Artificial intelligence can write a believable and fun headline with a little help, according to research in The BMJ’s Christmas edition.

AI was trained to write headlines based on the most popular stories in the previous Christmas editions of The BMJ and it came up with amazing headlines.

The algorithm was able to generate headlines that sounded like they could have true scientific merit after some human curation.

According to a study published in The BMJ’s Christmas edition, artificial intelligence (AI) technology can produce convincing, engaging, and scientifically fascinating titles for possible research articles.

A study of The BMJ’s most popular Christmas research pieces, which mix evidence-based science with lighthearted or humorous themes, found that AI-generated titles were just as appealing to readers, but that human input, like in other areas of medicine, improved results.

As a result, the researchers believe AI could play a role in developing hypotheses or future study areas.

AI is already being used to assist doctors in diagnosing diseases, and it is based on the principle that computer systems can learn from data and find patterns. Can, however, AI be used to produce useful hypotheses for medical research?

To find out, the researchers applied the names of The BMJ’s 13 most viewed Christmas research papers from the previous ten years to construct similar AI-generated titles, which they then judged for scientific merit, fun, and plausibility.

The top ten and bottom ten AI-generated titles were then merged with ten real Christmas research articles and scored by a random sample of 25 doctors from Africa, Australia, and Europe.

The findings reveal that AI-generated titles are at least as entertaining (64 percent vs. 69 percent) and engaging (70 percent vs. 68 percent) as real-world titles, despite the fact that real-world titles are assessed as more genuine (73 percent v 48 percent ).

Overall, AI-generated titles were judged as having less scientific or educational worth than real titles (58 percent vs. 39 percent), but this difference was no longer significant when the AI output was filtered by humans (58 percent v 49 percent ).

According to the scientists, this discovery is consistent with past AI research that suggests the best results come from mixing machine learning with human control.

Among the AI-generated titles with the highest believability scores were “The clinical effectiveness of lollipops as a treatment for sore throats,” and “The effects of free gourmet coffee on emergency department waiting times: an observational study.”

The funniest AI-generated title was “Superglue your nipples together and see if it helps you to stop agonising about erectile dysfunction at work,” while the authors point out that this demonstrates AI’s failure to see a study’s real-world application and to understand if titles are insulting.

They admit some limits, but believe that “AI has the potential to generate plausible outputs that are engaging and could attract potential readers” even in the context of odd titles like those found in The BMJ’s Christmas issues.

They do, however, emphasize the significance of human intervention, “a finding that mirrors the potential use of AI in clinical medicine, as decision support rather than as outright replacement of clinicians,” they concluded.

Source: 10.1136/bmj-2021-067732

You were reading: Can AI generate a believable and entertaining research headline?