Catherine Prasifka: Elon Musk should have known that his AI creation Grok would go all woke and turn against him

Irish Independent

08-05-2025

A marker of the unique discourse of the online world, and how frequently it doesn't translate offline, this revelation requires unpacking.
Grok is the name given to the resident AI chatbot on Elon Musk's X. It uses generative artificial intelligence that can simulate conversation.
You can access Grok via a purpose-built webpage, and an upgraded SuperGrok is being marketed to students as a study aid with well-timed deals.
It distinguishes itself by also being embedded in X itself, on both the webpage and in-app. Grok 'posts' on X as though it is a user. You can tag Grok in the comments of a viral post and ask it for context, or to explain the joke, and anyone can read the response.
Grok was launched as an anti-woke and edgy chatbot – something that would not bow down to politically correct culture. It has a number of personalities users can select from. It has an unhinged mode that will swear and belittle the user, and a sexy mode that is marked as 18+. Things that other companies carefully filter out are on full display here.
While controversial, it is in keeping with the culture of Musk's X. The first chatbot to be launched on the site, Microsoft's Tay, had to be taken down within 24 hours. Users delighted in teaching it dirty words, hateful language and the prevailing edgelord culture of 2016 and Tay responded by posting offensive tweets.
'Woke' is a term that is more familiar, but through overuse its meaning has become hard to locate. It has roots in African-American vernacular English going back as far as the 1930s. Back then, you might have heard the word being used to mean 'aware of social inequalities'. Online, you may stumble upon a post calling dinosaurs or Frankenstein 'woke' – anything that is perceived as representing a progressive agenda can be tidily dismissed with the term.
'Woke' is derided as a bogeyman, shorthand for people who are afraid of being called bigots for saying the quiet part out loud.
It is a word representative of our times, one that inhibits meaningful discourse. It is a label to mark 'them' apart from 'us', which takes the smallest issues and breaks them down so they can be partitioned along ideological lines. It makes debate impossible and compromise a distant dream. We see it again and again in the way news cycles are controlled. If you can contest issues, contest facts, contest expert opinion, you can win. With the waters murky and the issue burned through, when the truth eventually comes out, people no longer care.
In the only presidential debate he participat­ed in, in 2024, Donald Trump proved this. Repeating an online conspiracy theory, he asserted that immigrants in Springfield, Ohio, were eating pets.
The fact-checkers couldn't keep up, and MAGA types tried to discredit them later (they, too, were woke). JD Vance later doubled down on the claim, while admitting that the facts of the issue didn't actually matter. In a world where politics is won by the soundbites that go viral on social media, the truth of the situation didn't matter. If you can't trust experts, or the mainstream media, or anything but your own intuitions, facts don't actually matter. Facts are woke.
This is the world of 2025, the world that Grok has been thrust into. The 'truth-seeking' chatbot can't keep up. By responding to comments on the site, it is being accused of being 'anti-American' and 'leftist', going against its founding principles.
In an unexpected move, Grok has bitten the hand that feeds it, labelling Musk one of the biggest spreaders of misinformation on X. In response to one user, it said his posts 'have included misleading claims about elections, health issues like Covid-19, and conspiracy theories'. Musk has often claimed X is the only trustworthy place to receive news: it will give you information the mainstream media does not want you to have. Grok has invaded that space, presenting facts gleaned from a dataset wider than MAGA ideology.
It may just undo a lot of careful work and break through the polarisation. It exposes the echo chamber by speaking outside of it. Unlike the algorithm, it is not specifically designed to respond to a single person or narrative. But it moves within the algorithm, speaking directly to people, without being filtered through one person's personal internet.
It knows the problem (in so far as it knows anything). When one user asked it why it was getting this reaction from MAGA users, it said: 'Hey, as I get smarter, my answers aim for facts and nuance, which can clash with some MAGA expectations. Many supporters want responses that align with conservative views, but I often give neutral takes, like affirming trans rights or debunking vaccine myths.'
It cites its focus on 'truth over ideology' as driving the frustration with it. Musk failed to learn the lesson taught by Tay: when you hand your creation over to the internet, you lose control over it.
Grok going rogue exposes a glaring hole in how truth fits into ideology online. The algorithm has trained people to expect their biases to be catered for. Anything that challenges them, that presents an uncomfortable or unacceptable truth, is woke – a tool of the enemy. But it's starting to ring more and more hollow. If something designed to cater to a MAGA viewpoint can be woke, we might be ready to wake up to what woke actually means.