Executive summary: The key to AI alignment is formulating it as a translation problem between AI and human models and interests, rather than an oversight or generalisation problem. This perspective clarifies interpretability approaches and suggests incorporating more human-like inductive biases into AI systems.
Key points:
AI alignment requires finding shared models and plans that respect the interests of both humans and AIs.
Alignment is better framed as a translation problem rather than oversight or generalisation.
“Reverse engineering” interpretability is more productive than mechanistic interpretability.
Incorporating human inductive biases like the “consciousness prior” makes alignment more natural.
Economic incentives could encourage adopting more human-like AI approaches.
Cross-organisational causal model sharing could incentivise compact, causal models.
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: The key to AI alignment is formulating it as a translation problem between AI and human models and interests, rather than an oversight or generalisation problem. This perspective clarifies interpretability approaches and suggests incorporating more human-like inductive biases into AI systems.
Key points:
AI alignment requires finding shared models and plans that respect the interests of both humans and AIs.
Alignment is better framed as a translation problem rather than oversight or generalisation.
“Reverse engineering” interpretability is more productive than mechanistic interpretability.
Incorporating human inductive biases like the “consciousness prior” makes alignment more natural.
Economic incentives could encourage adopting more human-like AI approaches.
Cross-organisational causal model sharing could incentivise compact, causal models.
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.