AI has more creative ideas than 50 scientists combined

The opportunities and risks of artificial intelligence are currently being hotly debated. A study now shows that AI has enormous potential for research.

Since the release of ChatGPT, the topic of artificial intelligence has become omnipresent. This year alone, the AI ​​sector is expected to have a market volume of around 228 billion US dollars.

According to forecasts, this figure could reach up to 632 billion US dollars by 2026. Artificial intelligence therefore has enormous potential – and not just economically.

AI could also contribute to more innovation in science and research, as the results of a new study show. According to the study, artificial intelligence can develop more research ideas than 50 scientists working independently.

Could AI enrich research?

For this study, the research ideas of an AI generator and scientists were compared. Reviewers were then asked to evaluate them – without knowing which idea came from a human and which from a machine.

According to the results of the study, AI-generated research ideas performed better. They were rated as more exciting, but suffered losses in terms of feasibility.

However, there are limitations to the results of the present study, which has not yet been peer-reviewed. The study only covers one research area. In addition, the human participants had to put forward their ideas relatively spontaneously.

This is what the study looked like

For their study, researchers at Stanford University recruited more than 100 colleagues. 49 of them were asked to come up with research ideas on one of seven given topics within ten days.

Meanwhile, the AI ​​idea generator was supposed to search for relevant articles on these research topics with the help of an AI-supported literature search engine. These were then supposed to form the basis for the generator to generate 4,000 ideas for each research topic.

But although the reviewers rated the AI ​​ideas as more original and exciting, a disadvantage emerged. Of the 4,000 AI ideas, only about 200 were actually unique. The researchers therefore assume that the originality of the AI ​​decreases as the number of ideas increases.

What potential does AI have in research?

However, it remains questionable whether AI systems can actually develop research ideas that can keep up with those of experienced scientists in a direct comparison. Chenglei Si, co-author of the study and computer scientist at Stanford University, explains that this is mainly due to the very subjective evaluation of ideas.

We are trying to get the community to think more deeply about what the future should look like when AI can take a more active role in the research process.

This can also only be done by researchers who have sufficient expertise in the respective field. “The best way to contextualize such skills is a direct comparison,” Si explains to Nature.

Also interesting: