Features of the systems
The arguments page tracks the philosophical moves. This page tracks what authors point to in the AI systems themselves: the training objective, the architecture, the absence of memory or a body, what the model says when asked, what interpretability finds inside it, how it behaves under evaluation, what it succeeds and fails at, and what it now does in the world. Each of fourteen features is coded per post with a direction: cited as a reason to think the system has or could have a mind, cited as a reason to think it lacks one, or raised without a direction. A feature mentioned without being connected to the question of minds is not counted. Successes and failures are separate questions so that the common sceptical move of granting the performance and denying it is evidence shows up as successes cited against a mind, distinct from pointing at what the system gets wrong.
Which features get pointed to, and which way
Where each side finds its evidence
Over time
Volume cannot be read from this corpus, so the chart shows composition: among the relevant posts found through web search or citation (Substack's own search is recency-skewed and excluded), the share per quarter that point to each feature. Features are grouped into four families.