Artificial intelligence can generate plausible, entertaining and scientifically interesting headlines for research papers


Find the difference: Can AI generate plausible BMJ Christmas headlines?

The AI-generated titles were as engaging and entertaining to readers as the real titles, but performance was enhanced by human intervention; AI could play a role in generating hypotheses or guidance for future research.

Artificial intelligence (AI) technology can generate plausible, entertaining and scientifically interesting headlines for potential research papers, according to a study published in the Christmas issue of BMJ.

A study of the BMJ’s most popular Christmas research papers – which combine evidence-based science with light or quirky themes – found that AI-generated headlines were also appealing to readers but, as in other areas of medicine, performance was improved by human input.

As such, the researchers say AI could play a role in generating hypotheses or guidance for future research.

AI is already being used to help doctors diagnose diseases, based on the idea that computer systems can learn from data and identify patterns. But can AI be used to generate interesting hypotheses for medical research?

To find out, the researchers used the titles of the 13 most-read BMJ Christmas research papers over the past 10 years to invite similar AI-generated titles, which they noted for their scientific merit, their entertainment and their plausibility.

The 10 highest rated and 10 lowest rated AI generated titles were then combined with 10 real Christmas research papers and were reviewed by a random sample of 25 physicians from various specialties in Africa, Australia and in Europe.

The results show that the AI-generated headlines were rated at least as pleasant (64% vs. 69%) and appealing (70% vs. 68%) as the real headlines, although the real headlines were rated as more plausible ( 73% versus 48%).

The AI-generated titles overall were rated as having less scientific or educational merit than the real titles (58% vs. 39%), but this difference became insignificant when humans staged the release of the game. ‘IA (58% vs. 49%).

This finding is consistent with previous work on AI suggesting that the best results come from the combination of machine learning and human monitoring, according to the authors.

Among the headlines generated by the AI, the most plausible were “The clinical effectiveness of pacifiers as a treatment for sore throats” and “The effects of free gourmet coffee on emergency room wait times: an observational study” .

The funniest AI-generated headline was “Superglue Your Nipples Together and See If It Helps You Stop Having Erectile Dysfunction At Work,” although the authors note that this illustrates AI’s inability to see the actual application of a study and understand if the headlines are offensive.

They recognize some limitations, but say that even in the context of original headlines such as those appearing in the BMJ Christmas issues, “AI has the potential to generate plausible results that are engaging and could attract potential readers.” .

However, they stress the importance of human intervention, “a finding that reflects the potential use of AI in clinical medicine, as a decision aid rather than as an outright replacement for clinicians,” they conclude.

Reference: “Research: Ghost in the machine or monkey with a typewriter — Generating titles for Christmas research articles in The BMJ using artificial intelligence: observational study” December 15, 2021, BMJ.
DOI: 10.1136 / bmj-2021-067732

Comments are closed.