artifact descriptions that satisfy given goals and constraints, thereby resolving whether machines can design like humans.

Background

The paper frames machine design as constructing an artifact description that satisfies specified goals and constraints, noting that recent AI successes in designing materials, proteins, drugs, chip floorplans, and programs operate in comparatively smaller search spaces than those tackled by human designers.

The authors highlight that designing a CPU entails exploring an extraordinarily large search space (on the order of 10{10{540}} alternatives), far exceeding prior machine-designed objects, which motivates the longstanding question of whether machines can truly design like humans.

References

Though machines have already demonstrated their abilities in designing new materials, proteins, and computer programs with advanced AI techniques, the search space for designing such objects is relatively small, and thus, "Can machines design like humans?" remains an open question.

— Pushing the Limits of Machine Design: Automated CPU Design with AI  (2306.12456 - Cheng et al., 2023) in Abstract

Finally, traditional computational design is limited to optimization within predefined design configurations, with open challenges for high-level design of novel lattices, configurations, and accelerator components, and even exploration of novel beam dynamics and manipulations.

— Future of Artificial Intelligence for Science in Japan 2024 Community Report  (2608.27807 - Itow et al., 28 Aug 2026) in Section 4, subsection “Accelerator Physics,” paragraph “Design”