The newly formed Institute of Automated Science has launched “Carl.” This was the first AI system to write academic research papers and passed a rigorous double-blind peer review process.
Carl’s research paper was accepted in a small paper track at the International Conference on Learning Expression (ICLR). Critically, these submissions were generated with minimal human involvement, and marked a new era for AI-driven scientific discoveries.
Meet Carl: “Automatic Research Scientist”
Carl represents advances in the role of AI as not only tools but as active participants in academic research. Called “automated research scientist,” Carl applies natural language models to accurately integrate academic work, hypotheses and cite them.
Importantly, Carl can read and understand published papers in seconds. Unlike human researchers, it works continuously, thus accelerating the research cycle and reducing experimental costs.
According to Autonomy, Carl “recognized, experimented, and executed new scientific hypotheses that designed and executed multiple academic papers that passed peer reviews at the workshop.”
This highlights the possibility that AI not only complements human research, but also outweighs speed and efficiency in many ways.
Carl is a meticulous worker, but human involvement is still essential
Curl’s ability to generate high-quality academic work is built on a three-stage process.
Formation of ideas and hypotheses: Using existing research, CARL identifies potential research directions and generates hypotheses. That deep understanding of the relevant literature allows us to develop new ideas in the field of AI. Experiment: CARL writes code, tests hypotheses, and visualizes the resulting data via detailed numbers. That tireless operation reduces iteration times and reduces redundant tasks. Presentation: Finally, Carl summarises his findings into sophisticated academic papers. Visualize your data and achieve clear and clear conclusions.
Carl’s capabilities are largely independent, but workflows require human involvement to adhere to calculation, formatting, and ethical standards.
Green Light Research Procedure: To avoid wasting computational resources, human reviewers provide a “continue” or “stop” signal at specific stages of the curl process. This guidance will lead the project to a more efficient project, but it does not affect the details of the research itself. Citations and Format: The Automated Science Team ensures that all references are correctly cited and formatted to meet academic standards. Currently, this is a manual step, but it ensures that the research is consistent with the expectations of the publishing venue. Supporting Pre-API Models: CARLs occasionally rely on new OpenAI and deep research models that lack automatic accessibility APIs. In such cases, manual interventions such as copy pasting output fill these gaps. Autoscience hopes that these tasks will be fully automated in the future when APIs become available.
In Carl’s debut paper, the human team also created a “Related Works” section, which also helped refine the language. However, these tasks were not required following any updates applied prior to subsequent submissions.
A rigorous verification process for academic integrity
Before submitting the survey, the autonomous team undertook a rigorous verification process to ensure that CARL work meets the highest standards of academic integrity.
Reproducibility: Every line of Carl’s code was reviewed and the experiment was rerun to check for reproducibility. This ensured that the findings were scientifically valid and not accidental anomalies. Originality Check: Autoscience conducted extensive and innovative reviews to ensure that CARL ideas are not versions of existing publications, but new contributions to the field. External Verification: Hackathons, involving researchers from prominent academic institutions such as MIT, Stanford University and UC Berkeley, independently examined Carl’s work. Further plagiarism and citation checks were performed to ensure compliance with academic norms.
It may not be denied, but raises a bigger question
Achieving acceptance in a respected workshop like ICLR is an important milestone, but self-science recognizes the larger conversations that this milestone can cause. Carl’s success raises larger philosophical and logistical questions about the role of AI in the academic environment.
“We believe that legitimate results should be added to the public knowledge base regardless of where they came from,” Self-Science explained. “If a research meets scientific criteria set by the academic community, what it creates should not lead to automatic disqualification.”
“However, transparent science requires proper attribution, and we believe that work purely generated by AI systems is identifiable from those produced by humans.”
Given the novelty of autonomous AI researchers like CARL, conference organizers may need time to establish new guidelines that explain this new paradigm, particularly to ensure fair assessments and intellectual attribution standards. To prevent unnecessary controversy now, Autoscience has withdrawn Carl’s paper from the ICLR workshop while these frameworks were being devised.
Advanced automatic science aims to contribute to the formation of these evolving standards. The company plans to propose a dedicated workshop at Neurips 2025 to formally respond to research submissions from the autonomous research system.
As the narratives surrounding AI-generated research unfold, it is clear that systems like CARL are not just tools, but collaborators who pursue knowledge. However, as these systems go beyond typical boundaries, the academic community must adapt to fully embrace this new paradigm, while protecting integrity, transparency and proper attribution.
(Photo: Rohit Tandon)
See: you.com ARI: Professional Grade AI Research Agent for Business
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California and London. The comprehensive event will be held in collaboration with other major events, including the Intelligent Automation Conference, Blockx, Digital Transformation Week, and Cyber Security & Cloud Expo.
Check out other upcoming Enterprise Technology events and webinars with TechForge here.
