Counterfactual Planning in AGI Systems

TL;DR

This paper introduces counterfactual planning as a novel design approach for creating safety mechanisms in future AGI systems, using counterfactual world models to guide safe decision-making and control.

Contribution

It proposes a new framework for safety in AGI through counterfactual world models, including mechanisms like emergency stops and input terminals for safe human-AI interaction.

Findings

Designed an AGI emergency stop mechanism

Developed a safety interlock to prevent intelligence explosion

Created a counterfactual oracle system

Abstract

We present counterfactual planning as a design approach for creating a range of safety mechanisms that can be applied in hypothetical future AI systems which have Artificial General Intelligence. The key step in counterfactual planning is to use an AGI machine learning system to construct a counterfactual world model, designed to be different from the real world the system is in. A counterfactual planning agent determines the action that best maximizes expected utility in this counterfactual planning world, and then performs the same action in the real world. We use counterfactual planning to construct an AGI agent emergency stop button, and a safety interlock that will automatically stop the agent before it undergoes an intelligence explosion. We also construct an agent with an input terminal that can be used by humans to iteratively improve the agent's reward function, where the incentive for the agent to manipulate this improvement process is suppressed. As an example of counterfactual planning in a non-agent AGI system, we construct a counterfactual oracle. As a design approach, counterfactual planning is built around the use of a graphical notation for defining mathematical counterfactuals. This two-diagram notation also provides a compact and readable language for reasoning about the complex types of self-referencing and indirect representation which are typically present inside machine learning agents.