Executive summary: In this exploratory dialogue between Audrey Tang and Plex, they probe whether “symbiogenesis” (hyperlocal, community-first cooperation that scales up) can stably beat convergent, power-seeking consequentialism, with Plex remaining skeptical that bounded/steerable systems can survive competitive pressure without a unifying, theory-level alignment that scales to superintelligence, and Audrey arguing that practicing alignment on today’s systems, strengthening defense-dominant communities, and iterating hyperlocal “civic care” and Coherent Blended Volition (CBV) may bootstrap a viable path—while both endorse improved sensemaking, shared vocabularies, and cautious experimentation.
Key points:
Core crux: Can complex cooperation (“symbiogenesis”) remain stable against selection for unbounded optimizers? Plex doubts boundedness survives competitive dynamics without a top-level, enforceable norm; Audrey thinks hyperlocal alignment plus defense-dominant coordination can scale and police defectors.
Economic pressure vs. safety: Plex argues unbounded systems will outcompete bounded/steerable ones (profit and influence gradients), making mere norms or lip service insufficient; Audrey counters with Montreal-Protocol-style, technology-forcing governance and claims steerable systems can deliver value and thus win investment.
Robustness requirement: Plex maintains that before strong agentic AIs, we likely need a general alignment theory that “tiles” through self-improvement (avoids sharp left turns and Goodharted proxies); Audrey frames robustness as strategy-proof rules and bounded “Kamis” (local stewards) loyal to relationships and communities.
Hyperlocal morality as scaffold: Audrey claims solving morality locally (quasi-utilitarianism/care ethics, subsidiarity/Ostrom) can recurse up via “dividuals” to produce stable higher-level coherence; Audrey worries local wins may aggregate into alien global outcomes that today’s humans wouldn’t endorse.
Coordination + sensemaking now: Both see immediate value in aligning current recommender systems, building shared vocabularies across alignment subfields, and running safe simulations (e.g., Metta’s “Clips vs. Cogs”) to test group-dynamics claims—while noting experiments won’t replace a theory expected to scale.
Practical implications: Focus on defense-dominant pods, transparent dashboards with awareness of Goodhart risks, participatory “utopiography”/clustered-volition processes (Weval/Global Dialogue), and cross-ontology translation; plex recommends engaging with MIRI and similar communities, and he remains cautiously supportive of AudreyThis comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, andcontact us if you have feedback.’s approach as a pathway to buy time and improve global strategy.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, andcontact us if you have feedback.
Executive summary: In this exploratory dialogue between Audrey Tang and Plex, they probe whether “symbiogenesis” (hyperlocal, community-first cooperation that scales up) can stably beat convergent, power-seeking consequentialism, with Plex remaining skeptical that bounded/steerable systems can survive competitive pressure without a unifying, theory-level alignment that scales to superintelligence, and Audrey arguing that practicing alignment on today’s systems, strengthening defense-dominant communities, and iterating hyperlocal “civic care” and Coherent Blended Volition (CBV) may bootstrap a viable path—while both endorse improved sensemaking, shared vocabularies, and cautious experimentation.
Key points:
Core crux: Can complex cooperation (“symbiogenesis”) remain stable against selection for unbounded optimizers? Plex doubts boundedness survives competitive dynamics without a top-level, enforceable norm; Audrey thinks hyperlocal alignment plus defense-dominant coordination can scale and police defectors.
Economic pressure vs. safety: Plex argues unbounded systems will outcompete bounded/steerable ones (profit and influence gradients), making mere norms or lip service insufficient; Audrey counters with Montreal-Protocol-style, technology-forcing governance and claims steerable systems can deliver value and thus win investment.
Robustness requirement: Plex maintains that before strong agentic AIs, we likely need a general alignment theory that “tiles” through self-improvement (avoids sharp left turns and Goodharted proxies); Audrey frames robustness as strategy-proof rules and bounded “Kamis” (local stewards) loyal to relationships and communities.
Hyperlocal morality as scaffold: Audrey claims solving morality locally (quasi-utilitarianism/care ethics, subsidiarity/Ostrom) can recurse up via “dividuals” to produce stable higher-level coherence; Audrey worries local wins may aggregate into alien global outcomes that today’s humans wouldn’t endorse.
Coordination + sensemaking now: Both see immediate value in aligning current recommender systems, building shared vocabularies across alignment subfields, and running safe simulations (e.g., Metta’s “Clips vs. Cogs”) to test group-dynamics claims—while noting experiments won’t replace a theory expected to scale.
Practical implications: Focus on defense-dominant pods, transparent dashboards with awareness of Goodhart risks, participatory “utopiography”/clustered-volition processes (Weval/Global Dialogue), and cross-ontology translation; plex recommends engaging with MIRI and similar communities, and he remains cautiously supportive of AudreyThis comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.’s approach as a pathway to buy time and improve global strategy.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.