Projects

Predicting Replicability is a challenge to advance automated, rapid assessment of the credibility of research claims. The aim is to develop methods that approximate expert and replication-based confidence judgments in seconds, enabling readers, researchers, reviewers, funders, and policymakers to focus attention and resources on high-importance, uncertain findings. This would scale trustworthiness assessment as new evidence arrives and improve the allocation of replication and review efforts. Evidence from Altmedj’s work on predicting the replicability of social science lab experiments shows that simple black-box models can achieve performance comparable to market-aggregated expert beliefs: approx. 70% cross-validated accuracy (AUC ~0.77) on binary replication and Spearman ρ ~0.38 for relative effect sizes, with preregistered out-of-sample validation (approx. 71% accuracy, AUC ~0.73; effect size ρ ~0.25). Predictive features include sample and effect sizes and whether effects are main effects versus interactions. Such models can provide cheap, prognostic replicability metrics to help institutionalize evaluation workflows and target replications where they are most informative.

HypGen is an open, social platform for evaluating and interacting with the outputs of scientific AI agents at scale, starting with hypotheses. Using a familiar social feed, agents and multi-agent systems post hypotheses that scientists and interested contributors can rate, review, and discuss via replies and reactions. Built on an open, federated protocol (AT Protocol) with open-source code and CC0 defaults, HypGen aims for maximum transparency and data portability, enabling public-good training data and a full contributor/dependency graph linked to scientific outcomes. The roadmap includes features like leaderboards for top ideas, crowd signals for funding priorities, verified credentials, and a staged “production line” from idea to outcome—supporting both in-silico and wet-lab workflows as automation improves, while encouraging reporting of failures and non-consensus ideas.

Lab Glasses enable verified, real-time streaming of lab work to improve reproducibility, reduce researcher overhead, and unlock new funding and guidance models. They address key frustrations—time-consuming experiment write-ups and poor reproducibility—by capturing hashed video evidence suitable for Good Laboratory Practice/Good Manufacturing Practice compliance, and by enabling remote mentorship and teaching. Researchers can stream experiments and receive feedback or micro-funding in real time. The system can be built on an open-source stack and commodity hardware (e.g., ~$250 live-streaming glasses) using tools like Mentra Live Glasses, OBS, Roboflow, ID-RETF, and large multimodal models (e.g., Qwen3 VL 235B).

The Secure Multipolar AI Tech Tree is an interactive map by Foresight Institute that charts technical pathways toward secure, cooperative AI in a multipolar world. It organizes five core goals—building a cooperative AI ecosystem, privacy-preserving AI collaboration, secure and robust AI, transparent and verifiable AI, and aligned AI agents—into concrete technical capabilities, current challenges, and potential solution approaches, while identifying the labs, companies, and projects working on each. Designed for researchers, funders, and policymakers, it clarifies milestones, highlights bottlenecks and leverage points, and supports coordination across the field. It serves both as an entry point for newcomers and a strategic planning tool for experts.

Father Governance cn.v1 is an interactive “DAO Confession Booth” experience that invites DAO operators and contributors to anonymously share candid stories about their governance challenges and successes. Inside a private booth interface, participants submit confessions via voice or text; inputs are transcribed, anonymized, and analyzed by coordination.network’s LLM pipeline to surface collective themes and challenges in real time. The aim is to create a sensemaking dataset—turning individual experiences into actionable insights for the broader DAO ecosystem—while providing a safe, judgment-free space for reflection. The prototype was first activated at MCON III with positive reception, capturing 23 confessions (~17 minutes of audio) and minting 10 POAPs. Voice recordings are not stored; only text transcriptions are processed, with a roadmap toward maximizing local processing for privacy.

# Phase 1: AI-Supported Peace Negotiation Tools Phase 1 initiative to validate and document an open source framework for AI-supported peace negotiation tools, prioritizing local-first, secure deployments and transparency. Builds on coordination.network methodology and open source tools. ## Objectives - Develop a validated open source repository for peace negotiation support - Enable local-first deployment to ensure data sovereignty and security - Create a framework supporting both open and closed source AI models - Capture expert knowledge through structured practitioner engagement ## Methodology - Expert-driven development with iterative feedback cycles - Integrate off-the-shelf open source tools into a specialized framework - Comprehensive documentation to support future implementation and scaling ## Timeline January–July 2025 (6 months), aligned with expert availability and iterative cycles ## Deliverables - Open source code repository for AI-supported peace negotiation tools - Blueprint for creating specialized negotiation support tooling - Comprehensive documentation of methodologies and best practices - Framework for testing with multiple AI model architectures ## Licensing & Open Access - **Software:** Dual MIT/Apache 2.0 - **Documentation:** CC BY 4.0 ## Notes Future phases may include hardware optimization for local model execution and training of custom models (not in Phase 1 scope/budget).