AI Scientists: Has Science's 'Full Self-Driving' Era Begun?
Introduction:
If you are a scientist, you have surely fantasized about this: could you just propose a bold hypothesis, and have AI automatically handle everything else—searching the literature, designing experiments, operating lab robots, writing the paper, and fixing bugs?
At a workshop at ICML 2026, this sci-fi premise became reality. Researchers from top global labs jointly defined a new species: the "AI Scientist." They were no longer discussing "can AI assist research," but rather "should AI be our tool, a co-author, or the future founder of laboratories."
---
#### 1. The 'New Face' in the Lab
AI is no longer just a research assistant. Across physics, chemistry, and biomaterials, a wave of models with autonomous discovery capabilities is emerging.
- Autonomous hypothesis generation: They can scan millions of papers, spot "logical gaps" humans never noticed, and propose counterintuitive new material combinations.
- Closed-loop experimentation: By connecting to Self-driving Labs, AI can directly command robotic arms to synthesize and test samples 24 hours a day.
- Tool theorists: AI will always be just a more advanced slide rule.
- Partner theorists: AI's creative contribution already exceeds that of many junior researchers, so it deserves co-author status.
- Founder theorists: A more radical view—future startups may be built around a "domain-specific AI Scientist," with humans serving merely as its "administrators."
- Drug discovery: Screening speed for rare-disease drugs improved 100x.
- Superconductor search: Crystal structures that once required three generations of physicists to hunt down can now be shortlisted by AI in a single afternoon.
- Automated knowledge replication: Using the ARA protocol (Agent-native Research Artifact), the moment a discovery is published, AI scientists worldwide can immediately replicate and verify it locally.
#### 2. From 'Assistant' to 'Named Author': The Ethical Deep Water
The workshop tackled a highly controversial question: if 90% of the work behind a major discovery was done by AI, whose name goes first on the paper?
#### 3. Results: A 'Cambrian Explosion' in Research Efficiency
The consequences of this fully automated research model are staggering:
#### Editorial Commentary (Zhichai)
The essence of science is exploring the boundary of the unknown. In the past, that boundary was set by human mental and physical capacity combined. Now AI is taking over the "physical labor" and "junior mental work."
This does not mean the end of human scientists—it means an upgrade of the scientific paradigm. Humans will be freed from tedious experimental details and move toward higher-order "aesthetics, intuition, and philosophical judgment." The great scientists of the future may resemble directors with impeccable taste, with AI as their most prolific production team.
If one of your paper's co-authors were an AI, what would you most fear it might do? Share your thoughts in the comments!
--- *Note: This article is based on the spirit of the ICML 2026 "AI Scientists" workshop.*