The 163rd Block: Feedback, feedback loop, out of the loop
Alternatively, why Maria Ressa is critical of the "Reuters Institute Digital News Report 2023"
This week…
Your reading time is about 7 minutes. Let’s start.
Reuters Institute for the Study of Journalism (RISJ) released the annual “Digital News Report.” Its fiercest critic? Nobel laureate Maria Ressa, who last year quietly resigned from the RISJ advisory board because of concerns over the report’s methodology. For one, the report rated Rappler, the news site Ressa founded, the least trusted media in the Phillippines. Ressa said the report failed to consider the impacts of disinformation campaigns in countries where independent media is under constant attack by government machinery.
In the Phillippines, the previous administration under the populist Rodrigo Duterte targeted Rappler for its extensive reporting on the government’s corruption, fake news campaigns, and the drug war that led to extrajudicial killings. Ressa herself faced multiple arrests and convictions for “cyber libel” and “tax invasion,” some of which she has been acquitted. As one of the most outspoken critics of Duterte, international media observers consider her legal issues to be politically motivated.
For their part, RISJ director Rasmus Kleis Nielsen said caveats are included in the report to discuss threats to news outlets such as Rappler and noted that “scores should not be seen as a measure of quality or trustworthiness of the content.”
Ressa, though, warned that the report could be weaponised against reporters and undermine free media in corrupt regimes. See the story below. Then check out other top stories on my radar, a few personal recommendations, and the chart of the week.
ICYMI: The Previous Block introduced newsletter redesigns, and Substack’s challenger, WordPress. CORRECTION NOTICE: Fixed broken link.Nobel laureate Maria Ressa says research by Oxford institute can be used against reporters
Emma Graham-Harrison for The Guardian, taken in part:
The approach and methodology that make the report so dangerous to journalists in her country, also provide weapons to any government that wants to attack and undermine free media, Ressa said.
“We are not alone. This ‘study’ is like giving a loaded gun to autocratic governments trying to silence independent journalists not just in the Philippines but in countries like Brazil and India, where information operations and the lawfare are used to persecute, harass, and chill.”
“It admits that social media has taken over distribution, but it is not critical of biases that are implicit in social media and how this has turned the world upside down,” she said.
And more:
Emily Bell, director of the Tow Center for Digital Journalism at Columbia, which studies the impact of social media on journalism, said the report offered only a partial perspective on the contemporary media industry.
“[This report] is interpreted as being a comprehensive view of everything going on in journalism but, if you only take an audience view of the threats to journalism, specifically in markets where you do not have strong protections for a free press, you are at high risk of ending up with a distorted picture,” she said.
“This began as an Anglo-centric, Eurocentric report. They have been able to expand it to include more countries but you do have to question if it can be presented as something comprehensive. I would argue that journalism research should be advocating more strongly for journalism.”
“You also have to look at the funding for the research,” she added.
The report was partially funded by Google, which was also the largest single sponsor.
Vietnamese YouTuber is filling information voids with Newsmax and Breitbart
Lam Thuy Vo for The Markup:
“Let me show you a beautiful photo of Donald Trump as he is playing golf,” [Sonia Ohlala] said in one video published on May 26, 2023. “He bends down and picks up golf balls, like it’s not a problem.[…] He’s a very healthy person at 76.” Even though he has not been president in the past two years, most of Ohlala’s YouTube broadcasts revolve around the former president, translating content from far-right outlets like Newsmax, The Gateway Pundit, or Breitbart, according to her public YouTube community page.
For some Vietnamese immigrants to the U.S., like Oakland resident Duyen Hoang, 56, Ohlala’s videos are one of the few U.S. news sources she can access, because they’re in Vietnamese. Hoang watches Ohlala regularly on YouTube and believes Donald Trump is a good and accomplished man, even if she knows that Ohala’s videos may not always be based on reliable information—or be giving her the local news she sorely needs.
“I really just need more information about my hometown Oakland,” Hoang said.
Hoang granted the author access to her entire YouTube archive to “understand the context of how these videos show up in her feed.”
The AI feedback loop: Researchers warn of ‘model collapse’ as AI trains on AI-generated content
Remember when I asked about what will happen when AI trains on AI-generated content (in a far less elegant way) in “The 161st Block?” Well, here’s Carl Franzen for VentureBeat on the AI feedback loop, discussed in a recently published paper in the open-access repository arXiv:
“Original data generated by humans represents the world more fairly, i.e. it contains improbable data too,” [one of the researchers, Ilia Shumailov] explained. “Generative models, on the other hand, tend to overfit for popular data and often misunderstand/misrepresent less popular data.”
Shumailov illustrated this problem for VentureBeat with a hypothetical scenario, wherein a machine learning model is trained on a dataset with pictures of 100 cats — 10 of them with blue fur, and 90 with yellow. The model learns that yellow cats are more prevalent, but also represents blue cats as more yellowish than they really are, returning some green-cat results when asked to produce new data. Over time, the original trait of blue fur erodes through successive training cycles, turning from blue to greenish, and ultimately yellow. This progressive distortion and eventual loss of minority data characteristics is model collapse. To prevent this, it’s important to ensure fair representation of minority groups in datasets, in terms of both quantity and accurate portrayal of distinctive features. The task is challenging due to models’ difficulty learning from rare events.
You can imagine the implications, especially for the marginalised, but I don’t think anyone is surprised.
What I read, listen, and watch…
I’m reading the “Reuters Institute Digital News Report 2023.”
I’m listening to “Why preserving Indigenous languages is so critical to culture” on The Conversation’s Don’t Call Me Resilient.
I’m watching “The Billion Dollar Scam” on BBC.
Other curious links:
”Generative AI tools are perpetuating harmful gender stereotypes” by Marie Lamensch for CIGI.
“The dark world of deepfaked debt” by Snigdha Poonam for The Dial.
”The illusion of moral decline” by Adam M. Mastroianni and Daniel T. Gilbert, published in Nature.
“‘If they come for me, will they kill me?’ – How a cricket dream became a nightmare for Afghanistan’s women’s team” by Katya Witney for Wisden.
“La transformación digital requiere de acuerdos analógicos” por Matías Bianchi en El País.
« Post-vérité et dépérissement du politique » par France Giroux dans Le Devoir.
Chart of the week
According to the “Reuters Institute Digital News Report 2023,” “Facebook remains the most important network (aggregated across 12 countries) at 28 per cent, but is now 14 points lower than its 2016 peak (42 per cent).”
The report explained that “Facebook has been distancing itself from news for some time, reducing the percentage of news stories people see in their feed (3 per cent according to the company’s latest figures from March 2023), but in the last year it has also been scaling back on direct payments to publishers and other schemes that supported journalism.”
And one more thing
