That last group is one of my pet peeves, especially since GenAI fits in that group.

We have always had people explaining human experience they have never had. Colleges do not have a monopoly on this, but they are a good place to start. Some instructors have done the work they teach. Some have studied it deeply without doing it. The credible ones tell students the difference. They can explain the research, the law, the theory, and the limits of their own experience without implying that reading about an event is the same as being there.

Academic research matters as a researcher may know the published law better than the police officer applying it. A professor may know the history of a profession better than the person working in it. Teaching from books is not dishonest. Teaching from books while implying that you possess experience you do not have is. The problem begins when credentials are used to turn secondhand knowledge into firsthand authority.

Some of them dismiss the people who actually did the work as “anecdotal.” Apparently, experience is unreliable when it comes from a practitioner but becomes knowledge when an academic interviews the practitioner and publishes it.

I once read a doctoral dissertation about DFIR whose central findings rested on interviews with nine security investigators.

Nine.

Only three were forensic examiners. The others were incident responders and triage analysts, all grouped under the broader label of “digital forensic analysts.” The study reached conclusions about skills common across the profession.

The author had substantial experience in network defense, security operations, and incident response. What the dissertation did not clearly establish was how much conventional digital forensic (i.e., legal casework) examination the author and participants had actually performed, distinct from incident response and network-security investigations.

Interview nine practitioners’ experience, filter their experience through an academic framework, call it saturation, and that same anecdotal experience returns wearing a doctoral robe. That is not the same body of work as producing forensic examinations that have to survive cross-examination, chain-of-custody challenges, and a judge who has never heard of a $MFT.

That should aggravate anyone who has ever done difficult work and then listened to someone who has never done it tell you what your experience is really like.

All I Needed to Know About Police Work I Learned in a Ride-Along

When I was both a police officer and a student, I enrolled in a criminal procedure class at the University of Washington. I dropped out of that class.

As I remember him describing it, the professor’s law enforcement experience came from thesis research involving the Oakland Police Department in the 1970s. He went on ride-alongs as a student-researcher before eventually teaching law enforcement procedure. That was his description of his practical law enforcement experience, but he taught it as if he lived it during that semester of ride-alongs.

That was the experience behind what he taught about police work. His explanations were far enough from the work I was doing that I could not sit through the course after several arguments. Even if every observation he made in Oakland had been accurate, he was teaching in a different state, decades later, under different laws, policies, court decisions, technology, training, and public expectations.

It is one thing to know the law governing deadly force. It is another to be entrusted with a firearm and required to make a deadly-force decision. The researcher may know every published court decision. The police officer may know what happens when those decisions collide with a dark street, incomplete information, an armed person, a frightened witness, a supervisor, a prosecutor, a defense attorney, and eventually a jury.

Experience does not make someone automatically right. Practitioners can be biased, lazy, careless, jaded, or stuck in the past. Thirty years on the job can mean thirty years of learning. It can also mean the first year repeated thirty times.

But lived experience still matters. Reading about something, watching someone do it, and summarizing what other people said about it are not the same as living it. I can tell you that reading about someone getting shot and actually getting shot at are two different things.

GenAI Puts the Problem on a Conveyor Belt

You no longer need to pay tuition to be taught about an experience by something that has never experienced anything. Ask a chatbot and it will assemble a plausible answer from patterns in its training data, your prompt, and whatever sources it can access. It can merge, summarize, and imitate understanding without understanding any of it. We are beginning to treat the appearance of understanding as understanding.

My Corner of Kindle Unlimited Was a Slop Factory

Amazon’s Kindle Unlimited is a great idea. Pay a monthly fee and read as many books as you want. I subscribed because I thought it would be useful for research.

I found books on digital forensics, investigations, technology, and law with strong titles, professional covers, and descriptions containing every keyword a reader might search.

Then I read them.

Many appeared to have been written largely, if not entirely, with GenAI and barely touched by a knowledgeable human. Some contained technical information that was plainly wrong. In several cases, I could not verify the author’s identity, qualifications, professional history, or any meaningful connection to the subject.

That does not prove the authors were fictitious. It does not prove AI wrote the books. It does mean I had no reason to trust those books as technical sources. If I cannot determine who the author is, what the author knows, how the author knows it, or where the claims originated, I am looking at packaging instead of authority.

Citation Laundering

I use “citation laundering” to describe the process by which an unsupported claim acquires the appearance of authority through repetition.

An AI-generated book cites a paper that does not exist. A blogger repeats the claim without checking the paper. A trainer cites the blog. A professor puts the claim into a presentation. A government report cites the professor. Then everyone cites the government report.

By that point, the claim looks legit. It is still the same bad claim, but now it carries a thicker résumé backed by acronyms of credentials and by being cited by numerous “sources.”

Consider the six-foot COVID distancing rule. In congressional testimony in 2024, Anthony Fauci said he did not recall how the exact distance was selected, was not aware of studies establishing six feet as the threshold, and said the rule “sort of just appeared.” He described it as an empirical decision rather than one based on specific data. This was not said at the time. Six feet was not a scientific force field. It was an operating rule marketed as law of nature.

That does not mean distance was irrelevant. The risk of infection generally increased with proximity, and public-health officials had reasons to recommend physical separation. But six feet was presented to the public as something close to a scientific boundary when it was closer to a practical rule of thumb.

Because these were delivered as scientific proof and packaged with institutional authority, we treated it like fact. And when people didn’t bow to the COVID mitigation rules, the consequences were physical. Security guard Calvin Munerlyn was shot in the back of the head and killed in Flint, Michigan, trying to enforce an unverified mandate. Teenaged McDonald’s employees in Oklahoma City took shrapnel and gunfire because a lobby was closed under the same absolute certainty. We turned low-wage retail workers into frontline infantry for an empirical guess, all because nobody asked to see the source code.

Now add GenAI.

GenAI can create claims, explanations, summaries, and citations faster than anyone can verify them. Those claims enter books, articles, courses, legal filings, technical reports, and government publications. People cite them. Future AI systems may ingest them and repeat them.

Every repetition moves the claim farther from its origin while making it look more established.

“Polished” once suggested that someone had spent time researching, checking, writing, editing, and building something. Now it may only mean that someone spent five minutes writing a prompt.

The Consequences Still Land on Humans

This is not merely an argument about bad books and junk science.

In a 2025 federal case involving the use of force by immigration agents in Chicago, U.S. District Judge Sara Ellis noted that, in at least one instance, an agent asked ChatGPT to compile a report narrative from a brief sentence and several images. After comparing official accounts with body-worn camera footage, she wrote that using ChatGPT further undermined the agents’ credibility and might explain inaccuracies in the reports.

A police report can influence an arrest, a charging decision, a plea, or a sentence. An inaccurate HR report can cost someone a job. A flawed forensic conclusion can change the direction of a case. A false statement in an affidavit can put the power of the government behind something that never happened.

The consequences land on a human being. Nothing happens to the AI.

When lawyers are sanctioned for submitting fake cases or fabricated citations, I am happy to see it. Consequences are sometimes the only training that works. But the errors must first be found, opposing counsel must recognize the problem, have the resources to investigate it, and then be willing to fight.

Many cases settle and many reports go unchallenged. Some people may never learn that an official document used against them was partly, and perhaps inaccurately, generated by AI.

Government-Grade Slop Is Still Slop

Government use of GenAI can affect thousands or millions of people at once.

In 2025, the White House’s Make America Healthy Again Commission released a report concerning childhood chronic disease. Reviewers found broken links, mischaracterized research, repeated citations, and references to studies that apparently did not exist. Some citation links reportedly contained “oaicite,” a marker associated with output from OpenAI tools. The report was later updated, and government officials characterized the problems as minor citation and formatting errors.

The presence of those markers did not prove that AI wrote the entire report. But nonexistent and mischaracterized sources proved that someone did not adequately check it before the federal government published it.

That is the part that matters.

Saying that a report was created with “AI assistance” tells me almost nothing. Which system was used? What did it generate? Which sources did it access? What did a qualified human verify? Who takes responsibility when something is wrong?

Without those answers, disclosure becomes another layer of “polish.” Maybe instead of “polish” we should call it “lipstick” (on a pig). Government reports influence laws, budgets, public health, criminal justice, and wars. “We used AI” is not an excuse and “the AI made a mistake” is not accountability.

The Human Version of Model Collapse

Researchers use the term “model collapse” for degradation that can occur when generative models are recursively trained on model-generated data. Errors and distortions can compound as the systems move farther from original human-created data.

I think we are creating a human version of it.

AI summarizes human experience. People rely on the summary instead of talking to the humans who had the experience. Other people quote and cite the summary. AI later consumes those new sources and produces another version.

Eventually, no one knows who performed the work, tested the method, witnessed the event, verified the claim, or accepted responsibility for the conclusion. The content looks better even though its connection to reality gets worse.

I use GenAI for research, editing, brainstorming, and testing ideas. But I do not get to blame AI when something carrying my name is wrong. The responsibility is mine.

Before relying on a book, report, course, website, expert opinion, or being taught by someone, I want to know:

• Who created it?
• What relevant work has that person actually done?
• Where did the claim originate?
• What did AI generate?
• What did a qualified human verify?
• Who will answer for it when it is wrong?

Lived experience is not perfect. It does not replace research, nor does it excuse bias, poor judgment, or outdated practices. Academic research and lived experience should challenge and test each other. One should never impersonate the other.

DFIR Should Know Better

The DFIR community should be among the first to question what we are told. We trace artifacts to their sources, test competing explanations, and change our conclusions when the evidence changes. We demand provenance from evidence. We should demand it from information and from ourselves. Changing your mind when the evidence changes is the job.

Yet we accept headlines, research, government reports, and AI-generated answers when they confirm what we already believe, or more accurately, what we are told to believe and what we want to believe to be true as long as it aligns with our biases.

We humans are already predisposed to believe what confirms what we want to believe. If the news tells us a politician is good or evil, many of us accept it. If a scientist tells us something is safe or dangerous, we accept the authority along with the answer. If someone gives us a version of history that fits our worldview, we eat it up.

We should be asking: What is your source? How does that source know? What was left out? What is your intention in the way you are telling me? And who benefits if I believe you?

Now AI can manufacture the appearance of authority without possessing experience, judgment, intention, or accountability. It can tell a convincing story about a place it has never been and an experience it has never had, and can be abused in a way that we may never see coming.