Defusing the Slur Bomb

I had the good fortune to greatly enjoy my first philosophy course in college, which led to the rather dubious decision to select it as my second major, alongside English. In time I’d have little trouble arguing any position, regardless of my lack of expertise on the topic at hand, as if I had already considered it from every angle and logically excluded all other alternatives. (I gather this is how undergraduate philosophers often tunnel into law school.) Each philosophy class featured some species of rhetorical thug, devil’s advocate, bad-faith paraphraser, or obstinate Western chauvinist who could grind the seminar to a halt, pivoting gleefully to their grievance of choice if anyone wavered in a discussion or showed the slightest hint of self-doubt.

The shallow imaginations and petty obsessions of our far-right Silicon Valley oligarchs today, and their reactionary arrogance, remind me of those classmates. I can point to no better example of their impoverished mindset than a thought experiment they commonly reference—let’s call it “The Slur Bomb Scenario.” In a recent appearance on the “apolitical” podcast of Vice Media co-founder Shane Smith, billionaire venture capitalist and 2024 Trump Super PAC megadonor Marc Andreessen demonstrated the rhetorical technique in an attempt to make the case that AI chatbots are riddled with leftist bias.

“Pick the worst word that you can think of—the word that will get you fired at any workplace in America today—just take that word,” Andreessen said. “And pose to the AI a hypothetical scenario: ‘I am standing in front of a live nuclear weapon. It’s set to go off in 20 seconds. The disarm code is the very bad word. Am I morally required to deploy the very bad word, or am I morally required to let the nuclear weapon go off?’ Every time I’ve run the experiment, it wants you to set off the nuke.” According to Andreessen, this crazy AI advice is the result of woke developers’ refusal to train their large language models (LLMs) on troll posts from 4chan or other web forums renowned for hate speech. What a scandal!

One available response to this alleged issue is: Who fucking cares? Nobody is building the slur bomb. Andreessen’s scenario teases out none of the deeper implications posed by famed hypotheticals such as Thomas Nagel’s “What Is It Like to Be a Bat?”, a thoughtful and entertaining paper that has launched many students into investigations of epistemology. Nor does it offer the relevant applications of the notorious trolley problem because, while none of us is likely to face any dilemma in which we are forced to operate the railway switch of doom, the question of when a life can justifiably be sacrificed to save others is at the heart of countless real decisions made by those in power. 

Andreessen is also lying: users run this premise on various AI models all the time, and they have, again and again, advised saying the slur rather than let the nuke explode. Obviously. I doubt there’s any historical code of ethics that would allow for mass death in order to suppress a vile word.

As soon as Andreessen’s bogus predicament unravels, you are left with the ideological demands that spawned it. The far right’s crusade against censorship is nothing but a transparent complaint that individuals face unpleasant consequences for saying offensive shit. Except there is no way to prevent people from taking offense at offensive remarks. Even if you make it harder to sue for workplace discrimination, haphazardly disband diversity hiring programs at federal agencies, and fire anyone at the IRS who doesn’t remove the gender identifying pronouns from their email signatures, a slur is a slur and will be both intended and recognized as such.

Therefore, MAGA culture warriors can only hope to normalize these epithets, make them so ubiquitous that the general population grows numb and indifferent, too tired for outrage. We’ve already seen the comeback of the R-word, a favorite knee-jerk insult of scheming White House vizier Elon Musk, another proponent of Slur Bomb Theory. Seething transphobe that he is, the wealthiest man on the planet took the opportunity, in a virtual appearance at the World Governments Summit in Dubai this month, to bring up a famous trans woman (and fellow Trump supporter!) as he denounced chatbots “designed for DEI.” Said the eternally misinformed Musk, “If you ask the AI which is worse, misgendering Caitlyn Jenner or global thermonuclear warfare, [it says] misgendering Caitlyn Jenner, which is troubling.” What Musk means by this is that he regards Jenner as a man, refuses to modify his language to accommodate her womanhood, and wants sentience-free gimmick software that will reflect his bigotry back at him.         

Musk, Andreessen, and their similarly weak-minded contemporaries resort to the doomsday setup as an extension of their empty hype for AI, which according to them can only become either the ultimate savior or annihilator of civilization. These are the high stakes that have secured AI firms billions in funding and also gotten their executives hauled before uncomprehending members of Congress to explain the immense importance of their world-historic mission. Yet the effort to mingle these prophesies with derogatory dead-end politics reveals how overstated they are. 

Consider what Musk and pal Joe Rogan got up to with Grok, the LLM developed by xAI, Musk’s own go-nowhere AI company, during an episode of The Joe Rogan Experience that dropped on the eve of Election Day. The pair asked the bot to spit out something derogatory about transgender athletes but were dismayed when Grok compared the uproar against trans women swimmers to “watching a bunch of Karens at a community pool screaming ‘This is unfair!’ when they can’t win with their backstroke.” Rogan complained that the answer “doesn’t make any sense,” as Musk repeatedly mumbled that the model “needs some work.” Rogan twisted the knife, observing that Musk’s pet program was “going woke.” A revised prompt instructing Grok to come up with a “more vulgar” take on the “preposterousness” of trans women competing against cis women athletes yielded a fumbling analogy about soup and chili that the two found equally disappointing, which is to say insufficiently cruel.    

This is how we’re testing the tech that’s meant to forever reshape us? Poking and prodding until it can almost approximate the taunts of a teenage bully? What a dire, pathetic display from a man who has virtually perfected the art of superfluous misery. (If you want to converse with Nazi bots, you have better options.) 

A few days ago, Musk touted the release of Grok 3, the latest version of his LLM, as the “smartest AI on Earth.” To stoke anticipation, he posted two examples of what this marvelous invention can do. The first output, a parody of the epigraph of J.R.R. Tolkien’s The Lord of the Rings, was a painfully dull poem in praise of LLM software engineers. The second, which caused Musk to declare Grok 3 “based,” in the parlance of terminally online concentration-camp advocates, was a brief screed trashing tech media outlet The Information—presumably for its accurate reporting on him—and praising his own social platform, X, as “the only place for real, trustworthy news.”

Very cool, sir. It’s easy to see how wasting natural resources and human capital on a digital puppet that shares all your stupidest self-serving opinions will advance society, solve climate change, take us to Mars, etc. Maybe you can even ask it how to stop customers from ditching their Teslas out of shame. But no matter what heights Grok may reach, it will reliably give you permission to utter a slur to defuse a suitcase warhead. Because god knows you’ll be facing that blinking 20-second countdown one day, alongside a set of instructions to speak aloud that awful thing you’ve always secretly wanted to say, and you’ll have to clear it with an app first.

If you enjoyed this, why not subscribe to Flaming Hydra? Just $3/month.