←
MH Tessler 🇺🇦861 posts
MH Tessler 🇺🇦 avatar

MH Tessler 🇺🇦

@mhtessler

research scientist @GoogleDeepMind. previously MIT, Stanford. interested in society, language, humans, and coffee. enjoying living in the 🇬🇧

◷ Joined June 2011▧ 66 Photos and videos
MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

AI isn't a panacea, but it offers a powerful tool to navigate democratic trade-offs. We need more work on oversight and understanding strategic behavior in AI mediated deliberation, but the path to better citizen engagement is a little more clear we hope. 8/9

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

One intriguing possibility for scale is to use hierarchical aggregation, an "old idea" from the GPT-3 era arxiv.org/abs/2109.10862 but this raises scalable oversight questions: How do we ensure the will of a large population is well captured by AI aggregation? 7/9

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

In our paper, we discuss the following parallels: Equality is about Fairness in AI systems, which the HM has a social choice mechanism to address. ⚖️ Deliberation is about AI assistance. Can AI help us reason through issues better? 🤔 Participation is about Scale... 🌍 6/9

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

We examine the trilemma through the lens of the "Habermas Machine" (HM), an AI system designed to help diverse groups find common ground that we published about in 2024. Can AI help address the challenges of equality, deliberation, and participation? 5/9

I am really beyond words to be able to share with the world what @bakkermichiel, @summerfieldlab, and I and a truly world-class team @GoogleDeepMind worked on for the past 2 years. Out today in @ScienceMagazine, “AI can help humans find common ground in democratic deliberation”

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

Deliberation and Equality can be maximized through structured deliberation protocols (like a Deliberative Poll), but these events are limited in scale due to high costs imposed on participants, not everyone has that sort of time. bloomberg.com/news/newslette… 4/9

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

Participation and Deliberation can be maximized in online fora (like X). Lots of people can participate, and the discussions can go deep. But not all voices are heard equally. 3/9

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·Jan 12

The trilemma was noted by Fishkin (2011). Participation and Equality can be maximized through elections: lots of people can participate in elections and each vote has equal weight, but elections aren't by nature deliberative. People may not know what they are voting for. 2/9

Ryan Lowe avatar
⟳ MH Tessler 🇺🇦 retweeted
Ryan Lowe@ryan_t_lowe·11 Jul 2025

Introducing: Full-Stack Alignment 🥞 A research program dedicated to co-aligning AI systems *and* institutions with what people value. It's the most ambitious project I've ever undertaken. Here's what we're doing: 🧵

Post attachment
Sian Gooding avatar
⟳ MH Tessler 🇺🇦 retweeted
Sian Gooding@SianGooding·2 Apr 2025

🚨 I’m hosting a Student Researcher @GoogleDeepMind! Join us on the Autonomous Assistants team (led by @egrefen ) to explore multi-agent communication—how agents learn to interact, coordinate, and solve tasks together. DM me for details!

Sian Gooding avatar
⟳ MH Tessler 🇺🇦 retweeted
Sian Gooding@SianGooding·26 Mar 2025

New paper alert from @GoogleDeepMind! 🚨 We've put LLMs to the test as writing co-pilots – how good are they really at helping us write? LLMs are increasingly used for open-ended tasks like writing assistance, but how do we assess their effectiveness? 🤔 arxiv.org/abs/2503.19711

MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·20 Mar 2025

Going beyond demographics to model individual variation in human judgments for AI alignment. Very happy to see this great work by Taylor reach the outside!

🤔🤖Most AI systems assume there’s just one right answer—but many tasks have reasonable disagreement. How can we better model human variation? 🌍✨ We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝 arxiv.org/abs/2503.15484 (1/?)

Post attachmentPost attachment
MH Tessler 🇺🇦 avatar
MH Tessler 🇺🇦@mhtessler·11 Mar 2025

Very interesting demonstration of using AI for peace. Incredible effort

Can AI help find common ground during active, real-world conflicts? We worked with @ALLMEP to find common ground between Israeli & Palestinian peacebuilders after Oct 7 2023 using an iterative deliberative process combining LLMs, bridging algorithms, and collective dialogues 🧵

Post attachment