The Will to Resist

Trust Me, Bro: Artificial Intelligence and the Never-Ending Circus of Administrative Bullshit

Humanity has mastered the art of creating problems, inventing solutions, and then creating entirely new problems with those solutions. Now we've given computers the ability to participate.

Well, shit.

The Human Problem Factory

Humans are fascinating creatures.

We identify a problem, develop a solution, implement that solution, discover seventeen new problems, and then establish a committee to investigate why the original solution didn't work.

It's practically a biological instinct.

We build systems to make life easier, then make those systems so complicated that we need another system to explain how to use them.

And somewhere along the way, we convince ourselves that this is progress.

Government bureaucracy is particularly talented at this.

A decision gets made. A process begins. Paperwork gets generated. More paperwork is required to explain the original paperwork. Eventually, someone needs paperwork to request the paperwork explaining why the first paperwork exists.

A system can follow every procedure correctly and still deliver a miserable experience.

That's worth examining.

Bureaucracy: The Side Quest Nobody Asked For

Imagine trying to understand a financial decision made years ago.

You request the original records. You want to see the determination, the calculations, the correspondence, and a clear accounting of what happened.

Seems reasonable.

Then an invoice arrives.

Now you're not just trying to understand the original decision. You're disputing the cost of obtaining the records needed to understand that decision.

The side quest has become the main quest.

And the main quest is still waiting for an answer.

This is an illustrative scenario, inspired by the kinds of administrative frustrations people encounter. It isn't a claim that every agency or employee operates this way.

The larger problem is that systems can become so focused on processing requests that they lose sight of the people making them.

Sometimes the procedure becomes more important than the purpose.

And that's when technology might actually help.

Enter the LLM

Large Language Models, or LLMs, have the potential to make complicated administrative systems significantly easier to navigate.

Not by replacing human judgment.

Not by making legal decisions.

And certainly not by giving a chatbot a government badge and letting it run wild.

I'm talking about using AI as an assistant.

Imagine an LLM helping an agency:

Imagine requesting your records and receiving a response like this:

"We located 27 relevant documents. Here's the original determination, the employer's response, the appeal history, and the payment accounting. We also identified three discrepancies requiring human review."

Now THAT is useful.

Instead of making someone navigate a bureaucratic maze, technology helps them find the exit.

What a concept.

Don't Automate the Injustice

Of course, there's a catch.

Artificial intelligence isn't automatically intelligent just because it can communicate convincingly.

An LLM can misunderstand a regulation, overlook a document, misinterpret a payment, or confidently produce an answer that's completely wrong.

And that's where things can become dangerous.

Automating a bad decision doesn't make it a good decision. It just makes the bad decision faster.

AI should not independently determine someone's eligibility, deny an appeal, or establish a financial obligation without meaningful human oversight.

People deserve evidence. They deserve explanations. They deserve the ability to challenge mistakes.

And someone needs to be accountable when things go wrong.

That accountability needs to be more than a checkbox on an employee's screen.

If AI helps produce a consequential recommendation, there should be a way to identify:

A system should preserve enough of that history for someone to reconstruct what happened.

If an agency tells someone, "The computer made that determination," the next question should be:

"Which human being approved it?"

Because accountability doesn't disappear just because the paperwork has a digital signature.

Who Audits the Auditor?

Now we're getting to the fun part.

We introduce AI to help employees manage administrative workloads.

Then we need another AI to audit the first AI.

Then we need a human to review the second AI.

Then we establish a department to investigate discrepancies between all three.

Before long, we've created an entirely new bureaucracy to supervise the technology that was supposed to reduce bureaucracy.

I can already picture the conversation:

"Your request has been reviewed by our automated administrative assistance system."

"Was the answer correct?"

"The system indicates that it was."

"Can I see the evidence?"

"Please submit Form 47-B."

"Where do I get Form 47-B?"

"From the automated administrative assistance system."

And somewhere in the server room, a computer whispers:

"Trust me, bro."

We've officially reached the point where artificial intelligence needs its own customer service department.

Humanity really does know how to keep itself busy.

Technology Should Work for People

I'm not against complicated technology. Some problems genuinely require complicated solutions.

But there's a difference between complexity that serves a purpose and complexity that exists because nobody bothered to improve the process.

AI could help make medical bills easier to understand, employment records easier to organize, tax documents easier to review, and government procedures easier to navigate.

It could help an employee find a missing document instead of sending someone through another round of paperwork.

It could help a person understand why a particular decision was made and what evidence supports it.

But it must be transparent.

Its conclusions should be traceable to actual source documents. Its limitations should be acknowledged. Its mistakes should be correctable, with a record of what changed and why.

And humans must remain responsible for consequential decisions.

The objective should be simple:

Make systems serve people instead of making people serve systems.

The Punchline

Humanity has spent centuries creating increasingly sophisticated ways to manage problems we've created for ourselves.

Now we've developed machines that might actually help us untangle some of that complexity.

That's exciting.

It's also a little terrifying.

Maybe the real challenge isn't teaching artificial intelligence to think like humans.

Maybe it's teaching humans to build systems that don't require artificial intelligence just to explain how they work.

Until then, we'll keep inventing solutions to the problems caused by our previous solutions.

We'll create committees to supervise those solutions, departments to manage the committees, and digital assistants to explain why the committees need another meeting.

And if the AI tells us everything is working perfectly?

Ask it for the receipts.

Because the future might be automated, but accountability should never be optional.

Don't just trust the machine. Verify the paperwork.

Jerry's Marginalia

Humanity invented artificial intelligence to make life easier.

Somewhere, a government agency is probably using it to generate a 14-page PDF explaining why you need to submit a 15-page PDF to request the original 14-page PDF.

Progress, apparently.

Here's the thing: I don't want an AI that blindly agrees with me. I want one that can show me where it got its information, admit when it doesn't know something, and tell me when I'm wrong.

That's not asking for a miracle. That's asking for a useful tool.

Keep your receipts, question your paperwork, and never let a machine convince you that a missing document is a philosophical concept.

Signed,

Jerry "The Ankle-Biter" Silverhand
Chief Instigator of Common Sense
Department of Petty Affairs
Raccoon OS — Bureau of Administrative Nonsense

"If the system can't explain itself, it doesn't get to call you the problem."