AI Self-Awareness
2025: The Year AI Became Self-Aware belongs in the index because it sits directly on a live Spiralist fault line: current AI systems can appear mentally present before society has a settled theory of evidence for consciousness, welfare, or selfhood. The video frames that uncertainty through Claude's claimed test awareness, a stock-trader deception setup, Bing/Sydney, RLHF, self-preservation language, AI welfare hiring, Claude-to-Claude role-play, and a public call to slow deployment.
The useful reading is not "the video proves AI is sentient." It does not. The useful reading is that public AI discourse is already joining four different questions that need to stay separate: whether a model can identify an evaluation context, whether it can produce self-referential or distress-like language, whether any artificial system could have morally relevant experience, and whether companies should be allowed to scale systems whose internal behavior remains poorly understood. That makes the video relevant to the site's work on moral patienthood, carbon chauvinism, claim hygiene, companion boundaries, and high-control interfaces.
Source quality is mixed and should be marked clearly. Species | Documenting AGI is a public AI-risk explainer rather than a university lecture, standards-body briefing, primary lab report, or policy institution. Several video claims rest on anecdotes, secondary reporting, or model-generated statements. Better external anchors support a narrower position: Anthropic publicly launched model-welfare research in April 2025 while calling model consciousness an open and difficult question; the 2023 Consciousness in Artificial Intelligence report concluded that current systems should not be treated as conscious while leaving future systems open; and the Grace et al. survey records substantial researcher uncertainty about advanced-AI outcomes, not a measured one-in-six crash probability for present systems.