In our research we investigate the output accuracy of discrete event simulation models and agent based simulation models when studying human centric complex systems. In this paper we focus on human reactive behaviour as it is possible in both modelling approaches to implement human reactive behaviour in the model by using standard methods. As a case study we have chosen the retail sector, and here in particular the operations of the fitting room in the women wear department of a large UK department store. In our case study we looked at ways of determining the efficiency of implementing new management policies for the fitting room operation through modelling the reactive behaviour of staff and customers of the department. First, we have carried out a validation experiment in which we compared the results from our models to the performance of the real system. This experiment also allowed us to establish differences in output accuracy between the two modelling methids. In a second step a multi-scenario experiment was carried out to study the behaviour of the models when they are used for the purpose of operational improvement. Overall we have found that for our case study example both discrete event simulation and agent based simulation have the same potential to support the investigation into the efficiency of implementing new management policies.

The RADAR project developed a large multi-agent system with a mixed-initiative user interface designed to help office workers cope with email overload. Most RADAR agents observe experts performing tasks and then assist other users who are performing similar tasks. The interaction design for RADAR focused on developing user interfaces that allowed the intelligent functionality to improve the user's workflow without frustrating the user when the system's suggestions were either unhelpful or simply incorrect. A large evaluation of RADAR demonstrated that novice users confronted with an email overload test performed significantly better, achieving a 37% better overall score when assisted by RADAR.

The RADAR project developed a large multi-agent system with a mixed-initiative user interface designed to help office workers cope with email overload. Most RADAR agents observe experts performing tasks and then assist other users who are performing similar tasks. The interaction design for RADAR focused on developing user interfaces that allowed the intelligent functionality to improve the user’s workflow without frustrating the user when the system’s suggestions were either unhelpful or simply incorrect. For example with regards to autonomy, the RADAR agents were allowed much flexibility in selecting ways to assist the user, but were restricted from taking actions that would be visible to other people. This policy ensured that the user remained in control and mitigated the negative effects of mistakes. A large evaluation of RADAR demonstrated that novice users confronted with an email overload test performed significantly better, achieving a 37% better overall score when assisted by RADAR. The evaluation showed that AI technologies can help users accomplish their goals.

These unintended negative consequences of the introduction of intelligence often have no direct relationship with the intended benefits, just as the adverse effects of a medication may bear no obvious relationship to the intended benefits of taking that medicine. Therefore, these negative consequences can be seen as side effects. The purpose of this article is to give designers, developers, and users of interactive intelligent systems a detailed awareness of the potential side effects of AI. As with medications, awareness of the side effects can have different implications: We may be relieved to see that a given side effect is unlikely to occur in our particular case. We may become convinced that it will inevitably occur and therefore decide not to "take the medicine" (that is, decide to stick with mainstream systems). Or most likely and most constructively, by looking carefully at the causes of the side effects and the conditions under which they can occur, we can figure out how to exploit the benefits of AI in interactive systems while avoiding the side effects.

Everyday social interactions are heavily influenced by our snap judgments about others' goals. Even young infants can infer the goals of intentional agents from observing how they interact with objects and other agents in their environment: e.g., that one agent is'helping' or'hindering' another's attempt to get up a hill or open a box. We propose a model for how people can infer these social goals from actions, based on inverse planning in multiagent Markov decision problems (MDPs). The model infers the goal most likely to be driving an agent's behavior byassuming the agent acts approximately rationally given environmental constraints andits model of other agents present.

In partially observable worlds with many agents, nested beliefs are formed when agents simultaneously reason about the unknown state of the world and the beliefs of the other agents. The multi-agent filtering problem is to efficiently represent and update these beliefs through time as the agents act in the world. In this paper, we formally define an infinite sequence of nested beliefs about the state of the world at the current time $t$ and present a filtering algorithm that maintains a finite representation which can be used to generate these beliefs. In some cases, this representation can be updated exactly in constant time; we also present a simple approximation scheme to compact beliefs if they become too complex. In experiments, we demonstrate efficient filtering in a range of multi-agent domains.

Young children demonstrate the ability to make inferences about the preferences of other agents based on their choices. However, there exists no overarching account of what children are doing when they learn about preferences or how they use that knowledge. We use a rational model of preference learning, drawing on ideas from economics and computer science, to explain the behavior of children in several recent experiments. Specifically, we show how a simple econometric model can be extended to capture two- to four-year-olds’ use of statistical information in inferring preferences, and their generalization of these preferences.

Extensive games are often used to model the interactions of multiple agents within an environment. Much recent work has focused on increasing the size of an extensive game that can be feasibly solved. Despite these improvements, many interesting games are still too large for such techniques. A common approach for computing strategies in these large games is to first employ an abstraction technique to reduce the original game to an abstract game that is of a manageable size. This abstract game is then solved and the resulting strategy is used in the original game. Most top programs in recent AAAI Computer Poker Competitions use this approach. The trend in this competition has been that strategies found in larger abstract games tend to beat strategies found in smaller abstract games. These larger abstract games have more expressive strategy spaces and therefore contain better strategies. In this paper we present a new method for computing strategies in large games. This method allows us to compute more expressive strategies without increasing the size of abstract games that we are required to solve. We demonstrate the power of the approach experimentally in both small and large games, while also providing a theoretical justification for the resulting improvement.

Solving multi-agent reinforcement learning problems has proven difficult because of the lack of tractable algorithms. We provide the first approximation algorithm which solves stochastic games to within $\epsilon$ relative error of the optimal game-theoretic solution, in time polynomial in $1/\epsilon$. Our algorithm extends Murrays and Gordon’s (2007) modified Bellman equation which determines the \emph{set} of all possible achievable utilities; this provides us a truly general framework for multi-agent learning. Further, we empirically validate our algorithm and find the computational cost to be orders of magnitude less than what the theory predicts.

The Minimum Description Length (MDL) principle selects the model that has the shortest code for data plus model. We show that for a countable class of models, MDL predictions are close to the true distribution in a strong sense. The result is completely general. No independence, ergodicity, stationarity, identifiability, or other assumption on the model class need to be made. More formally, we show that for any countable class of models, the distributions selected by MDL (or MAP) asymptotically predict (merge with) the true measure in the class in total variation distance.