Using AI to Predict the Future of AI: Researchers have created a tool to better digest emerging AI research and propose new research ideas.
One of the biggest problems faced by researchers studying AI is the navigation of the vast amount of AI research published every year. Since McCulloch and Pitts first published a paper in 1943 discussing “artificial neurons” that could complete simple tasks, each year has seen an increase in AI research publications. For instance, there was a 34% increase in AI journal publications from 2019-2020.
Noticing the problem
Several researchers, including Mario Krenn from the Max Planck Institute, realized the need to create a tool that could sort, read, and help researchers understand thousands of AI research publications. Using this tool, the research group hoped to make AI research more efficient and productive, and even propose new research ideas to researchers.
Tackling the problem
In developing their tool, the group explored popular AI models, such as OpenAi’s GPT-3, for text processing. Despite this approach being useful for identifying existing ideas, it failed to go to the next level: recommending new concepts based on connections between ideas found in various papers.
In response, the group focused on the creation of a semantic knowledge network, which tied papers from different scientific fields that mentioned similar concepts. Simply put, they created a “web of ideas”, where the AI tool became better at processing large datasets of research publications, thus recommending new research ideas. Moreover, the semantic network approach helped develop a trend in AI research through a measuring benchmark named Science4Cast, which helps graph trends for future AI research.
Interested in learning more? Check out the research below: