Job SummaryThe Alignment of Dynamical Cognitive Systems program seeks a Research Engineer to develop novel AI platforms addressing critical alignment challenges and practical LLM agents. Combining platform development with alignment research, this role requires expertise in agent-based systems and multi-objective optimization. You'll co-design and develop systems supporting safety research, internal tools, and community resources -- creating environments to shape multi-agent dynamics and implement cooperative AI architectures with robust alignment properties. You'll also build technical infrastructure to investigate experimental AI alignment and control approaches while collaborating with researchers to bring theoretical approaches into practice. This position offers the chance to develop foundational technologies essential for promoting safety as AI capabilities rapidly advance.About CARMAThe Center for AI Risk Management & Alignment (CARMA) works to help society navigate the complex and potentially catastrophic risks arising from increasingly powerful AI systems. Our mission is specifically to lower the risks to humanity and the biosphere from transformative AI.We focus on grounding AI risk management in rigorous analysis, developing policy frameworks that squarely address AGI, advancing technical safety approaches, and fostering global perspectives on durable safety. Through these complementary approaches, CARMA aims to provide critical support to society for managing the outsized risks from advanced AI before they materialize.CARMA is a fiscally-sponsored project of Social & Environmental Entrepreneurs, Inc., a 501(c)(3) nonprofit public benefit corporation.ResponsibilitiesResearch and develop AI systems and platforms variously for: safety research, internal tooling, and safety community supportDesign and implement architectures for agent execution environments and interaction platformsDevelop optimization algorithms for multi-objective and cooperative AI systemsCreate mechanisms for conflict resolution and preference aggregation in multi-agent settingsImplement testing frameworks and example environments to validate theoretical approachesBuild middleware components that facilitate secure and efficient agent communicationArchitect and develop reusable software components for AI alignment researchDocument system architectures, APIs, and implementation detailsCollaborate on technical publications and research presentationsContribute to both practical implementations and theoretical research in AI alignmentSupport the evaluation and refinement of research prototypesRequired QualificationsMS or PhD in Computer Science, AI, or related field (or equivalent experience)Experience with one or preferrably more of: dynamic plan recognition, activity recognition, dynamic multicriteria decision making, multi-agent systems, or AI builder platformsStrong programming skills in both Python and JavaDemonstrated ability to relatively independently drive technical projects from concept to implementationFamiliarity with techniques for AI alignment, control, safety, or related research areasExperience developing middleware, frameworks, or platformsBreadth of experience in machine learning and AI systems architectureLiterate in semi-formal semantics Excellent communication skills and technical documentation abilitiesPreferred QualificationsPublications (formal or informal) in AI alignment, AI safety, or related research areasFamiliarity with varied multi-objective optimization techniquesExperience with contextual pragmatics and/or constraint satisfaction problemsDeveloped novel multi-agent architecturesExperience with LLMs and integrating them into agent architecturesExperience with distributed systems and platform developmentKnowledge of security and privacy methods for multi-party computationUnderstanding of basic game theory and/or social choice theory conceptsFamiliarity with metacognitive systems, metareasoning, or dynamical cognitive approachesDemonstrable open source project contributionsExperience building systems that actively manage trade-offs between competing objectives
21 days ago