AI Safety Lunch + Coworking @ CIC

​Join AI safety researchers, policy makers, communicators, field builders and others working on making AI go well.

10:00–18:00 Free coworking within the Æthos space @ CIC
13:00 Introductions and pitches (all optional)
13:15 Lunch @ CIC - buy lunch from Coco Restaurant (~10€, self-paid) or bring your own lunch
14:00 Talk - Non-Agentic Components for Safe-By-Design AI? by Jobst Heitzig (see details below)

We recommend to arrive at 10:00 for coworking or at 13:00 for the intro round + lunch, but you can also join later, and stay as long as you want.

​Talk: Non-Agentic Components for Safe-By-Design AI? 🗣️

​A central idea of many "safe-by-design" approaches is that the smartest component of the system remains non-agentic.

​One example is LawZero's "Scientist AI", which is meant to be a disinterested oracle assessing potential harms caused by other (less capable) AI agents' possible actions. Another is the "truthful world model" that ARIA's "Safeguarded AI" and my own human-empowerment-based approach require.

​In this talk, Jobst will describe why such designs might be safe(r), then discuss challenges in building such non-agentic components, including his own solution to one particular challenge: useful predictions are typically performative—the prediction influences what it predicts.

​This will lead into a lively general discussion about the overall idea of safe-by-design AI systems involving non-agentic components. Read this paper if you want to dive deep beforehand.

​👋 Newbies welcome! 👋

People new to the AI risk / AI safety community are welcome. If you're new to the field, we recommend having a quick look at the introductory resources on aisafety.berlin/learn to get a rough overview (you don't need to read everything). Come say hi to one of the hosts (Alex, Trevor or Manuel). We'd be happy to get to know you, answer questions and introduce you to other community members.

Directions to the CIC 📍

​The AI Safety coworking happens on the 3rd floor of Cambridge Innovation Center in the Æthos space.

If you arrive at 10a for coworking or at 1p for lunch, wait in the lobby and we'll come down to pick you up. If you arrive at another time, message Trevor (@FastFedora) on Telegram or via SMS at +49 163 627 6801 or Manuel (@manuelallgaier) on Telegram or via SMS at +49 160 9234 3485.

​About coworking at the CIC / Æthos 🧑‍💻

We're renting desks within the Æthos space at the Cambridge Innovation Center (CIC), directly adjacent to the Foresight Institute's Berlin Node for studying AI for Science & Safety. Researchers, policy makers, advocates, field builders and others interested in AI safety & governance are welcome to join us! Ask Trevor or Alex for details.

About AI Safety Berlin 👬

We're a network of Berliners working in research, policy and field building to ensure that future AI systems are safe and beneficial for all. If you're new to the AI Safety Community, have a look at aisafety.berlin/learn and aisafety.com beforehand. This can be helpful but it's not required.

Feedback & Contact ✉️

Feedback? Suggestions for future coworking, events, speakers or topics? --> Reach out! You can also submit anonymous feedback via this form.

We're looking forward to meeting you!

Manuel, Alex, Trevor & the rest of the AI Safety Berlin Team

(Note that we may take photos, some of which we may use for our website. If you don't want us to use your photo, we can AI remove you, just let us know at the event)

------------

Join the AI Safety Berlin Community Chat on Telegram for announcements and discussions, or for announcements only, subscribe on Telegram, Signal or Whatsapp.