Photo By: Marco Oriolesi
A lot of the conversation about “AI slop” in Congress is focused on whether lawmakers should use AI to write laws. That is an important question, but I think there is a bigger issue that we are missing.
The problem is not only that AI could create bad laws. The problem is that the government already has a huge amount of laws, regulations, policies, guidance, contracts and other rules that can be hard to understand. Now we have technology that can create even more of that material very quickly. Before we use AI to add more rules, we should use it to understand the rules we already have.
Joanne M. Frederick, CEO of Government Market Strategies, puts it well: “We shouldn’t use 21st-century technology to accelerate a 20th-century habit of layering new rules on top of old ones.”
This is something government contractors deal with all the time. Before they can respond to a government requirement or build something new, they have to figure out what the rules actually say. That can mean going back to the original law or regulation, looking at later changes, checking other documents that are included, and figuring out which requirements actually apply. Sometimes different rules do not even fit together very well.
So the first step is not always to create something new. Sometimes the first step should be understanding what is already there.
The same idea should apply when the government starts using AI to create new laws and policies.
For decades, the government has added new rules on top of old ones. Congress passes a law. An agency creates regulations to carry it out. Later, Congress changes the law. The agency changes its regulations. Then more guidance is added. Another law creates an exception. A court makes a decision that changes how part of the law is understood.
Over time, this can create a very complicated system.
Some of that complexity is necessary. Government has to deal with complicated problems, and rules need to cover a lot of different situations. But it can also become difficult to understand how everything fits together.
What rules already apply? Which agency is responsible? Has the government already tried to solve this problem? Is there already a program for it? Do two rules conflict? Would a new law actually help, or would it just create another layer?
These are questions AI could be really good at helping answer.
Instead of asking AI to immediately write a new bill or regulation, lawmakers could first ask it to help them understand the system they already have. An AI system could look across laws, regulations, agency guidance and court decisions. It could find similar rules, point out possible conflicts and show when a new proposal might duplicate something that already exists.
That would be a different way of using AI.
It would not mean letting AI make decisions for lawmakers. It would mean giving lawmakers a better understanding of what they are working with.
This is also why the debate about AI-generated legislation should not just be about whether AI writes good or bad language.
People have always been able to write bad laws. Congress can create unnecessary rules. Agencies can create confusing regulations. Lawyers and other experts can produce huge amounts of complicated language.
The difference with AI is how quickly it can create more of it.
If the problem is that government does not have enough information, that could be helpful. But if the problem is that government already has too much information and has trouble understanding it, creating more information may not solve anything.
It could make things worse.
AI could instead be used as a tool for remembering and organizing what the government already knows. There are decades of laws, regulations, policies, court decisions, contracts and other documents spread across the government. AI could help connect all of those pieces.
For example, it could tell a policymaker that there are already programs trying to solve the same problem. It could point out that a new proposal might conflict with an existing rule. It could explain how a law has changed over time. It could show that two agencies have similar or overlapping responsibilities.
Most importantly, it could help answer a basic question: What already exists?
This is something government contractors understand well. Before responding to a government requirement, they have to figure out which rules actually apply and how those rules connect to each other. Finding a document is not enough. You have to understand what it means and how it fits with everything else.
The government should take the same approach with AI.
The biggest opportunity may not be using AI to write legislation faster. It may be using AI to help government understand the huge amount of law, regulation, policy and guidance that already exists.
Where do the rules connect? Where do they conflict? What is outdated? What has already been tried? And does the government really need another rule?
AI should not decide what government should do. People still need to make those decisions and take responsibility for them. But people can make better decisions when they understand the system they are changing.
There is also a potential benefit beyond legislation. The same technology could help agencies and contractors understand complicated requirements before money is spent, contracts are written or programs are created. Instead of spending so much time searching through documents and trying to figure out which version of a requirement is current, people could use AI to help organize that information and explain how the pieces fit together.
That could make government more efficient without simply creating more government paperwork.
It could also help reduce one of the biggest risks of using AI in government. If AI is mainly used as a writing machine, the natural result will be more documents, more rules and more language. If it is used as a tool for understanding, it could help people question whether another document or rule is even necessary.
That is an important difference.
The goal should not be to use the newest technology to produce government faster. The goal should be to use it to make the government easier to understand.
So maybe the basic rule for using AI in government should be simple: understand before you generate.
Before writing a new regulation, understand the regulations that already exist. Before creating a new program, understand the programs that are already there. Before adding another requirement to a contract, understand the requirements that already apply.
And before asking AI, “What should we write?” maybe the first question should be:
“What do we already have?”
The first approach gives government more text.
The second could give the government something much more important: understanding.