According to an AI-powered large-scale analysis of disruptions, women are less likely to speak out on CNN, Fox News and MSNBC

The Research Brief is a brief overview of interesting scientific work.

The big idea

My colleagues and I used artificial intelligence to analyze hundreds of thousands of cable newscast dialogues to better understand the nature of disruptions in political discussions. We found that in these situations, women have far fewer opportunities to express themselves than men, and may therefore interrupt more often than men.

Analyzing disruptions on this scale provides powerful insights into subtle conversational dynamics and how they vary by race, gender, occupation, and political affiliation. In addition to gender differences, we found that conversations between people of opposite political affiliations on CNN, Fox News, and MSNBC are riddled with far more intrusive and unfriendly cuts than conversations between people who share a political affiliation.

I am a computer scientist and I use AI to investigate social science issues. In collaboration with student AI researchers at Carnegie Mellon University, we have developed AI methods that reliably distinguish intrusive and unfriendly interruptions from harmless ones. Intrusive interruptions aim to take over a conversation or choke the speaker, and harmless interruptions aim to assist the speaker with helpful information or signs of approval.

Over years of work, we analyzed 625,409 interrupted dialogues found in 275,420 transcripts from the three cable news channels from January 2000 to July 2021. We found that female speakers on the networks averaged 72.8 words per speaking opportunity compared to 81.4 for male speakers. We also found that female speakers were interrupted in 39.4% of dialogue compared to 35.9% for male speakers. However, women had a better ratio of benign to intrusive disruptions than men: 85.5% to 75.4%.

This political discussion on CNN between people of different genders and political viewpoints has numerous intrusive interruptions.

Why it matters

Our AI techniques could be used to provide real-time interrupt analysis of talk shows, interviews and political debates. Post-debate analysis found Donald Trump interrupted twice as often as Joe Biden during the third US presidential debate of 2020. Real-time analytics can be helpful in calling out serial breakers, informing the audience during debate, and perhaps helping to ensure civil discourse.

We also examined the evolution of unfriendly interrupts over these two decades. This research shows that the rate of unfriendly or intrusive interrupts has gradually increased, with the period during the 2016 Trump-Clinton campaign generating the largest increase in intrusive interrupts among commentators.

This finding points to the widening of the political divide in the US, previously documented in research on news consumption patterns, media portrayals of important issues such as policing, discussions of events on social media, and the language of partisan news audiences.

What other research is being done

Other researchers have examined breaks in political speech in contexts other than cable news broadcasts, including parliamentary speeches.

While disruptions have been extensively analyzed in the social science literature for decades, our study used AI techniques to examine disruptions on an unprecedented scale.

What is not yet known

Interruptions could be categorized more granularly than just considering them as intrusive or harmless. Our current methods are not robust enough to reliably detect these nuances.

Our analysis also suffers from a selection bias as it only considers individuals who have appeared on major news networks and are therefore likely to exert significant social influence. We do not know whether our results would generalize to broader groups, for example from male politicians to all men.


Leave a Reply

Your email address will not be published. Required fields are marked *