Administrative Law's Fourth Settlement: AI and the Scrutable State

Imagine the government as a giant, complex machine designed to solve big problems, like fixing traffic, keeping the air clean, or helping people when they get sick. For a long time, the people running this machine (the experts in agencies) have had a tricky job. They need to be super-smart and use advanced tools to fix complicated issues, but if they get too smart and use tools that are too hard to understand, regular people and the judges who watch them can't see what's happening. It's like a magician pulling a rabbit out of a hat: if the trick is too complex, the audience starts to wonder, "Is it real? Is it fair? Who is actually doing this?"

This paper explores a famous problem in government called the "capability-accountability trap." "Capability" means the government's ability to do hard, technical work. "Accountability" means the public's ability to understand and check that work. For over a century, whenever the government got better at doing hard work (like using computers or statistics), it became harder for regular people to understand how it worked. To fix this, the government started adding a lot of paperwork and rules (procedures) to prove they were being honest. But now, there is so much paperwork that the government is slow, confusing, and hard to fix. The paper argues that the government is currently trying to shrink itself to make it easier to understand, but this might make it too weak to solve new, scary problems like climate change or pandemics. The big question is: Can we use a new tool called Artificial Intelligence (AI) to make the government both super-smart and super-transparent at the same time?

---

The Story of the Government's Growing Pains

This paper tells the story of how the American government has tried to manage its own brainpower for the last 140 years. It argues that every time a new technology arrives, the government gets a "growth spurt" in its ability to do things, but it also gets a "fog" that makes it hard for regular people to see what's going on.

The First Three Settlements (The Old Ways)
The author traces three big moments in history where the government had to adapt:

1. The Railroad Era: When trains first zoomed across the country, they were too fast and complex for local judges to understand. The government created expert commissions (like the ICC) to manage them. To keep these experts in check, the law said, "You must keep a record and explain your reasons." This worked, but it started a trend of adding more and more rules.
2. The Great Depression: When the economy crashed, the government needed to help millions of people at once. They used new statistical tools and punch-card machines to manage Social Security and other programs. But because these decisions affected so many people, regular folks couldn't understand the math. The solution was the Administrative Procedure Act (APA), which said, "As long as you follow the right steps and keep a paper trail, you're good."
3. The Computer Age: When computers arrived to handle things like air pollution and nuclear safety, the problems became even more scientific. The government started using complex models. The courts, realizing they couldn't understand the science, decided to just check if the government followed the rules. This led to the famous Chevron decision, which basically said, "If the law is unclear, we'll trust the experts to figure it out, as long as they have a good reason."

The Current Crisis: The "Sedimentary State"
Now, we are in a weird spot. The government is drowning in paperwork. Every time a new rule was added to make things fair, it piled on top of the old ones, like layers of sedimentary rock. This "sedimentary state" has made the government slow and confusing.

• The Problem: The government is so bogged down in rules that it can't move fast enough to fix new problems.
• The Backlash: Recently, the Supreme Court and the President have started tearing down these expert agencies. They are saying, "We can't understand what these experts are doing, so we are going to shrink them down to a size we can see." They are removing protections for independent experts and forcing more power back to the President or the courts.
• The Risk: The paper warns that this is a bad idea. By making the government smaller and simpler just to make it "scrutable" (easy to see), we might lose the ability to solve the massive, complex problems we face today, like climate change or new diseases.

The New Solution: The "Fourth Settlement" with AI

The author suggests that we don't have to choose between a smart government and a transparent one. We might be able to have both, thanks to Artificial Intelligence.

AI as the "Translator" and "Flashlight"
The paper proposes that AI can act as a special kind of infrastructure that makes the government's complex work understandable without making it simple.

• For the Public: Imagine a chatbot that can read a 200-page government rule about air pollution and explain it to you in plain English, telling you exactly how it affects your family.
• For the Judges: Imagine a tool that can read thousands of comments on a rule and summarize the main points, or check if the government's math actually adds up.
• The Goal: Instead of just checking if the government followed a checklist (procedure), AI could help us check if the government is actually doing the right thing (substance).

The Three New Rules for an AI Government
To make this work, the paper suggests three new rules for how the government should use AI:

1. The "Model and System Dossier" (The ID Card): Every time an agency uses an AI system, they must keep a detailed "ID card" for it. This document would explain what the AI is supposed to do, how it was tested, what mistakes it might make, and who is responsible for it. It's like a nutrition label for a government algorithm.
2. The "Material Model Change" Trigger (The Update Alert): AI systems change and learn over time. The paper says that if an AI system changes in a way that significantly alters its decisions (like changing how it approves disability benefits), the government must announce it and explain why. It's like if a car's engine suddenly started running differently; you'd want to know before you drive it.
3. Deference to Audit (The Trust-but-Verify Rule): Instead of judges trying to understand the complex math behind an AI decision, they should check if the agency hired independent experts to "audit" the AI. If the agency can show, "We tested this thoroughly, and here is the proof it works fairly," the courts should trust it. This shifts the focus from "Did you follow the paperwork?" to "Did you prove your tool is safe and fair?"

Why This Matters
The paper argues that this approach could break the "trap" that has held government hostage for a century.

• It suggests that AI could lower the cost of understanding. Right now, understanding a complex government decision is hard and expensive. AI might make it cheap and easy.
• It suggests that we might eventually move away from just checking paperwork and start checking if the AI is actually aligned with the public's goals.
• It warns that this isn't a magic fix. If the AI is biased, or if the government uses it to hide bad decisions (a problem called "value laundering"), we need strict rules to stop that.

What the Paper Does NOT Say
The paper is careful not to promise that AI will solve everything overnight.

• It does not say that AI will replace human judges or politicians. Humans must still make the final decisions on big, moral questions.
• It does not claim that AI is currently perfect at explaining itself. The technology is still developing, and there are risks like "hallucinations" (where AI makes up facts).
• It does not argue that we should stop all the old rules immediately. Instead, it suggests using AI to make the rules work better and less painfully.

The Big Picture

The story here is one of hope mixed with caution. The government is currently stuck in a cycle where it tries to fix its complexity by making itself smaller, which might leave us vulnerable to future crises. The author suggests a different path: use Artificial Intelligence not just to do the work, but to shine a light on the work. By creating a new set of rules that require transparency, testing, and auditing of AI tools, we could build a government that is smart enough to handle the future but clear enough for everyone to understand. It's a "Fourth Settlement" that tries to have our cake and eat it too: a government that is both capable and accountable.