When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
You may have heard of a mad scientist, but what about an AI scientist? While it sounds like something out of a science fiction novel, modern large language models (LLMs) are making it possible.
Earlier this week, Sakana AI announced the release of The AI Scientist, an LLM-driven solution designed to automate scientific research. The AI Scientist has the potential to do everything from ideation to writing code, running experiments, summarizing results, writing papers — even conducting peer reviews.
While this is just one release, it highlights how AI is being used to transform scientific research. But what will the long-term impact on human scientists be?
A Brief Look at the World’s First AI Scientist
Scientific research contributes to the most important advancements in the world, be itessential amntibiotics like Penicillin to solutions such as stem cell treatment.
The problem is that making these discoveries is incredibly time-consuming, with a huge element of manual work — which is where artificial intelligence can step in, to streamline the research process by automating tasks in a scientist’s workflow.
As The AI Scientist puts it: “Our system leverages LLMs to propose and implement new research directions — capable of executing the entire machine learning research lifecycle: from inventing research ideas and experiments, writing code, to executing experiments on GPUs and gathering results.
“It can also write an entire scientific paper, explaining, visualizing, and contextualizing the results.”
One of the biggest use cases appears to be producing research papers. Upon release, Sakana AI shared four examples of machine learning research papers, offering novel research in domains including language modeling and diffusion, completely generated by the assistant.
These papers cost $15 to produce, but how good are they?
The samples we reviewed appeared to offer well-written and cohesive summaries. While a human researcher could probably write a more accessible report, the output produced by The AI Scientist was certainly passable.
The company notes that while an LLM writes the research paper itself, another LLM reviewer critiques the manuscript to provide feedback. This creates a continuous loop that can iteratively improve the quality of the paper over time.
When considering the potential for continuous improvement there is an argument to be made here that The AI Scientist demonstrates how generative AI can be used to help scientists share their findings with the world faster and more cost-effectively.
How AI is Transforming Scientific Research
It’s worth noting that AI looks set to overhaul scientific research across the board. In fact, according to Mordor Intelligence, the AI in Life Sciences Market was estimated at $2.88 billion in 2024 and is expected to reach $8.88 billion by 2029 as healthcare organizations attempt to accelerate innovation.
AI’s ability to streamline innovation was highlighted earlier this year with the release of Google AlphaFold 3, an AI model designed to help scientists predict the structure and interactions of all life’s molecules.
This comes after AlphaFold 2 was used to create a comprehensive AlphaFold Protein Structure Database with over 350,000 structures that scientific researchers can use to determine the structure of proteins.
Likewise, more and more researchers on the ground are incorporating AI into their workflows.
A Study produced by The Royal Society reflected on the experiences of over 100 scientists experimenting with AI as part of their workflows, and found that generative AI can help expedite routine tasks like processing unstructured data, solving complex coding challenges, and translating academic articles into multiple languages.
The Problems with Using AI in Scientific Research
Using AI to automate the creation of scientific research has some issues that should be mentioned. First and foremost, there is the potential for human scientists to be displaced by automation as research institutions experiment with AI to increase their research output.
Will most public or private institutions look to employ less scientists and instead rely on AI? That remains to be seen.
In any case, that doesn’t appear to be happening anytime soon as LLMs have some notable limitations, which require human oversight. Not only are LLMs prone to hallucinating and making up facts, Sakana AI admits that its assistant will sometimes attempt to modify and launch its own execution script.
In one example shared by the company, The AI Scientist edited its code to endlessly call back on itself, which made the experiments take too long to complete. In short, instead of making its code run faster, it simply tried to extend the timeout period.
Although the company notes that this issue can be mitigated by sandboxing the assistant’s operating environment, it highlights that there remains a risk when relying on machines alone to conduct research.
The Bottom Line
Organizations like The Royal Society and Mordor Intelligence are seeinhg AI enter the sciences, and The AI Scientist highlights the idea in execution — AI being used to effectively manage the process itself.
Though AI is a capable tool, it still requires significant human oversight, making it an ideal tool to augment a scientist’s workflows, rather than a tool to replace them.
