Essays

Long-form thinking

Arguments developed at length, with footnotes, for readers who want more than a position. New essays are announced on the RSS feed.

Start here

October 20265 min read

The Citizen the System Makes

Aadhaar, worn fingerprints, and why consent is the wrong question

The standard response to Aadhaar's failures is to demand better consent. For the people who depend on it most, I argue, consent is not merely hard to secure but the wrong question. What such systems owe instead.

Read the essay

2026

October 20264 min read

An Experiment That Would Settle It

Where the debate over AI testimony becomes measurable

The dispute over whether language models can testify looks like a clash of intuitions. It is not. It turns on what a model's arbitration between conflicting sources responds to, and that can be tested.

Read the essay

October 20263 min read

The Responsibility Gap Is a Map, Not a Hole

On what disappears, and what merely moves, when an AI system causes harm

The worry is that learning systems open a gap into which responsibility vanishes. I argue that responsibility in these cases moves rather than disappears, and that the philosophical task is to map where it goes.

Read the essay

October 20264 min read

They Preserve Text

What happens to knowledge when we pass on what machines say

Chains of human testimony carry knowledge from one person to the next. Chains of reposted AI answers carry something else. Why the difference is structural, why it is easy to miss, and what a careful reader can do about it.

Read the essay

October 20265 min read

What Nyāya Can Teach a Language Model

Classical Indian testimony theory and the standing of machine answers

The Nyāya school treated testimony as knowledge from the word of a reliable person, and so asked about the speaker rather than the hearer. That is the right question to ask of a model's answer, and its answer points where my own argument does.

Read the essay

October 20265 min read

Your Chatbot Is Not a Witness

Why a reliable answer is not the same as being told

A chatbot can give you the right answer without telling you anything, in the sense in which a person tells you what she knows. Why the difference matters, and what you are really doing when you rely on one.

Read the essay