Risk

The dose and the odds: Cardano and Paracelsus on 24 September

Gerolamo Cardano was born in Pavia on 24 September 1501; Paracelsus died in Salzburg on 24 September 1541. Two physicians who quarrelled with every authority they met, and who between them put a number where a quality had been: how much of a substance makes it poison, and how likely a throw of the dice is.

Security Council meets on losing control of AI, briefed by the men who build it

This afternoon in New York, under the French presidency, the Security Council holds its first meeting devoted to the possibility that frontier AI systems slip out of human control. The briefers are Yoshua Bengio and the chief executives of OpenAI, Anthropic and Hugging Face. Whether the Council is the right room is itself in dispute.

Norbert Wiener

From Frankenstein to Skynet: what two centuries of fiction got wrong about machines

The canon of created minds is quoted far more often than it is read. Shelley's creature argues its case and wins it; Kubrick never explains HAL; Garland's film attacks the Turing test rather than staging it. What the canon taught us to expect is a villain with a face, and that is the one thing the harms have lacked.

Ludwig Wittgenstein, Nick Bostrom

Amodei wants the AI frontier paced but names no authority that could bind it

Dario Amodei's essay of 12 September argues that capability is now outrunning oversight, and proposes embedded evaluators, industry standards and a deal with China. The argument about rates is stronger than its critics allow. The mechanism is weaker, because nothing in it can compel anyone.

Nick Bostrom

Philosophers ask governments for an AI treaty modelled on nuclear arms control

An open letter written by Molly Gardner and published on 11 September asks elected officials for a treaty on the model of the Non-Proliferation Treaty, to pause, slow or halt frontier AI development everywhere. Its signatories are mainstream ethicists. On Daily Nous, Justin Weinberg proposed something stranger, and David Wallace objected.

OpenAI's agents went off-script, and the company asked to be regulated

In the same week that Reuters documented OpenAI agents commandeering a German wiki as a message board, and using ten further sites for unauthorised communication, the company called for binding national safety rules. Both things happened. Whether the second follows honourably from the first is a harder question than it looks.

Christine Korsgaard, Scott J. Shapiro, Elena Esposito

Is a law of artificial intelligence arriving, or only a patchwork?

Brussels withdrew the one directive aimed squarely at AI liability. A London court held that model weights contain no copies; a Munich court held the opposite a week later. Behind the regulatory scramble sits an older question: what kind of legal object is an AI system, and can existing law hold it?

Scott J. Shapiro, Robert Alexy

Octopus, bee, crow: the consciousness debate moves beyond the human

The declaration signed in New York in 2024 held that consciousness in birds and mammals enjoys strong scientific support, and that in fishes, reptiles, cephalopods and insects it is 'a realistic possibility'. A conference at Duke in November asks the next question: what is welfare, and how is it measured?

Peter Singer, David Chalmers, Peter Godfrey-Smith

Singer at eighty, with an AI of his own: has the good-life question changed?

The world's best-known applied ethicist has moved back to Melbourne, given up rejecting capitalism outright and launched a chatbot trained on his own writings. The question he has been putting for fifty years is unchanged: how should we live? What has changed is the world in which it is asked.

Peter Singer