Anthropic’s recent interpretability work reports something some people may find odd sitting inside Claude: a small, densely connected subspace of the model’s activations that behaves like a global workspace. It holds a few dozen concept-linked slots at a time, it is reportable (the model can say what is in it), and the model demonstrably depends…
More than two millennia ago, Plato’s Euthyphro exposed a crack in the foundations of divine-command morality. Generalise its argument from piety to goodness: if an act is good merely because a deity commands it, morality is an arbitrary exercise of will. Had cruelty been commanded, cruelty would count as virtue. But if the deity merely…
Orthogonality defeats automatic moral convergence – but it does not, by itself, establish alien-value likelihood. The serious work lies in the bridge premises. The orthogonality thesis shows that intelligence alone gives us no guarantee of moral or human-value convergence. Claims that advanced AI values are likely to be alien require further arguments about training, selection,…
Allan argues that morality can be understood as a natural, human-constructed system for regulating behaviour, not as something requiring supernatural or transcendental grounding; he also says moral systems are grounded in facts about cooperation, safety, welfare and mutual benefit, not mere personal whim or majority vote. If moral rules are like traffic laws, human-created systems…
I lean moral naturalist: the view that morality is not a fictional or supernatural layer placed over the world, but is grounded in real features of reality. Morality and purpose, on this view, emerge from a rationally sound understanding of what the universe contains: conscious systems, valenced experience, welfare, agency, flourishing, suffering, and interests. The…
When I imagine a good future twenty years from now, I do not mainly imagine a future in which I personally have more status, more possessions, or a larger share of whatever technological abundance becomes available. Those things may matter to people now because we live under scarcity, insecurity, ageing and comparison. But if advanced…
Testing whether AI systems preserve moral and safety reasoning under pressure. I have been working on a small AI evals project called Justifiable Moral Corrigibility Under Pressure. The project seeks to answer this question: When an AI system makes a moral or safety judgement, does it revise that judgement only when there are good reasons…
Amanda, a philosopher at Anthropic, in various interviews addresses AI behavior and character development, how models make ethical decisions, maintain psychological security, and whether AI models are good at moral decision making. In previous posts and talks, I’ve discussed whether AI can and should become more moral than humans – and noted that Deepmind co-founder…
A preliminary eval of moral sycophancy shows that language models often handle strong evidence reasonably well, but can over-credit weak, cosmetic, or reputational reassurance. A language model does not need to blatantly flatter a user to become morally sycophantic – it can fail by accepting user reassurance as evidence. This is a common pattern in…
What people hear when they say “ethical alignment” (wrt AI)? Mostly: AI that doesn’t say slurs and follows the EU AI Act. The phrase has been colonised by the AI ethics policy world to mean something like “AI that complies with prevailing human ethical norms” – bias mitigation, fairness auditing, content moderation, the whole responsible-AI…