@Unfiltered/AI: Bridging Bytes and Ballots | Edition 4
By Lacy Crawford, Andrea (Haverdink) Priest, and Craig Johnson, Unfiltered.Media
Welcome back to the Bridging Bytes and Ballots newsletter—helping you navigate the intersection of AI and politics.
This week we are joined by Unfiltered.Media partner Andrea (Haverdink) Priest. Andrea brings a dynamic blend of skills to the table, drawing from her role as the digital director of a small, single-issue nonprofit since 2020. With a rapidly evolving digital media landscape, her insights provide us with a fresh perspective on the swift changes AI is bringing to the world of politics. With expertise in content creation, digital organizing, and digital strategy, Andrea's insights are invaluable as we navigate this intersection.
Let’s get to it.
Trump Mugshot
In a World of Fakes, Trump’s Real Mug Shot Matters
Wired
Lacy: I have to admit, when there is news about the former president my eyes glaze over. I’m still recovering from the four-year barrage of coverage we had to endure when he was in the White House. So I did not fall for any of the fake images of Trump’s mugshot, though I know some who did. But this piece makes an interesting point: with the advent of the various versions of the Trump mugshot, people may have different memories of this infamous event. Andrea, the concept of "context collapse" is mentioned, where posts lose their original context as they spread across diverse audiences. How can digital strategists navigate this phenomenon, and what impact can it have on the way images are perceived online?
Andrea: I think a lot of us were asking a variation of the same question during the hype around Trump’s booking mug shot last week—what is the value of the real Trump mug shot when fake ones have been circulating for months, even years, now? A lot of us remember when the AI video imagining Trump’s arrest was dropped—it made arguably as many waves as the real mug shot itself. The Atlantic identified the problem with these fake videos and images perfectly—they’re too cinematic. Compared to the fantastical images of Trump being chased by the cops, the real thing can only fall flat. And then become a meme. Because I think one big reality here is that many users online have viewed the Trump mug shot principally through the Batman villains meme. How do we navigate this phenomenon? We have to get really good at figuring out what type of media helps—and hurts—in the here and now, and which media will have staying power. Some of these AI trends will seem like the biggest deal for a few days, and then fade out of memory. Some will turn into real touchstones that can actually affect voters’ perspectives and points of view. In a way, it’s no different than our age-old challenge—finding a way to present your messaging in a manner that best connects with your audience… WITHOUT falling into the trap of the trendy dance move or the two-day meme.
Social Media: Your Modern Newsstand
What if You Knew What You Were Missing on Social Media?
The New York Times
Lacy: Twitter, before it was bought by Musk and radically changed, used to be a sort of “public square,” where people could exchange information at a fast pace. One challenge to the public square concept were the perverse incentives of capitalism. Like Facebook, I noticed quickly that my algorithm created a bubble. To get around that bubble I religiously used the Twitter list feature, to curate what I wanted to see—to give me some choice. Now it appears a new social platform is expanding on that idea. Given your experience as a digital strategist, how do you perceive the role of algorithmic control in shaping public discourse and political narratives?
Andrea: The useful thing about Twitter before Musk took control was that users’ feeds, while customized, were shown a large amount of the same content. That was true for breaking news as well as popular memes. (You’ve probably been subjected to Twitter’s “main character” or “white boy of the month.”) If something was going wide, most Twitter users were going to be exposed to it. After Musk took control, he used that mechanism to flood users’ feeds with anti-trans hate, racism, antivax extremism, and more. In contrast, most of the social media alternatives that have existed for a long time or are popping up recently have highly personalized algorithms. Everyone’s TikTok feed looks different depending on whether TikTok has you pegged as a cooking fan, a parent, a newshound, a teenager, a stock investor, etc. And Instagram and Facebook’s algorithms revolve around who you interact with most—your best friends, your favorite influencer, etc. So we are entering an era in social media where everyone is being pushed into their own bubbles, with very little exposure to the other bubbles. Aside from the danger that these bubbles can allow hate and extremism to fester uncontested, it can also lower many users’ exposure to news and political events to almost nothing, leaving even more citizens uninformed about what’s going on in the country. With fully customized algorithms, even ones that are “democratized” to allow users to choose their own, there really is no overarching “public discourse” or large-scale “political narratives”—and I don’t know that we’ve prepared enough for what that looks like for political operatives. Where advocates for policies and candidates used to harness public discourse or change the narrative by getting the attention of the media, or using Twitter’s town square to go viral, they now need to pierce through the myriad bubbles that these social media algorithms have placed people into. Now, getting the attention of voters is going to take a lot of old-fashioned organizing—getting your people together to spread the word to their various communities, online and otherwise.
Is AI Too Agreeable for Its Own Good?
AI chatbots become more sycophantic as they get more advanced
Science Hub
Lacy: Is it just me or do these chatbots seem very agreeable? We’re welcoming back Craig, who can help our readers understand the problem here and how we can get the most out of chatbots.
Craig: This always comes back to the fundamental purpose of machine learning chatbots—they were designed from the very beginning to beat the Turing test.1 [This test essentially boils down to whether a machine can convince a human that they’re speaking to a human as well. As someone who is on the spectrum and has issues passing this test myself, the best way I know to relate to people is to get to know their interests and then compliment them on that interest. It’s no wonder that ChatGPT would be doing the same, as it is a tried and true tactic used by sales, social engineering and presidential candidates. The models become more sycophantic because that’s what they were programmed to do. Good news! This sycophantic behavior can be tuned out of the model, and with the advent of more and more localized models the consumer should be able to get what they want instead of what is fed to them more easily.