AI for Scientific Research
Artificial Intelligence (AI) can greatly accelerate scientific research if used properly.
This note aims to define AI, explain its role in scientific research, highlight recent advances, untangle hype from reality, and share tools to speed up your own research.
What is AI?
Artificial Intelligence (AI) is defined as the 'science of agent design' (Russell et al., 2022), where an agent consists of an environment, perception, and action, evaluated by a performance metric and improved through learning. The term 'AI' was first coined at the Dartmouth (USA) workshop in 1956. More broadly defined, AI refers to 'the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making' (Wikipedia, 2025). Thus, AI is really composed of different components, each suited to a different task.
AI in Scientific Research
Japanese AI company SakanaAI released the 'AI Scientist', an automated system for scientific research. One of its papers recently passed peer review (SakanaAI, 2025) – a first in AI history. The system combines multiple components: different Large Language Models (LLM), a coding environment, and a literature search engine. While impressive, it’s far from being a single intelligent entity. Human researchers shaped the whole process, it is focused on a narrow technical domain (Machine Learning), and it produced hundreds of invalid papers. Still, it showcases the potential of AI for generating ideas, writing code, and drafting manuscripts.
Myths vs Reality
AI can explore and review scientific ideas, generate coherent text and translations, check language errors, write experimental code, and create illustrative images. But it can also make serious mistakes!
AI cannot (yet) write papers in any given domain, conduct longitudinal studies, or run most real-world experiments. In particular, LLM lack rigorous logical reasoning and symbolic understanding—they're still just sophisticated word generators.
AI Tools to Accelerate Scientific Research
Here is a non-exhaustive collection of some AI toots that may speed up your research. Feel free to explore, and let me know if you have suggestions to add!
- Large Language Models (LLM): natural language generation through a chat interface → often multi-modal (text, audio, image, video)
- AI-Augmented Search: enhance literature search using semantics (not only keywords)
- AI Research Assistants: for literature search, ideation, summaries, review
- Retrieval Augmented Generation (RAG): interact with a trusted or custom data source (book, paper, database) through a large language model
- PerplexityAI, NoteBookLM by Google, ChatPDF
- Also with various LLM through uploading (drag & drop)
- Data Science
- AI detection
- Coding
GitHub Copilot, Cursor - Image Generation
- Midjourney
- Also with various LLM through prompting
- Text translation
→ Most tools can perform multiple tasks across this list. However, keep in mind that most AI-tools are designed for a specific function. While experimenting can lead to lucky accidents, choosing the right tool for the right task usually yield better outcomes.
Closing words
AI can significantly accelerate scientific research, and adapting to this technology appears inevitable. However, the role of human researchers is and will remain crucial. What are your thoughts?
References
- Russell, S. J., Norvig, P., Chang, M., Devlin, J., Dragan, A., Forsyth, D., Goodfellow, I., Malik, J., Mansinghka, V., Pearl, J., & Wooldridge, M. J. (2022). Artificial intelligence: A modern approach (Fourth edition, global edition). Pearson.
- SakanaAI. (2025-03-12). The AI Scientist Generates its First Peer-Reviewed Scientific Publication. SakanAi Blog. URL: https://sakana.ai/ai-scientist-first-publication/
- Wikipedia. (2025). Artificial intelligence. URL: https://en.wikipedia.org/wiki/Artificial_intelligence