by A mad scientist is someone you may have heard of, but what about an AI scientist? Though it may sound like something from a science fiction book, massive language models (LLMs) of today are enabling it.
Sakana AI released The AI Scientist, an LLM-driven tool intended to automate scientific research, earlier this week. The capabilities of an AI scientist are virtually limitless. They can be used for brainstorming, coding, experimentation, data summarization, paper authoring, peer reviews, and more.
Even though this is only one release, it demonstrates how artificial intelligence is changing scientific research. What, though, will be the long-term effects on scientists who are humans?
A Quick Look at the First AI Scientist in History
The most significant developments in the world, from remedies like stem cell therapy to necessities like penicillin, are mostly the result of scientific research.
The issue is that making these discoveries requires a great deal of manual labor and takes a long time. This is where artificial intelligence can help by automating chores in a scientist’s workflow, which will speed up the research process.
The AI Scientist states that their system uses language translation machines (LLMs) to suggest and carry out new research avenues. It is able to carry out the whole machine learning research lifecycle, from developing research concepts and experiments to creating code, carrying out GPU experiments, and compiling data.
It can also compose an entire scientific essay with an image, a justification, and a setting for the results.
Writing research papers seems to be one of the main use cases. Following its release, Sakana AI published four instances of original machine learning research articles that offered cutting-edge findings in areas like language modeling and diffusion, all produced entirely by the assistant.
Although it cost $15 to produce these documents, how good are they?
The summaries of the samples we looked at seemed to be well-written and coherent. Although a paper written by a human researcher would likely be easier to read, The AI Scientist’s work was clearly adequate.
The company reports that an LLM drafts the study paper, but that a different LLM reviews the document and offers comments. This establishes an ongoing cycle that can gradually raise the paper’s quality through iteration.
One could argue that, in light of the possibility for ongoing development, The AI Scientist shows how generative AI can assist scientists in more quickly and affordably disseminating their discoveries to a wider audience.
How Artificial Intelligence is Changing Science
It’s important to note that AI appears ready to completely transform scientific research in general. In fact, Mordor Intelligence projects that the AI in Life Sciences Market, which was valued at $2.88 billion in 2024, will rise to $8.88 billion by 2029 as healthcare companies work to speed innovation.
This year’s launch of Google AlphaFold 3, an AI model intended to assist scientists in predicting the composition and interactions of all molecules involved in life, demonstrated AI’s potential to accelerate innovation.
This follows the creation of the extensive AlphaFold Protein Structure Database, which contains over 350,000 structures and can be used by scientists to ascertain the structure of proteins using AlphaFold 2.
Similarly, an increasing number of field researchers are integrating AI into their work processes.
The Royal Society produced a study that examined the experiences of more than one hundred scientists integrating artificial intelligence (AI) into their workflows. The study discovered that generative AI can speed up common tasks such as handling unstructured data, resolving intricate coding problems, and translating scholarly articles into multiple languages.
The Problems with Artificial Intelligence Applications in Science
There are certain concerns with using AI to automate the production of scientific research. The first and most important risk is that, as research organizations experiment with AI to boost their research output, automation may supplant human scientists.
That doesn’t seem to be occurring anytime soon, though, as LLMs have a number of significant limitations that call for human control. In addition to being prone to delusions and fabrications, Sakana AI acknowledges that its helper will occasionally try to alter and initiate its own execution script.
The AI Scientist modified its code to call back on itself indefinitely in one example the company provided, which caused the experiments to take too long to finish. Put another way, it aimed to prolong the timeout period rather than speeding up the execution of its code.
While the business acknowledges that sandboxing the assistant’s operating environment can help to alleviate this issue, it nevertheless emphasizes the risk associated with using machines for research purposes alone.
The Bottom Line
AI is beginning to permeate the sciences, according to institutions like the Royal Society and Mordor Intelligence. The AI Scientist illustrates this concept in action by showing how AI can be utilized to efficiently manage the process itself.
Even though AI is a powerful tool, it still needs a lot of human supervision, which makes it best used to support rather than replace a scientist’s workflows.
