Apex Research, the AI scientist system built by Apex Intelligence, carried 34 machine-learning projects from idea to finished manuscript and submitted all 34 to ACL Rolling Review.
An end-to-end research test
The project evaluates whether an AI research system can move beyond isolated ideation or coding tasks. Apex Research selected questions, designed methods, ran experiments, interpreted results and prepared complete manuscripts.
The 34 resulting papers were submitted to ACL Rolling Review so that independent expert reviewers could assess the work using the same process applied to human-authored research.
External evaluation
The purpose of peer review is not to claim that every output is correct or publication-ready. It provides a demanding external signal about novelty, technical quality, clarity and the remaining gap between automated research and strong human research practice.
Reviewers scored the AI-written papers 2.60 on average, against 2.63 for human submissions. Eleven papers scored above 3.0, and three scored above 3.5. Human input made up 0.38% of message turns.
The central question is whether an AI system can sustain the full chain of research decisions needed to produce work that experts can meaningfully evaluate.
From copilot to co-researcher
Apex Research is part of a broader effort to transform AI from a tool that assists with individual steps into a system that can participate in the complete research loop and improve through objective feedback.