One Decade of Universal Artificial Intelligence

Abstract

The first decade of this century has seen the nascency of the first mathematical theory of general artificial intelligence. This theory of Universal Artificial Intelligence (UAI) has made significant contributions to many theoretical, philosophical, and practical AI questions. In a series of papers culminating in book (Hutter, 2005), an exciting sound and complete mathematical model for a super intelligent agent (AIXI) has been developed and rigorously analyzed. While nowadays most AI researchers avoid discussing intelligence, the awardwinning PhD thesis (Legg, 2008) provided the philosophical embedding and investigated the UAI-based universal measure of rational intelligence, which is formal, objective and non-anthropocentric. Recently, effective approximations of AIXI have been derived and experimentally investigated in JAIR paper ). This practical breakthrough has resulted in some impressive applications, finally muting earlier critique that UAI is only a theory. For the first time, without providing any domain knowledge, the same agent is able to self-adapt to a diverse range of interactive environments. For instance, AIXI is able to learn from scratch to play TicTacToe, Pacman, Kuhn Poker, and other games by trial and error, without even providing the rules of the games.These achievements give new hope that the grand goal of Artificial General Intelligence is not elusive.This article provides an informal overview of UAI in context. It attempts to gently introduce a very theoretical, formal, and mathematical subject, and discusses philosophical and technical ingredients, traits of intelligence, some social questions, and the past and future of UAI.

Toward an Algorithmic Metaphysics

“…Work such as [8] and [7] extend Solomonoff's original insights for formalizing inductive reasoning. The Bayesian approach of [16] is closely related (as argued in [17]), and has demonstrated potential to characterize approximate truth [4] and perhaps more elusive philosophical fruit [3].…”

“…At least for the simple W , though, I think AIT can provide a relatively intuitive answer to this vexed question. 8 This definition of composite objects will serve as a kind of lynchpin for the synchronic metaphysics that follow.…”

Universal Semantic Communication

“…The aforementioned constructions of optimal agents are also reminiscent of some results by Hutter [78]. There are a few differences: the first one is that Hutter's results are proved in a control-theoretic (reinforcement learning style) setting, in which the environment provides the agent with the value of its payoff at each moment in time.…”

“…One difference in our work is that the goals, rather than the agents, come to play the central role in our study. In particular, a crucial difference between our setting and some settings considered elsewhere (e.g., the universal agents of Hutter [78]) is that we do not assume that "utilities" are computed for us by the environment-in our setting, in general it will be up to the agent to decide whether or not it is satisfied.…”

Sequential Decision Theory

“…The probabilistic formulation in Section 10.1 is a basic problem of stochastic control theory [11,27]. The framework is also popular in artificial intelligence [6,17,25,40]. For an early, influential work on stochastic control, see [14], in which the notion of sequential games against nature is developed.…”