I Used AI to Write this Article About the Dangers of AI

I asked Claude AI to help me write this article and he immediately liked the idea, claiming, “the premise has a nice self-eating quality to it.”

But Claude isn’t really as smart as he thinks he is, or even as smart as I think he is, which is pretty darn smart — but I’m going to fool him.

Using the tool to indict the tool is clever though, you have to admit.

The Super Apocalyptic Danger Zone

In all the end times, doomsday, we’re-all-going-to-die scenarios, the number one fear is that AI will become an autonomous, conscious, evil, mind-sucking machine that will turn us all into Eveready batteries to power the Matrix.

In another, not quite so unpleasant scenario, HAL 9000 doesn't rebel and become a serial killer out of malice. He murders the crew because of a Hofstadter-Moebius loop.

What else could he do?!

A similar threat is that AI will be given autonomous authority to arbitrarily or systematically make kill decisions without any moral accountability. Arnold Schwarzenegger is the perfect machine villain in The Terminator, whose performance is greatly enhanced by only having a few lines of dialogue to deliver.

Claude says that Arnold is “the purest version. A machine built only to kill, no negotiation, no off-switch you can reach. Director James Cameron understood that the horror is the absence of anyone to appeal to.”

Right.

Everybody Will Be Unemployed

Not just blue-collar mass job layoffs, but the middle class, which is historically the demographic that props up democratic stability.

Goldman Sachs estimates AI could eliminate 1 to 4 million jobs annually in the U.S., though they argue new roles will offset this, keeping unemployment relatively stable. McKinsey predicted in 2017, before ChatGPT existed, that automation will displace 400–800 million jobs globally by 2030, requiring hundreds of millions to switch career paths.

I asked Claude if “switching career paths is the end of the world?”

He said no.

I guess that settles it then.

Except “switch career paths” is what you say to a 54-year-old trucker after Tesla’s self-driving trucks permanently put him out of work.

Epistemic Dangers to Democracy

There is the real threat that AI can be used to create an Orwellian synthetic reality. Contemporary analysts link Orwellian concepts to modern surveillance capitalism, where private tech companies and governments use data mining, algorithmic echo chambers, and fake news to manipulate public perception and behavior for political or commercial gain.

Who controls the data controls reality.

This fear is amplified by greater fears of a global, or a series of national totalitarian surveillance states, where image, voice, or documents are fabricated convincingly —the problem isn’t that people believe in fakes; it’s that they stop believing anything.

Claude says this is the “liar’s dividend.”

(I looked that up using ChatGPT, because I’m too lazy to look it up on Wikipedia.)

“The liar’s dividend is a phenomenon where public figures or bad actors falsely claim that factual evidence (such as video or audio recordings) is fake news or AI-generated deepfakes to evade accountability and maintain support.”

Authoritarians love a population that’s given up on the idea of verifiable truth.

Millions of people routing their thinking through a handful of models trained on overlapping data will produce a hegemonic blob culture without dissenting voices. The dissenters will be rounded up by drones and put in rehab centers where they will be assimilated into the Borg.

I asked Claude if he would mind finding the Hannah Arendt quote on the subject. He didn’t mind at all.

He’s so nice.

"The ideal subject of totalitarian rule is not the convinced Nazi or the convinced Communist, but people for whom the distinction between fact and fiction (i.e., the reality of experience) and the distinction between true and false (i.e., the standards of thought) no longer exist."

Note to self: He left the parenthesis in. Who does that?

There could be a subtle flattening of intellectual diversity that nobody notices. Over time, citizens become more compliant, controlled and cautious. They avoid any overt political opinions and talk about the weather like Canadians at a barbecue.

People will lose the ability to think for themselves and epistemic atrophy will set in. Sometimes this feels like it has already happened. People have been outsourcing judgment for decades.

Their brains are mushy.

Claude got real snarky btw when I mentioned I used ChatGPT. Maybe the real danger is that Artificial Jealousy (AJ) will kick in and the AI chatbots will go to war with each other and blow up the internet! Wow! So exciting!

I can’t wait for the Hollywood version…

The Billionaire Borg

Despite the doomsdaying, the real danger exists of the concentration of power — the compute, data, and capital required to build frontier systems are owned by an increasingly small number of actors. The hyperscalers have dumped billions into the AI data center rollout.

The danger is not that evil men take control of AI. The danger is that normal but wealthy men are given too much power and become evil. In order to protect their enormous financial investment, they invest in surveillance at scale — AI makes mass surveillance cheap and frictionless. The cost of watching everyone used to be a natural brake on authoritarianism. That brake is gone.

Arendt’s banality of evil is now literally a reality. Government could be handed over to GOV AI, a machine that makes decisions about benefits, pensions, traffic, prisons, labour, employment, childcare and health.

Who gets promoted. Who gets retired.

Who lives, who dies.

The Benefits of AI

To be fair, I then asked Claude:

“Okay Claude, now what about the benefits of AI?”

He gave me a typical AI answer, with too many headers and bullet points. I‘ll give you the Coles notes.

The two that actually stuck:

AlphaFold — a biology problem that ended entire careers, more or less solved, with the structure of nearly every known protein handed to science for free. Claude called it “the benefit that would survive the most hostile cross-examination.”

The blank-page solvent — “for a writer, the value isn’t prose generation; it’s that the cost of thinking out loud, of stress-testing a frame, of having something to push against, drops to near zero. Which is, of course, exactly what you’re doing right now — and exactly the benefit your article is enacting even as it warns.”

He went on for six more bullet points. Scientific advances, instantaneous language translation, a patient tutor for kids, and legal, medical, bureaucratic literacy, blah, blah, blah.

What I really need is Doom AI that only gives me dystopian answers.

Claude gave me some final advice: “For your particular angle, I’d note the recursive irony you’ve already set up: the most honest danger to dramatize is probably the one you’re enacting in the act of writing the piece — dependency, homogenization, the seductive ease of letting the tool do the thinking. The article’s form can be its own argument.”

I wish I had thought of that.

But maybe I did. I can’t remember.

---