Towards Shutdownable Agents: Generalizing Stochastic Choice in RL Agents and LLMs
For AI safety researchers, this provides early evidence that agents can be trained to be shutdownable while maintaining performance, addressing a key alignment concern.
The paper introduces DReST, a reward function that trains agents to be neutral about trajectory length (stochastic choice) while remaining useful. DReST-trained RL agents achieved 11% (PPO) and 18% (A2C) higher usefulness than baselines, and fine-tuned LLMs achieved maximum usefulness with near-maximum neutrality.
Misaligned artificial agents might resist shutdown. One proposed solution is to train agents to lack preferences between different-length trajectories. The Discounted Reward for Same-Length Trajectories (DReST) reward function does this by penalizing agents for repeatedly choosing same-length trajectories, and thus incentivizes agents to (1) choose stochastically between different trajectory-lengths (be Neutral about trajectory-lengths), and (2) pursue goals effectively conditional on each trajectory-length (be Useful). In this paper, we use DReST to train deep RL agents and fine-tune LLMs to be Neutral and Useful. We find that these DReST agents generalize to being Neutral and Useful in unseen contexts at test time. Indeed, DReST RL agents achieve 11% (PPO) and 18% (A2C) higher Usefulness on our test set than baseline agents, and our fine-tuned LLM achieves maximum Usefulness and near-maximum Neutrality. Our results provide some early evidence that DReST could be used to train more advanced agents to be Useful and Neutral. Prior theoretical work suggests that these agents would be useful and shutdownable.