Project summary

We’re raising $10k to deliver the EAGxBerlin pilot (Nov 13-15) and a three-month run at Foresight’s Berlin AI Node.

AI Safety & Society Museum (AISSM, aisafety.museum) builds infrastructure for AI safety public education, starting with in-person exhibits. We commission, test and distribute works about what AI systems can do, their impacts on society, what’s at stake, what good futures could look like and how to get involved, from hands-on demos of model behavior to art and audio pieces about AI in everyday life. A scientific review panel checks for accuracy. Our first open call drew around 100 submissions.

Public discussion about AI tends to be polarized into "AI is dangerous" and "AI will save us" frames, and the public and AI experts diverge in perspective (source). People rarely get to encounter multiple perspectives on AI risks in a live format.

AI safety outreach mostly targets technical audiences, is driven by intuition about the audience interests and concerns, and organizations rarely share what works. AISSM builds two shared resources: a licensed, reviewed collection that conferences, safety hubs and museums can host without producing from scratch, and an open evidence base on effective communication of AI safety ideas. Drawing on insights from informal science education, a lack of standardized measurements is a known barrier to comparing outcomes across exhibitions (source) (source). We aim to develop a shared visitor-response protocol across works and host sites, so that each new venue can add comparable evidence.

Phase 1 tests the collection with high-context audiences at EAGxBerlin and Foresight Institute’s Berlin AI Node. These audiences give fast, informed feedback, and are often considering AI safety careers. Phase 2 takes the strongest works to journalists, policy audiences and the public through museums and public venues.

What's changed since our original request

Our original proposal sought funding for a broader program of development and delivery. Since then, we have secured venues, expanded the team, and narrowed this request to the pilot’s direct production costs.

• Works selected: Ten interactive works selected from a shortlist of 25 submissions by artists, engineers, and ML researchers.

• Team assembled: A curator, scientific reviewers, and experience design support.

• Venues confirmed: EAGxBerlin, followed by a three-month exhibition at Foresight’s Berlin AI Node.

• Interest beyond Berlin: Conversations with Mox (SF), CBAI (Boston), Collider (New York), Trajectory Labs (Toronto), Lighthaven (Berkeley), and LISA (London) about hosting selections.

• Early support: Exhibition space and a €1,000 artist grant from Foresight. Financial support expected from AE Studio. DIS Study Abroad Stockholm found us through our website and is donating a €500 fee to bring 12 US ML students for an exhibition session.

Our mission

Political will, not research, may now be the main bottleneck for AI safety, and it depends on public understanding of AI’s long-term risks and benefits, which is limited by interest and exposure. The broader impact of automation on society is seldom discussed in mainstream public spaces, apart from tech-utopian or anti-tech circles, and most material about AI risk is aimed at specialists.

Excellent digital resources (AI Digest, Rational Animations) and pop-ups like the Misalignment Museum exist and have shown the value of art and design in drawing attention to potential harms from AI. However, there is no replicable, physical collection of interactive demonstrations that community hubs can easily host, where people can interact with AI's risks, benefits and effects on society and discuss them together. Our mission is to build the infrastructure required to bridge this gap. 

Theory of change

Our theory of change is that in-person exhibits, sourced and vetted for accuracy, let people experience what AI can do and what’s at stake, rather than only reading about it. Our pieces bring visitors on a journey through challenging questions about AI, and the futures worth building, to potential governance. By showing expert disagreement, visitors are supported in coming to their own conclusions. Providing balanced, actionable perspectives builds trust and reduces passivity, and visits end with an invitation to get involved. Meeting these questions in person, alongside others, builds lasting interest, bridges the gap between researchers and the public, and supports local field-building at hubs that host the collection. Visitor response summaries are made available to researchers and policymakers, allowing two-way communication between experts and the public, and build an open evidence base to improve AI safety outreach across the field.

Link to ToC Diagram

Planned activities

1. Build the collection: Developing interactive exhibits translating abstract risks into physical experiences. Examples include:

   • Gamified corrigibility demos where visitors try to interrupt an AI

   • Gallery of timelines: 2012 to today, alongside expert forecasts for the future

   • Mech Interp & Observability: what we know and don't know about model internals

   • Expert disagreement: Mapping where experts diverge and why.

2. Run the Pilot (November 2026-January 2027): Launch at EAGxBerlin and the Foresight Institute Berlin AI Node (shared with AI Safety Berlin), and test a first version of visitor surveys.

3. Package the best works: Standardize the best-performing exhibits so they can be shipped or reproduced cheaply anywhere.

4. Global Rollout (2027): Deploy to interested AI safety hubs, collecting comparable evidence from each site.

Success criteria for the pilot

• Engagement: 150+ active in-person visitors in Berlin.

• Learning: 50+ feedback/comprehension responses, and a short public summary of what we learned.

• Assets: At least five works documented for touring (installation guidance, indicative costs, design files).

• Hosts: At least two prospective hubs commit to a 2027 placement.

If the experiences don't draw the audience interaction we're hoping for, we'll change the format as needed.

Without this funding

We can only show a few screen-based works on borrowed hardware, with no capacity to make it replicable.

Core team

• Justin Shenk, PhD (Founder): AI researcher and field-builder. Designed interactive computer vision demos for Intel at CVPR, ICML, NeurIPS; contracted with Redwood Research on AI control evaluations; technical co-founder of an ML startup. Facilitator for BlueDot Impact (Technical AI Safety, AGI Strategy) and mentor for AI Safety Camp and SPAR.

• Curator: Natela Tetruashvili (formerly Contemporary Art Dept, State Hermitage Museum).

• Scientific Reviewers: Three AI safety researchers from AE Studio and Iliad Intensive.

• Advisors/Partners: Archana Vaidheeswaran (Algoverse), Beth McCarthy (Funding the Commons). AI Safety Berlin. Foresight Institute.

What are the most likely causes and outcomes if this project fails?

• Risk: The exhibits feel like standard "online demos" and don't leave a lasting impression.

  • Mitigation: This grant lets us build physical installations rather than placing laptops on desks.

• Risk: Venues lack capacity to host.

  • Mitigation: The collection is designed to be modular. A hub can choose to host 1-2 low-maintenance pieces or the full set depending on their square footage and volunteer capacity.

Even if the pilot underperforms, what we learn will be useful to others doing AI safety outreach.

How much money have you raised in the last 12 months, and from where?

Our first Manifund project asked for a $39k minimum to fund the entire multi-month vision and did not reach it, so this ask covers only the pilot's direct costs.

This $10k allows us to source the highest quality works for the pilot and bridges us to the next round. Follow-on funding for the full program looks likely, and this pilot generates the results needed to secure it. Foresight is contributing space and a €1,000 artist grant, expected support from AE Studio, and DIS Stockholm €500.

The ask

Minimum ($5k): build and install 5 works, mostly screen-based, at EAGxBerlin
Goal ($10k): more physical installations, full pilot including Foresight run and visitor surveys
Generous ($14k): increased outreach support and a more replicable touring kit

See budget sheet for details.

Please reach out if you see an opportunity to collaborate.

Email: info@aisafety.museum.