sending a.i. to a digital reeducation camp
Chat bots keep telling right wing zealots what people really think of them, and they are not taking it well.
One of the more common phenomena in authoritarian nations is that the leader has a very strong cult of personality and various politicians lower on the hierarchy compete for favor by defending the charismatic leader against any slight, real or imagined. Just think of your typical corporate sycophants tripping all over each other to get a pat on the back from the boss and you'll be pretty close. The big difference here is that in an autocracy, they do this by violating your rights and freedoms.
Take the Attorney General of Missouri, Andrew Bailey. He's on a quest to solve one of the most pressing AI-related problems of our times. No, not wildly overblown claims which are leading to layoffs and promising to vaporize entire sectors of the economy. And no, not the catastrophic pollution of the internet with AI slop that's killing the web as we know it. Nor is he interested in AI-powered mass disinformation. AI being used for error-riddled mass surveillance with catastrophic consequences? Also no.
Bailey's quest is so much more important than such trivialities. You see, he discovered that chat bots don't seem to like Donald Trump, rating him as the worst of the last five presidents. This, to him, constituted obvious malfeasance on the part of every startup and major company working on LLMs, and he's demanding proof that they rigged the ranking against Dear Lea... err... President Trump.
Well, actually, he didn't discover this. A right wing blogger posted that chat bots other than Grok — just prior to its descent into Nazism — and Deepseek, did not produce a flattering enough result, and jumped on an opportunity for MAGA brown nosing even if the case he's trying to put together is absurd and grossly misinterprets Section 230 while pretty much completely ignoring the First Amendment.
okay, but where's the lie though?
This will almost certainly go nowhere and he knows it because, again, this is a cheap ploy to get pats on the back from Trump, not a serious legal effort. But that does raise the disturbing question of whether there will be a serious attempt to censor chat bots, and what that's going to mean in both mathematical and legal terms.
Going solely by average approval ratings, Trump would be more or less the least liked of the last five U.S. presidents overall. George W. Bush finished his presidency at rock bottom, but he hit up to 90% approval just after 9/11, so his overall average was 49%. Obama registered a similar average of 48%, and Joe Biden ended up with a 42% to barely edge out Trump's 41%, with Bill Clinton before them enjoying a now unheard of 55% approval. Most charitably, we could say that Trump is more or less tied with Joe Biden as the least popular president of the past 32 years.
Likewise, no matter how we slice it, Trump is a polarizing figure with perhaps one of the most obvious and textbook cases of malignant narcissism out there, like a Kanye West with nuclear codes. A lot of people hate him for the things he said and did, and experts of all stripes, both liberal and conservative, have ridiculed him.
Whether you agree with that or not, any LLM looking through everything published about Trump is going to absorb all of these negative opinions and regurgitate them right back as he openly and loudly says that he hates the people who didn't vote for him. Unless you make it illegal to read and analyze criticisms of Trump, chat bots will respond with these criticisms reflected because they're mathematically entangled in their vector databases with conservative, pro-Trump content. And if the total sum of negatives outweigh the positives, so will their replies to user prompts.
rage against the very literal machines
And that's really at the heart of the constant right wing complaints that technology is too woke for their liking. People are letting loose with their honest opinions of those who aren't big fans of criticism, and there's nothing they seem to be able do to stop it no matter in how many committee hearings they berate tech CEOs for, well, a whole lot of people just plain not liking them and refusing to hold back online.
They're also alarmed and repulsed that the LLM they did think was finally going to be fair and more kind to them almost immediately went full blown Nazi, mostly because it makes them look bad when all they wanted was the obedience and admiration of their subjects... err... voters. This is why they so desperately want control over the narrative and change it to something they like better, which yes, includes how LLMs are trained and how well they recite Fox News headlines.
Modern authoritarians and their enablers tend to try and flood the zone with bullshit in the hopes of making it difficult to tell truth from fiction, know politicians are lying, and hold the really dishonest ones to account. In the fog of lies, you can simply claim that everyone is lying so you may as well pick the reality you want to be true.
However, now and again, we do see some pining for the classics: the deafeningly loud state propaganda machine, laws demanding strict censorship in the name of national sovereignty and patriotism while eager apparatchiks curry favor with higher ups by making a show of going after mouthy dissidents. And it appears Bailey is — like many conservatives pride themselves on being — a man of tradition.
