Thomas Arnold

Thomas Arnold

@thomasarnold.bsky.social

HRI, AI, religious studies, pragmatism | ethics-sullying at Tufts Institute for AI | he/him | current pencil: Tombow 2558 H | end theory: low | realism: visceral thomasarnold.scholar.st

721 followers680 following489 posts

Pinned

This is snake-handling 101. The snake-handlers believe snakes can be deadly. But they also need you to be in awe of their faithful handling, to confer authority upon them as the special, spiritually gifted guarantors of Snake Safety/Governance. Fear, fascination, and protection reinforce each other.

Dave Karpf@davekarpf.bsky.social

Three notes: (1) this is not *just* marketing. These guys really do mean it. They are convinced they’ve built a vengeful digital god. (2) but also they’re a cult. They are not impartial scientific experts. (3) and one of the byproduct of their endless confidence is marketing hype.

It is something of a truism these days that the world is on fire. The necessary and sufficient conditions for ekpyrosis, however, have been somewhat muddled in the flaming shards of secondary literature still legible. The so-called "austere incineration" view *disappears into a crevasse*

Thomas Arnold reposted

A future where humans live alongside robots is closer than ever, according to computer scientists studying AI. And these Boston-area scientists say the technology that enables those machines doesn't require data centers.

A huge amount of pre-registration (BrIan Cantwell Smith) goes into assuming LLM's model human cognition, especially with moral reasoning.

Benjamin Paaßen@bpaassen.bsky.social

Over one year ago, we got a lot of response for our preprint "Large language models do not simulate human psychology".

Chaplin's "Modern Times," understanding in 1936 that workplace surveillance would not look kindly upon private bathroom breaks. What it underplayed was how vulnerable bosses would feel if they were exposed to worker perception via a big screen.

Among this paper's many contributions, for someone working in ethics, is an indirect illustration of the psychological error behind the "bitter lesson" quest for preference "alignment" through ever more compute and behavioral data.

Iris van Rooij 💭@irisvanrooij.bsky.social

"Capacities are not tasks" -- van Rooij & Guest (2026) journals.sagepub.com/doi/10.1177/...

Reading "Foucault in California" and still trying to accept that, after a lecture at Claremont in 1975, the night after he trips balls in Death Valley with host Simeon Wade and his partner, Foucault is met by...Al Franken, who asks him "Who do you think will win the World Series?"

If one substitutes "people whom they consider odd or exotic are in pain" for "something is made with AI" one faces how excruciating certain knowing would be (ghost workers and women in my field hurt, do I want to know more?), what theaters of intellectual sophistication might help me not have to do.

One of the effects of people caring about if something is made with AI is that they seem to have to confront: they don't know. Playing with definitions as a segue to argument from ignorance won't cut it. We have to do the heavy lifting of saying what it is we object to and why.

LinkedIn posts as grimoire

Pavel@spavel.bsky.social

Anyone doubting that people still believe in magic need only attend a single boardroom meeting and listen to extremely rich and powerful people chant ritual phrases like "delight the user" and "seamless" and "single pane of glass" in the hope that they will magically do something

Thomas Arnold reposted

🌿🍄🌿 Finally went into the office + picked up my copy of Thinking With Plants + Fungi (and some other books). What a beautiful, colorful, joyful celebration of nature, history + science. Order your own copy + read my article on moss below (or on Amazon). 🌿🍄🌿 www.hup.harvard.edu/books/978067...

And "responsible use" brings more refusal problems with it-- when does a system allow for refusal within an interaction? Where is real-time consent, safety shut-offs, silence, not "guardrails" but exits from the road of sycophancy before it's too late to turn around? dl.acm.org/doi/abs/10.1...

dl.acm.orgdl.acm.org
Emily M. Bender@emilymbender.bsky.social

I was struck by the extent to which this reviewer (and many other people, I'm afraid) have lost sight of the possibility of refusal. And so I want to reaffirm: There is no responsible use without the possibility of refusal. >>

Thomas Arnold reposted

The recommendation of Sara Teasdale now seems unrealistic because of its restraint, rather than-- as it did in high school-- the breadth of the machine's poetic references.

This post is unavailable.

Thomas Arnold reposted

Really disappointed not to see @jgcarpenter.com quoted or cited here- sure seemed like she broke at least half this story on here almost 2 weeks ago

Carl T. Bergstrom@carlbergstrom.com

Y'all remember how they were going to put an AI teacher robot into a high school serving the Seneca Nation of Indians? OF COURSE it was built by a sexbot company. Gift link.

One of the many small details in @karenhao.bsky.social's Empire of AI that has stayed with me, amid many more gruesome others, is how OpenAI took the term "red-teaming" from cybersecurity without any adherence to its criteria and standards-- it just sounded cool, adversarial, accountability-coded.

Eryk Salvaggio@eryk.bsky.social

I think the HF hack was “spin” rather than a marketing stunt. OpenAI should have anticipated what happened, but instead engaged in some sloppy engineering, and then framed it as “but what could we dooooo though?” while shrugging and pointing at the models they built doing what they built it for.

Showing Morass sums up so much of AI vibe-metaphysics

Emily M. Bender@emilymbender.bsky.social

Lots of sophistry and philosophical games. So in a way it's reassuring to see him showing his ass like this. OTOH, he'd probably retreat into some definitional morass about what even is consciousness. >>

Join us for this conversation (DM me for host promocode). This got me to look back at More's Utopia, where I realized Sebald's "Austerlitz" might be riffing off it-- both start with meeting a mysterious man in Antwerp, who unveils truths from colonial exploration. interintellect.com/salons/nowhe...

interintellect.com[Upcoming Event] Nowhere Good: Rethinking Utopia in the 21st CenturyUtopia translates from the Greek outopos as ‘no place’. It is somewhere that can only be reached through the imagination. When he coined the term in his 1516 book Utopia, Thomas More told a story of a...

Thomas Arnold reposted

As a robotics researcher, learning scientist, and author who has followed Realbotix for years, I'm compelled to do a Bluesky thread on this whole pivot that Realbotix is now trying, turning from humanoid sex robots to classroom "education" robots. Buckle up. nysfocus.com/2026/07/14/n... 🧵

nysfocus.comRural NY School To Launch Humanoid Robot TeacherPlans to deploy a "lifelike" robot and avatar teaching assistant at Salamanca High School have sparked debate over the role of technology in schools.

Might be time for a Rorty reboot geared to AI, how the linguistic turn can serve a second-order "representation" and mirroring fantasy. Or a reboot of Wittgenstein, who would have vomited at the idea of LLM's vindicating anything from him other than his ear for Brahms.

Teresa Heffernan@tjheffernan.bsky.social

Finishing a paper on the myth that nature and Big Tech can ‘meet’ with no repercussions...a founding belief in the field is that AI replicates nature: ‘Getting machines to behave like humans and animals has been the quest of my life,’ writes Yann LeCun, former chief scientist at Meta’s AI lab.

Thomas Arnold reposted

Excellent new work out from my lab exemplifying how philosophy, cognitive science, and computer science must put their resources together in order to make headway on characterizing the capabilities of AI systems, especially when those capabilities are in a mentalistic idiom.

Tania Lombrozo@tanialombrozo.bsky.social

Do LLMs *understand* language? Do educational AI agents *understand* the material they teach (or their students)? Claims about what AI systems do or don't understand are pervasive, but assessing them requires an account of MACHINE UNDERSTANDING

It was a pleasure to host yesterday's @interintellect.bsky.social salon with @carissaveliz.bsky.social. The book is superb, an exemplary work of what we sweet summer children used to call a public intellectual.

Carissa Véliz@carissaveliz.bsky.social

...often acting as the avatars of hidden (or not so hidden) power. Written with verve, acuity, and humour, this is essential reading for the times we live in.” To pre-order the #book: www.penguinrandomhouse.com/books/759692...

The theological arachnics she would have unleashed on the name, message, and prophesying of Anthropic appearing at a joint press conference about a papal encyclical... My Adam Curtis doc would note she passed away in 2010, the year of the first ImageNet Large Scale Visual Recognition Challenge...

Thomas Arnold reposted

We are nearing capacity and the event date, and in addition to zeugma I have some free codes to offer. DM me in interested.

Thomas Arnold@thomasarnold.bsky.social

Looking forward to this book and the May @interintellect.bsky.social salon discussing it.

As tech companies have started blaming science fiction for their AI's shambolatry, it's time to face up to Melville's culpability for the Canvas hack.

Nate Wolff@natewolff.bsky.social

Canvas is down, ye say? Aye lad hearken to me and I'll tell thee of a time when a man might say canvas was down! When the leviathan booms his bulk upon ye an' thy splintered masts surrender their sails to the foaming, merciless drink!! then, lad, ye say might say with some authority that can (1/135)