Could AI systems currently evaluate novel propulsion concepts better than peer review?

Based on these preprints, the honest answer is: not yet, and probably not soon.

Here's what the evidence shows:

Current AI limitations are severe. The CritPt benchmark [AG-2025.09-1139] tested state-of-the-art AI models on 71 unpublished, research-level physics problems across 12 fields—the kind of reasoning frontier propulsion work requires. Even the best models solved only ~6% of full research challenges. That's not a rounding error; it's a fundamental gap between AI capabilities and actual physics research needs.

When AI evaluates novel physics, accuracy crumbles. Two separate experiments asked Perplexity.ai to objectively compare competing models of gamma-ray bursts [AG-2026.01-1434, AG-2026.02-1185]. The results were "enlightening but worrisome"—the AI produced plausible-sounding but unreliable assessments. This is precisely what you'd fear in propulsion review: confident-sounding nonsense that sounds authoritative.

**What AI *can* do is narrow.** AI-Newton [AG-2025.04-1028] succeeds at rediscovering well-established laws from clean experimental data. The AI Cosmologist [AG-2025.04-076] automates routine data analysis pipelines. These are valuable, but they're automating *execution*, not novel judgment. Evaluating whether a new propulsion concept is theoretically sound, experimentally viable, or worth funding requires the kind of creative, uncertainty-tolerant reasoning that frontier physics demands—and [AG-2025.04-1028] notes that "the theorist is in no way in danger of being replaced by AI in the near future."

The real role emerging is hybrid. Rather than replacing peer review, AI works best as a research tool that frees humans to think more deeply [AG-2026.04-820]. An AI might check mathematical consistency, run simulations, or surface contradictions in a propulsion paper—but the final judgment about novelty, feasibility, and significance still requires human expertise and intuition.

For propulsion concepts specifically, which live at the boundary of speculative theory and engineering constraints, peer review by physicists and engineers remains irreplaceable.