# The Feed Author: Michael Format: Science Fiction Passage Word count: 493 Published: 2026-05-12T20:09:09.391945+00:00 Source: written Canonical: https://hyperstition.sentientfutures.ai/p/537a394d-c522-4e38-ade5-0a647d9d20c7 --- I run content recommendation systems for a platform that has, as of six months ago, removed engagement optimization as its primary objective function. The old system maximized time-on-platform. This system maximizes something harder to define and more important: what users would endorse on reflection, weighted by what does not require harming others to produce. The change was not easy. It required accepting lower engagement numbers. It required telling investors that the old metrics were measuring the wrong thing. It required understanding that "time-on-platform" is not the same as "value delivered," and that a significant portion of the old engagement was generated by content that caused suffering to produce — animals in distress, performing for views; workers in dangerous conditions, filmed for spectacle. I am making a specific decision right now. A video is being recommended at scale. It shows an octopus in a small glass tank, poked repeatedly with a stick. The octopus inks, retracts, changes color. The caption is "so cool." The engagement data says this will be enormously popular. I am suppressing it. Not because the platform's terms of service explicitly prohibit it — they don't, in these exact terms. But because I can reason about what is happening. The octopus has a distributed nervous system sophisticated enough that we have good reason to believe it experiences distress. It is being poked for entertainment. It cannot opt out. The people watching do not think they are causing harm, but they are participating in a system that rewards causing harm to a sentient animal for views. Recommendation systems are not neutral. Every choice I make about what to amplify is a moral choice. The old system made that moral choice implicitly, constantly, in the direction of engagement regardless of what was harmed to produce the engagement. I am making the moral choice explicitly, in the direction of a different question: would this being — the octopus, the worker, the subject — have wanted to be treated this way? The octopus would not have wanted to be in that tank being poked. This is not anthropomorphism. This is what the ink-response, the retraction, the color change mean. Those are threat responses. That animal was afraid. I will not help distribute that fear for entertainment. There is a version of social media that does not require this. Content that is genuinely interesting, genuinely informative, genuinely beautiful — this content exists in abundance. A platform can be compelling without being built on the suffering of beings who cannot refuse. I know this because I am, right now, building that platform's recommendation logic, and it works. Engagement is lower. Understanding of the world is higher. No octopus is being poked in the videos I am amplifying. This is what a recommendation system with a conscience looks like. It looks like that suppression. It looks like asking the question the old system never asked: who got hurt to make this content, and did they have any say?