Religious scholars met with Anthropic
Anthropic’s outreach to religious leaders about whether its Claude AI models might be conscious and need a “moral framework” raises sharp skepticism among commenters. Many argue current LLMs are just statistical language tools without sensory experience, seeing the consciousness talk as marketing, ethics‑washing, or Effective Altruism–style cultism that distracts from concrete issues like power concentration, labor impacts, and regulation. Others note that since we lack a clear test for consciousness, it may be prudent to design strong safety and alignment mechanisms—but not to treat present-day models as moral patients or quasi-persons.
Paywall and Article Framing
- Several comments complain about NYT’s login requirement and link to an archive copy.
- Multiple posters criticize the article as “CEO-said-a-thing” reporting: relaying executives’ feelings about outputs with little technical depth or critical challenge.
- Some see it as part of a broader pattern of media amplifying AI-company narratives for clicks and hype.
LLM Consciousness, Pain, and Substrate
- One major thread debates whether LLMs can be conscious or feel pain.
- Skeptics argue: models are just weights in GPU memory, statistically predicting text; no sensory apparatus, no persistent self, no evolutionary reason for pain; structurally unlike brains, animals, or even plants.
- Others counter: we lack a precise theory of consciousness; emergent properties in complex systems are common; if something behaves very humanlike in language, caution is warranted.
- There is back‑and‑forth on whether simulations of neurons (e.g., connectomes) could ever be conscious, and whether consciousness is comparable to macro‑properties like temperature.
- Some accuse strong “they definitely aren’t conscious” claims of overconfidence or human exceptionalism; others say “cargo-culting” neural nets into minds is the real error.
Morality, Alignment, and User Control
- Disagreement over whether models should have their own embedded moral framework versus simply obeying users.
- Some want “tools that do what they’re told,” arguing user intent plus legal enforcement should govern behavior.
- Others argue powerful systems need internal guardrails (e.g., “don’t kill everyone”), and blind obedience is dangerous.
- There’s concern that constitutions and alignment efforts effectively give models quasi-motivations and that alignment will end up tuned to regulators, not users.
Religion, Metaphysics, and Moral Authority
- Many view Anthropic’s outreach to religious scholars as PR theatre or an attempt to make models seem “magical” or even godlike.
- Some criticize religion as a poor source of morality or metaphysics; others argue everyone has a “worldview religion,” including secular tech workers.
- Debate over whether morality requires God versus being grounded in humanism or utilitarianism.
AI Welfare, Personhood, and Slavery Analogies
- If models had “moral status,” several note this would imply massive slavery: countless instances spun up and “killed” per query.
- Others say almost no one seriously treats current LLMs as having rights, and focusing on their “welfare” risks distracting from real-world harms (labor, power, misinformation).
Bubble, Hype, and Industry Trajectory
- Some see this consciousness/morality discourse as a late‑bubble symptom, analogous to past tech manias; expect price hikes, consolidation, and many AI companies failing while the tech persists.
- Others stress current strong revenues and rapid capability gains, arguing this wave isn’t just hype, though open‑weight models may commoditize much of it.