AI Tests Lawyers’ Will to Keep Training the Models in Our Minds

David Wang, Cooley’s chief innovation officer, authored the below article for Bloomberg Law’s Professional Perspectives.

The Bottom Line

• AI has made producing legal work almost frictionless, but it remains up to lawyers to understand and stand behind that work.
• Lawyers and other professionals must choose every day to train the “models” in their heads and resist cognitively surrendering to AI.
• Lawyers don’t have to accept that the good and the bad of AI is a packaged deal — they can benefit from its strengths while still developing the knowledge to explain its output and identify its weaknesses.

Whether it is a resume, an article, or a contract, we have all encountered the phenomenon known as “AI slop.”

It has a recognizable texture: grammatically perfect, structurally turgid, and intellectually orphaned. The person who produced it may not understand or agree with every sentence. Sometimes, they haven’t even read it.

Consider the associate who circulates a clean diligence summary that mischaracterizes a change-of-control clause. Or the in-house lawyer who forwards an AI-generated regulatory overview that omits the one jurisdiction that matters. The finished product creates negative value. This is the worst-case scenario in the era of #sloplaw.

This is a problem with the AI model, right?

Actually, the most important model is the one in our heads. We have never had to pay attention to the model running in our human brains because there was no alternative before. AI offers us a machine that can help us learn, reason, and create at extraordinary speed. Like all powerful things, it offers tremendous temptation: In this case, the appearance of understanding without the work of understanding.

People have largely treated AI’s duality — the good and bad as a packaged deal. I reject that premise completely.

We can both capture the benefits and avoid the pitfalls of AI, but it will be a test of human will.

Production Isn’t Understanding

AI has collapsed the friction involved in producing words. It can draft the email, summarize the cases, and generate the contract without the need for a person to decide what any of it should say.

That is a genuine technological miracle. We have made rocks talk.

In a world where knowledge work is expressed in words, one could be fooled into thinking that the rocks are also doing the work. But it is called “knowledge work,” not “talking work,” for good reason.

As a foolish young man, I once thought that being a lawyer meant yelling “objection!” in a slick suit. As a slightly less foolish but far more arrogant law student, I thought I understood that it was knowing when one can yell “objection!” that was the key. It is only belatedly dawning on me now, in this AI moment, that it’s the act of standing behind one’s objection, in the representation of a real client, that distinguishes a lawyer from a perfect objection-yelling machine.

One can’t stand behind a thing one doesn’t understand. For most of history, production and understanding were intertwined. Drafting a complex document took long enough that a lawyer had to engage with its substance. Writing and revising created natural speed bumps: questions arose, assumptions were tested, and mistakes became visible. “Good writing is clear thinking,” the mantra goes.

Generative AI can now produce legal material without human understanding. For tasks that are ministerial, AI will be a boon to productivity. As long as it passes a spot check, the content doesn’t really matter. But for the work that we do, product without understanding isn’t productivity. Standing behind the thing is the actual point.

Reverse Distillation

I love telling people I’m training my own model. Living in Silicon Valley doing what I do, sometimes I get to push this joke pretty far: “500 trillion parameter model, totally proprietary.” The smartest victims get the joke first. “I’m actually distilling a mini version of this model” — he’s starting first grade in the fall.

Cognitive surrender occurs when a person uses AI not to extend thought, but to replace it. The tool produces a plausible answer, and the human accepts it without making the underlying knowledge their own. AI is efficient, comfortable, and corrosive. The intellectual equivalent of potato chips.

The alternative is the intellectual equivalent of getting up at 6 a.m. each morning to exercise: Train your own squishy, meat model using AI. I call it reverse distillation.

Consider a junior lawyer who receives an assignment they don’t fully understand. The easy path is to ask AI for the answer and forward it back, often resulting in complaints from supervising lawyers. “This is AI slop.”

In this scenario, the junior lawyer can’t answer basic questions about the assignment: What doctrines are embedded here? What assumptions are being made? What questions are implied? AI can surface all of this and teach it to the junior lawyer, but only if the lawyer asks it to. This isn’t “prompt engineering.” This is professional identity formation.

When the frontier foundation model providers complain about open-source model providers “distilling” their models as a competitive tactic, what they are complaining about is a model training technique that involves asking questions of the source “teacher” model to efficiently bring a “student” model up to the same intelligence. Funny enough, that’s exactly how humans should be using AI models.

The distinction is simple: Are you using AI to produce an answer, or to become someone capable of producing the answer? The former creates output. The latter creates capability.

Once this capability is in place, then one can fully leverage the accelerated production as well. To truly complete the assignment isn’t technically challenging, and it can’t be automated by AI. What the assignment requires is a professional sense of duty to understand and stand behind the answer, and the will to honor this duty always.

Never Surrender

Being a lawyer isn’t a credential you earn once. It is a choice you make every day.

In every case, on every assignment, you choose whether to understand or merely transmit; whether to exercise judgment or borrow the appearance of it, whether to take responsibility for the answer or surrender it to the machine.
The default choice will be to do the easy thing. The temptation will always be there, calling to you like the Ring of Power. Only you can choose to be the lawyer.

Legal education is beginning to grapple with this reality. The University of Chicago Law School’s recent AI strategy — combining device-free foundational courses with integrated AI instruction — is directionally right. Students must learn that they can think without assistance. But they must also learn to think in its presence.

Law firms face the same imperative. The traditional apprenticeship model relied on the fact that producing legal work took time. Learning happened along the way. AI breaks that linkage, but it also creates new opportunities, because if production takes less time, that time can be reinvested in understanding.

Training must become explicit rather than incidental. Firms can require explanation, not just output, and measure comprehension, not merely completion. They can also identify cognitive surrender early: work that looks polished but collapses under scrutiny, conclusions without reasoning, and answers without ownership.

The big law model has always been a crucible of sorts. Senior law firm leaders will recognize the quality one needs to succeed as a lawyer in the AI era: the commitment to be the best lawyer you can be, made new every day.

The Great Bifurcation

This phenomenon will define this coming era of legal work. On one side will be those who use AI to train their own models in their minds — becoming faster, sharper, and more capable with every interaction. On the other will be those who use it to replace thinking — producing more while understanding less.

At first, the difference will be subtle. Then it will be decisive.

Avoiding this outcome requires a conscious rejection of cognitive surrender and a deliberate commitment to continuously train and update your own model, with the level of intensity AI companies devote to theirs.

You can use AI and stand behind every word, like I did to write this article. When one has mastered one’s own model, AI use becomes a true tool and extension of capability. This commitment, to choose to use AI to produce and understand, will define careers.

The author used AI to draft a portion of this article.

Reproduced with permission. Copyright August 10, 2026 by Bloomberg Industry Group, Inc. (800-372-1033) https://www.bloombergindustry.com/