Memory Issues of Intelligent Agents

Abstract

Abstract. Theoretical models of artificial general intelligence, such as AIXI [3], typically consider an intelligent agent to have unlimited computational resources, allowing it to keep a perfect memory of its entire interaction history with its environment. In the real world, an agent's memory is part of the environment, which means that the latter can modify the former. This paper develops a theoretical framework for examining the implications of such real-world memory on universal intelligent agents. Within this framework we are able to show, for example, that in certain environments optimality can be achieved only with truly stochastic behaviors, and that guarantees about the trustworthiness of memories are difficult to obtain even with infinite computational power. To describe the probability of an agent's memory state, we propose an adaptation of the universal prior for the passive and the active case.

Intelligent agent-assisted organizational memory in knowledge management systems

“…One of the conclusions reached by the authors was the value of a "chief memory officer" or an organizational historian to act as a human agent and champion of organizational memory. It has since been identified that artificial IAs may aid human agents in the preservation of organizational memory (Orseau and Ring, 2012;Ring and Orseau, 2011).…”

Space-Time Embedded Intelligence

“…The notation is similar to that of Orseau & Ring [5,8,6]. At some time t the agent outputs actions a t ∈ A to the environment, which returns observations o t ∈ O to the agent.…”

“…Thinking of ρ as a non-universal but accurate model of the real world is also acceptable (although arguably non-realistic). 6 This is an uncomputable definition, but the solution of this equation is a computable optimal resource-bounded agent.…”

“…Such considerations were partially addressed in our definition of the optimal, selfmodifying, universal agent [5], whose source code was part of the environment itself, both readable and modifiable by it. A companion paper [6] considers the consequences of doing the same with the agent's memory of the past (but not its source code).…”

“…And in a companion paper [6], we consider agents whose memory is integrated into and can be altered by the environment. In the present paper, we formulate a more generalized frameworkmore reminiscent of physicalism [13] than dualism-where agents are fully integrated into their environment: not just modifiable by it, but actually computed by it.…”