Researchers built an ‘AI Scientist’—what can it do?

Could it be possible for science to be automated? Now the researcher has tried.

Tokyo company Sakana AI (an AI company based in Japan and Tokyo) created a model called an AI scientist in collaboration with academic labs in Canada and the United Kingdom recently performed full-cycle research on existing problems and developed a hypothesis for developing new solutions and writing a paper.

AI scientists even mimic the scientific discovery process. because AI scientists are based on a large number of language models, so they perform this job easily.

This is considered the greatest challenge to developing such variants that are able to do scientific research and knowledge.

Sakana AI outlined the new model and put it in an academic paper. The study shows how AI scientists are capable of suggesting novel ideas, writing code, executing experiments, visualising results, and doing research.

As Sakana claims, a research paper is produced for just 15 dollars by AI scientists based on the research and knowledge it uses.

In their blog post, Sakana AI’s team explains that developing AI scientists who are capable of performing research and employing a broad research direction and a starting codebase was a great challenge for them.

Jevin West says, “It’s impressive that they’ve done this end-to-end.” The AI scientist lacks the ability to do laboratory work– that most scientists would consider an important part of doing science work.

Automated Research:

AI Scientist developed as a large language model (LLM). There is a requirement of searching the literature for same work when descrines a machine learning algorithm by using a paper as a template.

A term “evolutionary computation” employed by the team on the basis of natural selection of Darwinian evolution and mutations. It is followed by in steps, applying a little bit, random changes to an algorithm and choosing the ones that provide an improvement in efficiency.

Additionally, AI scientists perform advanced experiments such as conducting its own eperiments to measure how the work by running the algorithms. Paper is produced at the end analyze it in a sort of automated peer review. This cycle starts again developing on its own results.

The statement is given by Authors that the papers produced by AI Scientists contained only incremental developments. Other researchers were withering on social media in their comments.As a reviewer, I would reject them,” saying one commenter on the website Hacker News.

“Science is more than a pile of papers,” says West. “You can have a 5-minute conversation that will be better than a 5-hour study of the literature.”

Both West and Memon encouraged the Authors for developing their codes and results fully open-West’s colleague Shahan Memon agrees. In that way they can analyze their AI Scientist’s result. Memon and West say about AI Scientist result that they preferes measuring whether the choice of AI Scientist’s were the most relevant ones.

Limitations:

It appeas that a lot of work needed to be done to compete this agants with genuine human scientists. For now, this model is limited to only AI software development, and may be there are hardware limitations.

Additionally, for their research, AI Scientists also needs a codebase as a starting point, in that way they lacks the abiloty to conceive new ideas.

As by the Sakana AI’s team, AI Scientsits lack computer vision capabilities. It can some time make comaparisons that are unfair to baselines. They also admit that there systems sometimes make “critivak errors” during evaluating results, especially during the comparisons of two magnitudes.

Still, Ceder says that even if AI won’t able to do the more creative part of the work any time soon, it could still automate a lot of the more repetitive aspects of research. “At the lower level, you’re trying to analyse what something is, how something responds. That’s not actually the creative part of science, but it’s almost 90% of what we do.” Lu says he got a familiar feedback from a lot of other some researchers, too. “People will say, I have 100 ideas that I don’t have time for my work. Get the AI Scientist to do those.”

West says, In the twenty-first century, it does force us to think what is science — what it could be, what it is, what it is not,” he says.

Main goal:

The primary objective of AI Scientists is to collect information by variuos innovative methods in order to enhance AI technology.

Also read: what is the Importance of artificial intelligence?

Also read: What are the Jobs that will be repalce by AI in the future?