How AI tools could help scientists—and where they may threaten research integrity

A survey maps how large language models may assist research from finding papers to reviewing scientific work, while outlining evaluation and misuse risks.

Editorial illustration — not from the study.

The survey reviews research on AI tools that assist with finding relevant literature, generating ideas, conducting experiments, writing scientific content, creating figures and diagrams, and evaluating scientific work such as through peer review. It organizes the field’s datasets, methods, reported results and evaluation strategies.

The authors also discuss limitations and ethical concerns, including the possibility that generative models could be misused in ways that harm research integrity. They present the survey as an accessible guide for newcomers and as a basis for developing future systems that integrate AI into scientific work.

Where AI may assist science

The survey identifies five main areas in which large language models and related tools are being developed for science: literature search; research-idea generation and experimentation; text-based content creation; production of multimodal materials such as figures and diagrams; and evaluation of scientific work, including peer review. Across these areas, it reviews representative techniques, datasets, reported results and ways researchers assess performance. It also catalogs limitations and ethical concerns, particularly risks to research integrity from misuse of generative models.

A map of the research process

The work provides a structured view of a broad and developing field rather than presenting a single new AI system. By considering assistance across the full research lifecycle—and pairing possible uses with evaluation practices and integrity risks—it can help researchers and others assess where these tools may fit into scientific work and what safeguards or tests may be needed.

What the survey can show

This is a survey article, not a new experiment or a direct test of one AI system. Its conclusions depend on the literature selected and summarized by the authors. The abstract does not provide quantitative comparisons, a complete account of the survey’s selection process, or evidence that the reviewed tools reliably improve scientific research in general. The authors explicitly note limitations, ethical concerns and research-integrity risks across the surveyed applications.

High-mounted cameras spot pedestrians on the device at busy intersections

Researchers introduced a 13,314-image dataset for testing AI systems that view urban intersections from above. The results show that detection can run on edge devices such as the Jetson Orin Nano, although pedestrians remain harder to detect than vehicles and performance changes with scene conditions.

5 hours ago·2 views

AI & Computing

Where smaller AI models can work alongside—and compete with—larger ones

A survey examines the roles smaller AI models can play alongside large language models. It focuses on both collaboration and competition, including how smaller systems may help researchers and businesses use computing resources more efficiently.

21 hours ago·6 views

AI & Computing

AI watermarks in Europe do not yet meet all four legal tests

The European Union’s AI Act calls for providers to mark and detect outputs from general-purpose AI models. A new analysis finds that no current watermarking approach for large language models meets all four of the law’s stated standards: reliability, interoperability, effectiveness and robustness.

2 days ago·17 views