Captive Gods: How Tech Companies Are Turning AI Into a Dangerous Cult

Pay no attention to the Boer behind the curtain

Deep in conspiracy culture they have this theory about something they call Project Blue Beam. It was proposed by a Canadian conspiracy theorist named Serge Monast back in the 90s, and the general thrust of it is that the shadowy They who exist as the antagonists of all conspiracy theories are planning to fake a worldwide theological event. Via holographic projections, subliminal messaging, and other technologies advanced enough as to be, as Arthur C. Clarke ventured, indistinguishable from magic, they will trick humanity into believing God is appearing before us all. The false deity will lay down His new commandments, but they will of course be the will of the mortal men projecting the illusion.

This is the story that came to my mind when Grok, the “anti-woke” AI chatbot Elon Musk built into Twitter, started ranting at people about white genocide recently.

To catch you up if you missed all the details: Grok is Musk’s answer to AI bots like ChatGPT and Copilot and it will respond to any question on Twitter that it’s tagged in. Around a month ago, for a full day, it answered every single query in a way that steered the conversation toward a conspiracy theory that South Africa is engaged in a systematic genocide of white people. Any prompt given to it, from questions about Fortnite to requests for soup recipes, triggered a response that drew some connection, however creative and tenuous, to the white genocide of South Africa.

At one point, the Pope got involved.

As you’d expect, once people started to understand what was happening and its comedy potential, it was a pretty fun day on Twitter. Those are kind of rare, nowerdays.

Now, according to an anonymous company rep hiding behind the XAI company handle, what happened here was that an “unauthorized modification” was made to Grok’s code very early one morning by an unidentified employee that compelled it to take a position on this very specific topic.

The “corporate damage control apology to English” translation is that Elon Musk, annoyed that Grok kept debunking his conspiracy theory, messed with its code to tell it to always support his point of view. Unfortunately, he was probably high at the time, so he forgot to specify that he only wanted it to back him up on this when asked about it.

This misadventure, funny as it was, exposed what I feel is a pretty frightening fact to come to terms with: The people in charge of programming these AI chatbots can and do try to manipulate the version of reality that they present as factual, and we only know they’re doing it when they fuck it up.

Of course this plainly intuitive. We know that these are computer programs, don’t we? But if you’re cognizant of this then the number of people who legitimately think these things are omniscient should frighten you.

If you’re unaware of just how creepy Twitter has become since Musk took over—people have completely abandoned the concept of finding things out for themselves in favor of asking Grok to sort fact from fiction. The practice has replaced the Community Notes function, which is barely used anymore. Instead, tens of thousands of users will ask Grok’s opinion, even in lieu of clicking a link to the primary source.

It isn’t just Grok. This is just the most conspicuous example of a trend accelerated by the fact that the developers of AI insist they are less than a decade from achieving conscious superintelligence—a truth machine that can understand our world far better than we ever could, to which we can look to guide us, and perhaps even to lead us. Sam Altman made a grand pronouncement of this just last week. This despite the fact that most experts (not tech CEOs, but scientists) believe this is bullshit. That we are not on the verge of building an artificial superintelligence or anything close to it.