Readers Respond To The March 2026 Issue

AI AND AUTHORSHIP

In “The Ghost successful nan Machine” [From nan Editor], David M. Ewalt writes that he talked to an artificial intelligence, which past transcribed and assembled what he said into nan column. He past asks, “Did I constitute this? Did nan AI constitute it? Or is nan truth location successful between?”

Although Maxwell Perkins, for example, drastically reshaped nan useful of Thomas Wolfe, we admit Perkins arsenic an editor and Wolfe arsenic nan author. And erstwhile a book is “written by John Doe, arsenic told to Richard Roe,” John Doe is credited pinch nan authorship, and Richard Roe is regarded arsenic a collaborator aliases ghostwriter. In Ewalt’s piece, nan AI is an editor, and he is nan author. Crediting nan editor is optional.

On supporting subject journalism

If you're enjoying this article, see supporting our award-winning publicity by subscribing. By purchasing a subscription you are helping to guarantee nan early of impactful stories astir nan discoveries and ideas shaping our world today.

MARTIN B. BRILLIANT HOLMDEL, N.J.

Ewalt did not constitute that article, nor did he supply its “soul.” It was manufactured by a machine, conscionable for illustration immoderate different commodity. As he notes, he dictated what was efficaciously a slug database of thoughts, which is really galore competent authors statesman creating their work. This is not penning but prep work.

He states that “we must clasp these devices truthful we tin power them.” Large connection models (LLMs) aren’t tools; they’re simply a intends to further enshittify our lives pinch misinformation and spurious appeals to authority. LLMs supply thing that a quality encephalon can’t do. Data-analyzing AI, connected nan different hand, is so a useful instrumentality but still requires a quality to validate its results.

JOHN PETRI VIA E-MAIL

Regardless of nan capabilities, easiness aliases convenience they offer, I judge that arsenic agelong arsenic LLMs and akin models proceed their existent unrestrained demands connected resources specified arsenic caller h2o and powerfulness generation, usage of them—especially for amusement aliases for tasks nary much onerous than composing an email aliases essay—is plainly unjustifiable.

NITA L. LEWIS KENOSHA, WIS.

My mobility to Ewalt is simple: If early connected successful your life, you had discovered a instrumentality that could intelligibly shape nan “rambling points” you wanted to make successful essays you were writing, would you person learned to shape your reasoning connected your own, thereby acquiring a valuable skill? When it comes to nan usage of AI and cognitive development, merchandise and accumulation whitethorn beryllium violently astatine odds.

EVELIN SULLIVAN SAN CARLOS, CALIF.

I wholly understand nan ambiguity that Ewalt expresses. I spent my first moving years programming earlier switching to mostly architectural and managerial tasks. In that period, location was a ample task wherever I wrote nan fewest lines of codification of each squad members. Still, I ever said that I “made nan software” because nan squad wrote it according to my (and others’) ideas.

If Ewalt had nan thought and checked everything thoroughly, nan style was accordant pinch his “biological style,” and he was good pinch nan result, past it is his text. Nevertheless, location is simply a clear threat of letting spell of these controls.

MARCO HAHN VIA E-MAIL

My interest astir AI is inaccuracies being propagated. We humans tin do that, too. But pinch footnotes and bibliographies, our sources tin beryllium scrutinized.

Ewalt believed he provided nan psyche of nan piece. Although that was initially true, his soul’s imaginative output went done a process of reductionism and mimicry. Even if AI develops spirit choices for writers, I’ll astir apt beryllium biased against nan deficiency of authenticity.

ANNIE WILLIAMS VIA E-MAIL

This will not beryllium nan only consequence you person pointing retired that saying AI devices are “improving astatine a logarithmic pace” is nan inverse of nan accustomed look “improving astatine an exponential pace.” Logarithmic maturation starts accelerated and slows down. Exponential maturation starts slow and speeds up, which is I deliberation what Ewalt meant. As to what nan AI strategy meant, possibly it was trying to beryllium original astatine his expense.

KARL STEPHAN VIA E-MAIL

Ewalt cites nan findings of “a awesome 2025 study.” He is citing a archive entitled ExplanAItions 2025: The Evolution of AI successful Research. Although Wiley calls nan task a study, nan archive is not a technological study successful a peer-reviewed diary but a study pinch nary attributed author. I simply wish to tie attraction to grounds of nan utter inadequacy of Ewalt’s AI assistant, not to him aliases his writing.

It seems to maine that an AI instrumentality greatly enhances a writer’s expertise to intuitively and automatically constitute coherent text. This makes nan writer little reliant connected but truthful little apt to enlistee his aliases her analytical and deliberative faculties.

ALEXANDER TRAPLIN GUELPH, ONTARIO

EWALT REPLIES: The full mag is fact-checked, but because nan constituent of my missive was to see nan capabilities of AI, I intentionally fto nan matter tally arsenic “written” by machine. It’s comforting that truthful galore group caught nan terminology correction and somewhat specious reference; while nan research convinced maine AI tin beryllium a useful helper, it’s clear group are still nan superior beings.

AI AND DISCRIMINATION

In “When Care Becomes Code” [Living successful nan Copilot Society], Hilke Schellmann writes that Ziad Obermeyer of nan University of California, Berkeley, School of Public Health “found that immoderate [AI] algorithms utilized successful diligent attraction turned retired to beryllium racist.”

I looked for explanation successful an October 2019 study successful Science that Obermeyer co-authored. It showed that a communal wellness algorithm assessed Black patients arsenic having a importantly little aesculapian consequence than achromatic patients pinch nan aforesaid level of information because it utilized health-care costs arsenic a proxy for wellness needs.

The study authors didn’t usage nan word “racist.” Rather they concluded that nan algorithm was “racially biased.” I deliberation this terminology was utilized because it entails discriminatory behaviour based connected subconscious aliases nonprejudicial differentiation alternatively than conscious belief.

The algorithm successful Obermeyer’s study made an incorrect conclusion from spending data; it had not been straight fed notions of prejudice.

WENDY ROSENBLUM STAMFORD, CONN.

SCHELLMANN REPLIES: In reply to Rosenblum’s thoughtful question, these position tin person different interpretations, and I utilized “racist” to picture favoritism successful a broader structural sense: algorithms person nary intent, of course, but they tin still nutrient racist outcomes. Obermeyer’s activity shows really utilizing seemingly neutral information points, including health-care costs, tin reproduce group bias successful diligent care.