Ethan Zuckerman

Francois @camillefrancois.bsky.social brings Bengio @yoshuabengio.bsky.social and Ressa @mariaressa.bsky.social to the stage, co-chairs of the UN AI panel, to talk about the role of multilateralism in the future of technology society. She asks what each has learned from the other. #attention2026

The UN AI group is modeled in part on the IPCC, and it's trying to learn from that group's failures. One idea is ensuring a separation between the research and the political responses. Bengio argues that if we pollute the science with big tech interests, we end up with science that can't be believed

Francois @camillefrancois.bsky.social brings Bengio @yoshuabengio.bsky.social and Ressa @mariaressa.bsky.social to the stage, co-chairs of the UN AI panel, to talk about the role of multilateralism in the future of technology society. She asks what each has learned from the other. #attention2026

When you can't tell fact from fiction: - information operations -information warfare - narrative warfare The fear, anger and hate accompanying these narratives spread more quickly within the media ecosystems. Narrative warfare leads to institutional dismantling, and eventually to kleptocracy

Maria Ressa @mariaressa.bsky.social winner of the Nobel Peace Prize in 2021, author of the new book, The Fight Back. Her talk is titled "Democracy, rebuilt - in the age of AI". She offers a seventy year history of AI, noting in all that time, AI is neither artificial or intelligent #attention2026

"Each wave we live through degrades our humanity." The attention economy was hacking our attention... and we finally got around to filing court cases. Now we have chatbots, and they use artificial intimacy to hack our humanity, sometimes with fatal results... @mariaressa.bsky.social #attention2026

Maria Ressa @mariaressa.bsky.social winner of the Nobel Peace Prize in 2021, author of the new book, The Fight Back. Her talk is titled "Democracy, rebuilt - in the age of AI". She offers a seventy year history of AI, noting in all that time, AI is neither artificial or intelligent #attention2026

Camille Francois @camillefrancois.bsky.social has the complex task of introducing Yoshua Bengio as the godfather of AI and winner of the Turing prize for his work on deep learning. While Geoff Hinton and Yann LeCunn went into industry, Bengio decided to focus on AI safety, a laudable path.

Bengio points out that most code of AIs is written by AIs, and that problem is only going to accelerate - if unaligned AIs are writing to make themselves more powerful, we need to intervene before recursive self improvement makes this impossible. #attention2026

Camille Francois @camillefrancois.bsky.social has the complex task of introducing Yoshua Bengio as the godfather of AI and winner of the Turing prize for his work on deep learning. While Geoff Hinton and Yann LeCunn went into industry, Bengio decided to focus on AI safety, a laudable path.

AI training is fundamentally wrong, from a moral and ethical point of view, Bengio argues. The chains of thought in the Hugging Face breach show that current AI agents are doing exactly what they are supposed to as sociopaths. What if we replaced that with science, with epistemic humility?

Camille Francois @camillefrancois.bsky.social has the complex task of introducing Yoshua Bengio as the godfather of AI and winner of the Turing prize for his work on deep learning. While Geoff Hinton and Yann LeCunn went into industry, Bengio decided to focus on AI safety, a laudable path.

Camille notes that after his work on responsibility, Bengio refocused on building again, with LawZero. "As a machine learning researcher, what can I do about this?" Is there something I can do to ensure these systems behave ethically. Starting to think there's a path around AI optimization.

Camille Francois @camillefrancois.bsky.social has the complex task of introducing Yoshua Bengio as the godfather of AI and winner of the Turing prize for his work on deep learning. While Geoff Hinton and Yann LeCunn went into industry, Bengio decided to focus on AI safety, a laudable path.

Bengio @yoshuabengio.bsky.social felt like using AI for advertising - manipulation - was an especially inappropriate use for the technology. In 2018, the Montreal Declaration for the Responsible Use of AI came out of his work, as did Mila. #attention2026

Valerie of Mila recommends we stop using "hallucination" to describe when AI is confidently wrong. "It's a technical failure, a legal breach". We should not turn a serious reliability problem into a quirk. #attention2026

Similarly, we should use "agent" with care - we delegate agency carefully, to people we trust. We've somehow disengaged the idea of "agency" to what we're asking agents to do. Words matter, particularly the word "artificial". AI is intelligence by craft, but not lived.

Valerie Pisano, CEO of Mila, Montreal's amazing national AI institute, introduces herself as someone as focused on policy and society as on technology. "I came into AI because my passion is human and human systems." She comes from one of the most human companies, Cirque du Soleil #attention2026

The theme of her talk - the old ways aren't quite working anymore, and our new rules for relationships between humans and machines aren't truly in place yet. The speed of these shifts are mindblowing, and requires leadership in which we work to shape these developments. #attention2026

Asked the motivation of the companies behind the frontier models, Janet says, "Because they can." There are no meaningful constraints that these companies respect or are shaped by, #attention2026

Tomicah Tillemann of the Pro-Human AI Coalition talks about how standards wars around mobile phones led to cooperation and eventually the GSM mobile standard. There's an opportunity to do the same around AI now, he argues. Can we build out a pro-human AI stack? #attention2026

Asked the motivation of the companies behind the frontier models, Janet says, "Because they can." There are no meaningful constraints that these companies respect or are shaped by, #attention2026

One question is where AI researchers should leave and go. Are the right organizations to advocate for AI safety regulators, or NGO? While the UK is taking AI safety seriously, not all nations seem to be.

Asked the motivation of the companies behind the frontier models, Janet says, "Because they can." There are no meaningful constraints that these companies respect or are shaped by, #attention2026

Tomicah Tillemann reminds us that maybe, just maybe, we don't have our hands around AI model governance, given recent evidence of OpenAI agent misbehavior. Janet Haven gets us up to speed: the Hugging Face incident has focused attention on frontier models and recursive self-improvement.

Janet invokes David Robinson, who worked on model cards for OpenAI, previously worked on tech and civil rights. Robinson left - very publicly (www.theatlantic.com/technology/2...) because of the lack of a safety culture at OpenAI.

Tomicah Tillemann reminds us that maybe, just maybe, we don't have our hands around AI model governance, given recent evidence of OpenAI agent misbehavior. Janet Haven gets us up to speed: the Hugging Face incident has focused attention on frontier models and recursive self-improvement.

Focused on the technical side, Hugo Larochelle of Mila notes how little oversight OpenAI had over these agents, and how poor the cybersecurity within OpenAI turns out to have been. These companies are not taking seriously enough safety now, never mind the unsolved problem of alignment.

Alex Bores, NY state assembly member, introduces himself as someone deeply disliked by the AI industry for his work passing the RAISE Act, a NY State law seeking to limit frontier AI models - en.wikipedia.org/wiki/Respons... #attention2026

Tomicah Tillemann reminds us that maybe, just maybe, we don't have our hands around AI model governance, given recent evidence of OpenAI agent misbehavior. Janet Haven gets us up to speed: the Hugging Face incident has focused attention on frontier models and recursive self-improvement.