Overview

Scientific papers sometimes lay dormant and largely unnoticed for long periods of time before suddenly attracting great attention — a phenomenon termed as "Sleeping beauties" in science by Van Raan. According to Ke et al.'s survey of nearly 23 million papers in both sciences and social sciences over the past 100 years, "Sleeping beauties" have occurred across all fields of study. Unlike the previous AI4Science workshops, which tackle scientific discovery based on AI methods, we want to ground our AI4Research into the scientific literature to uncover these potential "Sleeping beauties" and increase explainability. Finally, as Clauset et al. emphasize in their essay, we also want to ensure that "the usage of prediction tools does not inhibit future discovery, marginalize underrepresented groups, exclude novel ideas, or discourage interdisciplinary work and the development of new fields".

This workshop aims to help researchers explore and discuss the entire AI-assisted scientific research lifecycle, detailing how machines can augment every stage of the research process, including literature survey, hypothesis generation, experiment planning, results analysis, manuscript writing, paper evaluation, and fact-checking. We expect interdisciplinary collaboration to explore autonomous research for topics beyond existing natural science domains. This workshop solicits viewpoints from scientists and technology developers that go beyond technical issues to better understand the needs of the human-in-the-loop scientific research lifecycle. To this end, we want to broadly investigate several crucial yet potentially overlooked problems, including but not limited to:

• Challenges and potential solutions for autonomous scientific research: How can AI generate factually correct code/results for the scientific research lifecycle? How can AI overcome limited amounts of high-quality scientific data? Can we develop unified foundation models to incorporate domain knowledge and multimedia information effectively?

• State-of-the-art algorithms for each stage of autonomous scientific research: How can we develop end-to-end methods to enhance the scientific research workflow by grounding knowledge in scientific literature and external knowledge bases? Can we build frameworks to simulate scientific research procedures? How can we use structured and unstructured knowledge to enhance explainability in the automated scientific research lifecycle?

• Accessibility to scientists and collaboration between knowledge-grounded autonomous scientific research models: Researchers developing AI methods might struggle to understand the needs of scientists from other disciplines without a computer science background. Can we integrate AI methods into existing scientific platforms, such as PubMed, Google Scholar, OpenReview, etc? Can we design better human-computer interaction interfaces for scientists to use or even finetune those tools?

• The social impact and ethical considerations of knowledge-grounded AI methods in the scientific research lifecycle: What potential social problems could arise from using these tools, such as issues of factuality, plagiarism, pressure on peer-review, and privacy? How will the rise of these tools impact underrepresented groups and the creation of new research directions?

Apart from the directions listed above, we also welcome research work related to the following topics:

• Dataset: We will introduce a dataset track to emphasize new tasks in AI4Research.

• Evaluation: We will also accept new evaluation methods or metrics to measure the performance of AI methods for different stages of knowledge-grounded autonomous scientific research, including scientific hypothesis generation, automated review generation, etc.

Please check here if you are interested in the student mentoring session.  We especially encourage students from underrepresented groups to reach out.

Registration Link: https://aaai.org/conference/aaai/aaai-25/registration/

Organizers