Alignment Mapping pilot program tries to foster independent thinking in AI safety research

The Alignment Mapping Program (AMP) launched in 2023 to develop independent thinking skills among AI safety researchers through an intensive 8-week curriculum.

Program overview and mission: AMP bridges the gap between foundational AI safety courses and advanced research programs by emphasizing active construction of mental models rather than passive learning.

• The program targets a critical need in AI safety education: developing researchers who can critically evaluate existing frameworks and generate novel approaches
• AMP’s curriculum focuses on having participants build their own comprehensive maps of the AI alignment problem space
• The program ran five cohorts in 2024 with approximately 25 total participants

Core curriculum structure: The program follows a three-phase approach designed to build independent thinking skills.

• Phase 1 (Weeks 1-3): Participants create visual maps of AI alignment problems, potential solutions, and personal contribution paths
• Phase 2 (Weeks 4-7): Students analyze established researchers’ work by comparing it to their own models through “shoulder mentors” – simplified frameworks of researcher approaches
• Phase 3 (Week 8): Participants develop concrete research plans and goals based on their mapped insights

Key program outcomes: Initial data from the pilot program revealed both successes and areas needing improvement.

• High engagement and positive feedback characterized the first three weeks of the program
• 60% of survey respondents recommended making AMP a core component of AI safety researcher education
• The program experienced significant participant drop-off (-30%) after week three
• Participants found initial reading requirements excessive and some exercises lacking clarity

Implementation improvements: Program administrators are making data-driven adjustments based on participant feedback.

• Restructuring weeks 4-7 to focus on fewer researchers or shift to a project-based approach
• Reducing reading requirements to 2-3 hours per week with tighter integration into mapping exercises
• Developing clearer exercise instructions and detailed examples of problem maps
• Exploring both in-person workshop formats and condensed 5-week versions

Future development plans: The program seeks to expand its impact through strategic partnerships and format innovations.

• Program leaders are seeking partnerships with other AI safety initiatives to incorporate AMP into existing curricula
• New program variations are being developed to address retention challenges while maintaining educational impact
• The team is actively soliciting community feedback on concepts like “shoulder mentors” and independent thinking development

Critical considerations: While AMP shows promise in developing independent thinking skills for AI safety researchers, its effectiveness will likely depend on successful implementation of feedback-driven improvements and careful attention to participant engagement patterns.