Visual Narratives: A Fracturing of the Nightly News
When we tune into television news, how different are the stories and perspectives we see on each channel? This fundamental question lies at the heart of our understanding of the role the media play in our increasingly divided nation. Since Donald Trump’s election, the tone of news coverage has become darker and the media has fractured, with parallel universes that extend even to the pronouns each outlet uses.
What would a presumably nonpartisan artificial-intelligence viewer make of a decade of evening news broadcasts -- and what might its findings tell us about how the visual landscape of the news is diverging?
To explore this question, Google’s Video AI was used to “watch” the evening news broadcasts of ABC, CBS and NBC from July 2010 through present and describe the objects and activities it observed second by second. The source of all the broadcasts was the Internet Archive’s Television News Archive and they were analyzed by the GDELT Project in a special non-consumptive digital library system.
In all, Google’s AI watched 9,615 broadcasts totaling more than 18 million seconds of airtime. Each second contains a list of what the AI saw in it, such as police, protesters, or a podium. Unlike other tools, Google’s AI does not perform facial recognition, so it can’t recognize people, only basic objects and activities. Put together, Google’s AI recognized 11,215 distinct objects and activities and generated 477 million total descriptions, averaging around 27 descriptions for each second of airtime.
How do the three networks compare? The complete list of all descriptions found in each evening’s broadcast were compiled into a ranked list and compared using a Pearson’s correlation, which measures how similar they are (using a 0 to 1 scale, with the closeness to 1 indicating the greatest similarity).
How have those similarities changed over time? The timeline below shows the daily Pearson’s correlation of the three networks since July 2010 using a 14-day rolling average to make the trends more visible. (Click to enlarge).
From the 2010 start of the data through Barack Obama’s reelection in November 2012, the three networks moved largely in lockstep. In the weeks following Obama’s reelection, however, the networks became more divergent in their visual narratives. Similarly, in the aftermath of Trump’s election, the three networks became less and less similar over the course of his first year in office, before rebounding in late 2017.
The last two months appear to mark the lowest similarity of the three networks during the past decade. The timeline below zooms into this period from November 2019 to present.
The separation begins around Dec. 12, 2019, as the House debated the articles of impeachment, reaching its lowest point on Jan. 9, the day after Iran’s milder-than-expected retribution for the U.S. killing of Qassem Soleimani pulled the two nations back from the brink of hostilities. By the time of Trump's acquittal in the Senate, the three networks were back to their previous similarity levels.
What do these parallel media universes look like?
Take the evening of Dec. 18, 2019, in which all three networks focused heavily on impeachment. ABC’s broadcast that night emphasized imagery of individual politicians, with close-ups of Trump, Nancy Pelosi, Chuck Schumer and others, along with a segment on the weather. CBS opted for imagery of individuals, but rather than focusing on the impeachment participants, it interviewed ordinary citizens across the country. NBC used a split-screen arrangement, juxtaposing a wide shot of the House with close-ups of the major figures and commentators.
In short, all three networks covered the same story but used very different imagery to illustrate it.
On a more typical evening following a major news story, there is far more variety in the topics each network covers. Take the evening of Jan. 3, 2020, after the airstrike on Soleimani. All three networks opened with stories about the killing and its potential to widen into broader hostilities. ABC then moved on to the Australian wildfires and Cleveland kidnapping victims. CBS focused on U.S. justifications for the attack, a stabbing in Texas, divides in the Protestant Church, warnings about the flu season and a lighthearted look at a cat rescue facility. NBC focused on Iranian threats of revenge, the Australian wildfires and Carlos Ghosn’s escape from Japan.
While the major story of Iran-U.S. hostilities headlined all three networks, the rest of their broadcasts focused on very different stories, offering largely parallel ideas of the day’s biggest events.
Putting this all together, what can AI tell us about the parallel universes of the media? Perhaps the most important answer is that even when telling the same story, networks can use very different imagery that focuses our attention on very different elements of the same narrative. In the end, the graphs above remind us just how different the view can be depending on what station we tune into.