Beyond Hallucination: Semantic Laundering, Semantic Stewardship, and Prompting Aikido in Human-AI Collaboration
Abstract:
The primary conversation surrounding generative artificial intelligence centers on factual fabrication: hallucination. While hallucination matters, it represents a crude and easily detected failure mode. This paper identifies a second, quieter epistemic risk: semantic laundering. Large language models produce exceptionally fluent, coherent, and
It started with a routine task, as it usually does. I was using Claude to review a draft critique of an article I had some doubts about, checking the research behind a few of its assertions and testing its logic.
During the exchange, I noticed something simple: Claude was applying a strict standard of proof
People seem to like my AI session transcripts where I see big AI failures. I definitely don’t like living through them, but they can be instructive to readers of this blog.
Here’s an excerpt that will give you much to think about. Note my prompting style (my two prompts are in bold) as we reached
In the latest episode of The Kennedy-Mighell Report, Tom Mighell and I turned the spotlight inward to talk about the course I just finished teaching at the University of Michigan Law School called Legal Technology Literacy and Leadership. This course is considered a practical simulation class. It was built this past semester around
I’ve been tracking a subtle but dangerous shift in the legal tech landscape. For the last few years, the entire industry has been obsessed with “hallucinations”—obvious, glaring errors like fake case citations. But as Tom Mighell and I discussed on episode 418 of The Kennedy-Mighell Report, focusing
The Door in the Hallway: Agentic AI and the Illusion of Control
We used to talk a lot about the concept of a customized, bounded news feed designed to give professionals exactly what they need to know, and nothing they don’t. It was usually called the “Daily Me.”
Lately, I’ve been trying to build a
The prevailing narrative I hear in the legal world is that Claude is the “most human” of the LLMs and, especially, a nuanced, sophisticated writer. When I report that the system has begun to fail my specific research protocols, the common response is a suggestion that I am simply using the wrong version and a
Coherence degrades while fluency improves.
The central problem is not that AI systems sometimes fail. Of course they fail. Nor is the main problem that they occasionally hallucinate, wander, or produce obvious nonsense. Those are manageable problems because they announce themselves early. The more interesting and professionally dangerous problem is that a system can become
At a certain point in a long AI session, I can feel the texture change.
The words are still smooth. The tone is still confident. But something underneath has started to slide and give way. The session is still moving forward, yet the logic is no longer holding together in the same way.
That happened
A lot of the most interesting work in legal tech seemed to be coming from people who were newer to the field, earlier in their careers, or just not as widely known yet as
