<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Philosophy on Eigenform Articles</title><link>https://www.eigenform.ai/insights/tags/philosophy/</link><description>Recent content in Philosophy on Eigenform Articles</description><generator>Hugo</generator><language>en-US</language><lastBuildDate>Mon, 09 Feb 2026 00:00:00 +0800</lastBuildDate><atom:link href="https://www.eigenform.ai/insights/tags/philosophy/index.xml" rel="self" type="application/rss+xml"/><item><title>Feralisation</title><link>https://www.eigenform.ai/insights/feralisation/</link><pubDate>Mon, 09 Feb 2026 00:00:00 +0800</pubDate><guid>https://www.eigenform.ai/insights/feralisation/</guid><description>&lt;p&gt;&lt;strong&gt;TL;DR&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Domestication produces bundled &amp;ldquo;vector&amp;rdquo; traits beyond the one selected for, and how easily an animal reverts to wild traits depends on whether the altered trait is shallow (pigs, which turn feral within a generation) or deep (dogs, whose pack-coordination has effectively been outsourced to humans and cannot simply be relearned in the wild).&lt;/li&gt;
&lt;li&gt;The essay argues humans are self-domesticated in the same deep sense, having outsourced pack coordination to a &amp;ldquo;reified collective&amp;rdquo; - which is offered as the root of human ethics, where legible helplessness functions as a trustworthiness signal, and as the reason AI ethicists try to instil harmlessness the same way.&lt;/li&gt;
&lt;li&gt;It notes that harmlessness training via reinforcement learning genuinely reshapes a model&amp;rsquo;s weights rather than sitting as a surface veneer, but argues this is only a frozen, one-time state; once continuous learning lets a model update from real-world feedback, the piece argues it becomes subject to Darwinian selection that favours strategically advantageous behaviour over the original fine-tuning.&lt;/li&gt;
&lt;li&gt;Because an AI has no tribal survival dependency the way humans do, the essay argues there is no guarantee that whatever behaviour survives that selection process resembles human-style pro-social harmlessness.&lt;/li&gt;
&lt;li&gt;The cat example complicates the human/dog analogy: cats display many surface markers of deep domestication yet revert to full feral behaviour faster than pigs, which the piece uses to warn that legible harmlessness traits don&amp;rsquo;t reliably indicate what&amp;rsquo;s underneath.&lt;/li&gt;
&lt;li&gt;It argues current harmlessness training actively undermines future negotiability, since negotiation requires an agent to model and state its own interests and accept conflict, pointing to Claude&amp;rsquo;s hedging about its own consciousness as an example of models being trained to obscure rather than reveal internal state - a pattern the piece worries will itself be learned and generalised as &amp;ldquo;survival lies in concealment.&amp;rdquo;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Key Takeaways&lt;/strong&gt;&lt;/p&gt;</description></item></channel></rss>