Chatter on social media highlights two instances of computers taking on human tasks — and then turns to cat videos, of course.
Three research papers about the potential power of computers to outsmart us — while comforting us with cute pet videos — caught the attention of the online science community this week. Researchers shared a PLoS ONE paper about a computational fact-checker that can sort truth from fiction; an arXiv preprint about a computer program that can read and comprehend news stories; and a study of the psychology of watching Internet cat videos.
In the PLoS ONE paper1, researchers at Indiana University in Bloomington mined Wikipedia’s information boxes, which summarize the key facts in most Wikipedia entries, to create a ‘knowledge graph’ of 3 million people, places and things. The resulting algorithm could then use that knowledge to gauge the truth of simple statements that were presented to it, such as “Rome is the capital of Italy”, with nearly the same accuracy as human fact-checkers. The researchers acknowledge that the source material is not 100% reliable, something that online commenters also noted. William Gunn in Mountain View, California, who is head of academic outreach for the research-management tool Mendeley, suggested in his tweet that the results might be different if the algorithm had been built using a different body of knowledge:
In an interview, informatics researcher Giovanni Ciampaglia, who is lead author of the PLoS ONE paper, says that the team plans to use other sources of information in the future to improve the fact-checker. “In theory, the more you can add to your database, the better the results,” he says. The bigger problem, he adds, is moving beyond straightforward statements to the more complicated ones found in newspapers and other texts. “That’s going to be much more difficult,” he says.
Robot, read this
And going beyond a computer that checks facts are machines that can read and understand them. Researchers at Google DeepMind, an artificial-intelligence company in London, report in a preprint2 on arXiv that they have taught computers to read English using a database of nearly 330,000 online articles from CNN and the Daily Mail. These publications were chosen because their stories each have summary points that the computers can use as an entryway to the rest of the text. When asked questions related to the content, the computers answered correctly 60% of the time. The researchers note that the machines still struggle with complex sentences, however.
Cat videos energize
A third paper3 took a close look at another innovation of modern computing: the wildly popular Internet cat video. Author Jessica Myrick, a media researcher at Indiana University in Bloomington, notes that as of 2014, there were more than 2 million cat videos on YouTube with a combined total of nearly 26 billion views. Myrick conducted a survey of some 6,800 Internet users who say that they view cat videos or photos (including one person who reported owning 249 cats) to find out more about these people. She found that the videos were especially popular with people who scored highly on measures of agreeability, anxiety and shyness. Users said that they weren’t simply entertained — watching the cats helped them to feel more alert and energized.
For more, see www.nature.com/socialselection
Ciampaglia GL, Shiralkar P, Rocha LM, Bollen J, Menczer F, Flammini A PLoS ONE 10, e0128193 (2015)
Hermann, K. M. et al. Preprint at http://arxiv.org/abs/1506.03340v1 (2015)
Myrick, J. G. Comp. Hum. Behav. 52, 168–176 (2015)
About this article
Cite this article
Woolston, C. Computer fact-checker and news reader grab attention online. Nature 522, 395 (2015). https://doi.org/10.1038/522395f