This Week in Misinformation: Hallucinogenerative AI, COVID Origins Declass, Trump 'Arrest' Canard
23 March 2023
This Prism newsletter strives to be the paper of record for all that’s happening in misinformation in the United States. For any citizen whose life is impacted by misinformation, it helps you see how storylines evolve from multiple, sourced angles on important stories in one place. For amateur and professional misinformation watchers, it is your go-to resource for updates on peers, platforms, propagandists, and politicians. Learn more about Prism and our other products on our Substack page, follow us on Twitter, or like us on Facebook!
________________________________
Keeping up on misinformation is basically the best thing you can do for your brain. So glad you’re here!
Give this newsletter a share, won’t you? Posting a snippet to social can be a good taste for people in your network, or sending this email directly also tends to work well. 🙂
Reliability scores for media outlets cited in the summary are in parentheses for each, courtesy of the terrific folks at Ad Fontes Media.
Now, on to our top stories.
“Hallucinations” was the word of the week in the world of generative artificial intelligence.
It’s becoming better understood that AI likes to give answers that sound plausible, and that its answers are in no way tethered to the constraints of objective reality (Fortune, 44.49). Experts have started calling this tendency “hallucinating” and are documenting lots of examples like Stanford University’s now-retired Alpaca chatbot (Gizmodo, -). Google’s Bard is another; its claim about learning to generate answers by leveraging Gmail content was hallucinated--or so Google claimed (via Twitter).
Because OpenAI is apparently not slowing down enough to deal with even known problems with GPT-3, the newly released GPT-4 is somehow even worse than its predecessor when it comes to misinformation, according to one study (Axios, 44.15). Not promising for those of us who hope these companies can fix this instead of, for example, letting their bots cite one another’s hallucinations in a spiraling doom cycle (The Verge, 44.19).
Over on the AI-produced images front, one can see the enormous problems that all of this leads to. Fake pictures of Trump being arrested, made in seconds with Midjourney by journalist Eliot Higgins, went viral (Associated Press, 48.40) and earned Higgins a suspension from the tool (BuzzFeed News, 42.88). The pictures looked real enough that they are causing problems, and they will only get better from here.
COVID’s origins are coming into focus, but politics are getting in the way of the public gaining better understanding.
None of the theories about COVID’s origins are proven, but there is new evidence out that points, again, to animals, with raccoon dogs as the vector by which it was possibly transmitted to humans (Scientific American, 43.70). Notable: following the hullabaloo about lab leak after an intelligence assessment was leaked a couple weeks ago, Congress passed and the President has now signed a bill that will mandate declassification of information spy agencies have to work with (Reuters, 47.38). Solid development for transparency and, one hopes, for toning down the rhetoric.
It has all gotten very political, and therefore fraught with misinformation and agendas, as the New York Times (42.47) chronicles here. This of course isn’t new; the CDC found itself in the crossfire when the pandemic was just getting started (New York Times, 42.47), and it made its job of deciding on and communicating public health recommendations all the harder. This campaign season, it’s Ron DeSantis who has signaled he will ride anti-vaccine sentiment to victory at the polls (Vanity Fair, 32.40).
And all of this has consequences that can be measured in lives saved or lost, for example the sad tale of a hospital in DeSantis’s Florida (New York Times, 42.47), where “health freedom” activists are agitating for COVID patients to receive worse care. We’ve already seen a couple decades of conspiracy theory-driven opposition to vaccines and other standards of care becoming more mainstream (multiple authors via Lancet), but the way things are headed we seem destined to struggle with this even more going forward.
The former president said he was going to be arrested (but wasn’t).
Donald Trump, apparently reading or hearing about some “illegal leaks” (NBC News, 45.19) and reportedly without knowing anything beyond that (New York Times, 42.47), took to Truth Social to shout (ALL CAPS is a faux pas, sir) that the Manhattan DA would arrest him on Tuesday and it was all a politically motivated witch hunt that his supporters should rise up in protest against. FWIW: We learned days later the DA never planned to indict or arrest him this week, but the grand jury may hand down its decision next week (New York Times, 42.47).
It didn’t take long for the GOP, including House Speaker Kevin McCarthy, to rally behind him (NBC News, 45.19) against the supposedly corrupt prosecutors (there is no evidence of this)--a unity of purpose that caused some to worry we might be headed for another January 6th-type event to prevent the indictment from being enforced (Axios, 44.15).
Only problem is: unlike on January 6th, his supporters now know what happens when you break the law en masse, and they seem afraid to mobilize (BBC, 46.15). There were the usual excuses about the entire thing being a false flag or fed setup (Reutuers, 47.38), but I don’t blame anyone if the real reason is they don’t want to be sentenced to years in prison (CBS News, 44.22).
Some people say they can live without it, but I know you have to have your grab bag: Russia-friendly accounts helped steer Americans to anger over the Ohio train disaster; Jim Jordan is leaning into the anti-anti-misinformation aesthetic; four more Oath Keepers are convicted felons; the Proud Boys’ sedition trial is pretty all-around theatrical; some Jan. 6 Committee staffers are filling in the blanks after their work was published; a Fox producer claims the company coerced her testimony in the Dominion case; friend of Prism Brent Lee is the subject of a pretty neat BBC thing; and Elon Musk, who also bears partial responsibility for online attacks against a BBC disinformation beat reporter, has (unwittingly?) repeatedly boosts a QAnon influencer reinstated after he took over Twitter.
All that, and a lot more, below. This is This Week in Misinformation.
-- Kevin