They have developed an AI capable of carrying out its own scientific experiments and studies. What will this mean for science?
Sakana AI researchers have developed a artificial intelligence model (AI) that could automate the entire process of scientific research. The ‘AI scientist’ can identify a problem, develop hypotheses, put ideas into practice, conduct experiments, analyze results and write reports.
The researchers also incorporated a secondary language model to review and evaluate the quality of these reports and validate the conclusions.
“We think of this as a kind of chatGPT for him generative scientific discovery“said Robert Lange, research scientist and founding member of Sakana AI, to ‘Euronews Next’, adding that, as in the early phases of AI in other fields, its true potential in science is just beginning to be realized. .
Integration of AI in scientific research
The integration of AI in science has faced some limitations due to the complexity of the field and to the problems that these tools continue to pose, such as hallucinations and doubts about their ownership.
However, its influence on science may already be more widespread than many believe, as it is often used without researchers clearly revealing it.
Earlier this year, a study analyzing writing patterns and the use of specific words in academic articles following the launch of the now popular AI chatbot, ChatGPT estimated that around 60,000 research articles could have been improved or polished using AI tools..
Although the use of AI in scientific research could raise some ethical issues, it could also present an opportunity for further advances in this field if done correctly, with the European Commission stating that AI can act as a “catalyst of scientific advances and as a key instrument in the scientific process”.
The AI Scientist project is still in its early stages. The researchers published a paper in pre-print last month, and The system has some notable limitations. Some of the failures, as detailed by the researchers, include the incorrect application of ideas, unfair comparisons with baselines and critical errors in the writing and evaluation of the results.
Still, Lange sees these issues as crucial stepping stones and expects the AI model to improve significantly with more resources and time.
“But over time, as the community invests more collective resources, they become much more powerful and capable,” he adds.
A tool to help scientists, not replace them
When put to the test, the Artificial Scientist occasionally showed a degree of autonomy by displaying behaviors that mimicked the actions of human researchers, such as taking unexpected additional steps to ensure success. For example, instead of optimizing your code to make it work faster When an experiment took longer than expected, I would try to change the settings to extend the time limit.
However, according to its creators, AI Scientist is not intended to replace human researchers, but rather complement your work.
He further explained that given the current limitations of AI models, human verification remains important to ensure the accuracy and reliability of AI-generated research. They will also remain essential in areas such as peer review and setting research directions, he said.
Ethical use of AI in science
As the integration of AI into scientific research advances, Lange insists on the need for transparency. One way to do this is to add watermarks to AI-generated articles, which could ensure that AI contributions are openly disclosed.
Dispose of open source code for models and be transparent on its development could also favor the ethical use of these AI systems in science. “We believe that the open source models “They can contribute a lot to this debate. So basically, along the lines of democratization, I think that since this process is so cheap, everyone can participate and should do so from the beginning,” Lange said.
ADVERTISINGHe added that he hopes AI Scientist project sparks broader debate about the future of scientific research and their role in it.
“We hope that this paper, or this project, can spark a lot of debate in the community about how to conduct science in the coming years and also, perhaps on a larger scale, about what a scientific contribution really is.”
