Anthropoid agents

In the study of agents on the Internet, we often ascribe to them human qualities, such as beliefs and intentions. These qualities are best understood as metaphors that give developers a way to talk about and design the capabilities and applications of agents. Despite all the progress in computing, users have been slow to accept the technology. They have often accepted what was thrown at them, but only under economic duress. Bringing the technology closer to their emotional needs might ease this resistance. So how can we put a human face on computing? Maybe by putting an animated face on it! Thus, an interface may consist of an agent, which has an explicit presence (for example, as an on-screen animated figure) and appears to have a personality. In computer human interfaces, especially for education and commerce where a large variety of people must deal with computers, an anthropoid agent might be more inviting. Depending on the situation, the agent might appear shy, friendly, stern, or knowledgeable. For example, people might better accept advice offered politely by a shy agent, or heed warnings uttered seriously by a stern agent. And they might be more likely to purchase goods or services offered by a friendly, knowledgeable agent that could appear empathetic to their needs     

Let's talk! Socially intelligent agents for language conversationtraining

This paper promotes socially intelligent animated agents for the pedagogical task of English conversation training for native speakers of Japanese. As a novel feature, social role awareness is introduced to animated conversational agents, that are by non-strong affective reasoners, but otherwise often lack the social competence observed in humans. In particular, humans may easily adjust their behavior depending on their respective role in a social setting, whereas their synthetic pendants tend to be driven mostly by emotions and personality. Our main contribution is the incorporation of a “social filter program” to mental models of animated agents. This program may qualify an agent's expression of its emotional state by the social contest, thereby enhancing the agent's believability as a conversational partner. Our implemented system is web-based and demonstrates socially aware animated agents in a virtual coffee shop environment. An experiment with our conversation system shows that users consider socially aware agents as more natural than agents that violate conventional practices     

Specification by-example of virtual agents behavior

The development of virtual agents running within graphic environments which emulate real-life contexts may largely benefit from the use of visual specification by-example. To support this specification, the development system must be able to interpret the examples and cast their underlying rules into an internal representation language. This language must find a suitable trade-off among a number of contrasting requirements regarding expressiveness, automatic executability, and suitability to the automatic representation of rules deriving from the analysis of examples. A language is presented which attains this trade-off by combining together an operational and a declarative fragment to separately represent the autonomous execution of each individual agent and its interaction with the environment, respectively. While the declarative part permits to capture interaction rules emerging from specification examples, the operational part supports the automatic execution in the operation of the virtual environment. A system is presented which embeds this language within a visual shell to support a behavioral training in which the animation rules of virtual agents are defined through visual examples     

Building trustworthy software agents

As agents become more active and sophisticated, the implications of their actions become more serious. With today's GUIs, user and software errors can often be easily fixed or undone. An agent performing actions on behalf of a user could make errors that are very difficult to "undo", and, depending on the agent's complexity, it might not be clear what went wrong. Moreover, for agents to operate effectively and truly act on their users' behalf, they might need confidential or sensitive information. This includes financial details and personal contact information. Thus, along with the excitement about agents and what they can do, there is concern about the resulting security and privacy issues. It is not enough to assume that well-designed software agents will provide the security and privacy users need; assurances and assumptions about security and privacy need to be made explicit. This article proposes a model of the factors that determine agent acceptance, based on earlier work on user attitudes toward e-commerce transactions, in which feelings of trust and perceptions of risk combine in opposite directions to determine a user's final acceptance of an agent technology.     

PLANGENT: an approach to making mobile agents intelligent

Network environments give computer users the option of employing distributed information and services to complete a task. However, gathering information and using services distributed in networks requires knowing exactly what kinds of information and services are required for a task, where they are, and how they can be obtained or utilized. Tracking down the answers to these questions can be difficult, time consuming tasks for users. Mobile agent technology is expected to release them from having to do so. Instead, “intelligent” mobile agents will comprehend the user's requirements, search network nodes autonomously for appropriate information and services, and return with the answers. But several problems must be solved before we can expect agents to perform such actions effectively. We focus on the question of intelligence as a prerequisite for agent functions. What sort of intelligence is expected of agents? We have adopted a model based on the ability to make flexible plans. Specifically, we think mobile agents must be able to: understand user requirements; plan actions that will satisfy the requirements act according to the plan; modify the plan according to actual conditions when they differ from those initially expected; and execute the modified plan. We have implemented these functions in the Plangent system and validated their effectiveness in several example applications. We describe how we combined these planning functions with mobile agent facilities, and show how the agents behave intelligently in an example application of personal travel assistance     

A model driven engineering process of platform neutral agents for ambient intelligence devices

Ambient intelligence (AmI) systems are now considered a promising approach to assist people in their daily life. AmI proposes the development of context aware systems equipped with devices that can recognize your context and act accordingly. Agents provide an effective way to develop such systems since agents are reactive, proactive and exhibit an intelligent and autonomous behavior. However, current agent approaches do not adequately fulfill the requirements posed by AmI systems. From a modeling point of view, the aim should be to help in the design by providing adequate tools that assist in the development of important properties of AmI systems, such as context-awareness; and from an implementation point of view, agent technologies must be adapted to the diversity of AmI devices and communication technologies. As a solution to these issues we propose a Model driven engineering process, which supports the automatic generation of agent-based AmI systems. The source metamodel is PIM4Agents, a general purpose agent metamodel that we have adapted to support the explicit modeling of context aware systems, and the target metamodel is Malaca, an aspect-oriented agent architecture. Aspect-orientation makes Malaca platform-neutral for FIPA compliant agent platforms, simplifying the model driven process. The solution generates MalacaTiny agents, an implementation of Malaca that is able to run in AmI devices. We have evaluated the convenience of applying a model driven approach by measuring the degree of automation of our process and we have evaluated MalacaTiny for mobile phones by assessing different parameters, related to the scarcity of resources in AmI systems. All the results obtained are satisfactory.     

An adaptive social network for information access: Theoretical and experimental results

We consider a social network of software agents who assist each other in helping their users find information. Unlike in most previous approaches, our architecture is fully distributed and includes agents who preserve the privacy and autonomy of their users. These agents learn models of each other in terms of expertise (ability to produce correct domain answers) and sociability (ability to produce accurate referrals). We study our framework experimentally to study how the social network evolves. Specifically, we find that under our multi-agent learning heuristic, the quality of the network improves with interactions: the quality is maximized when both expertise and sociability are considered; pivot agents further improve the quality of the network and have a catalytic effect on its quality even if they are ultimately removed. Moreover, the quality of the network improves when clustering decreases, reflecting the intuition that you need to talk to people outside your close circle to get the best information.     

Coordination of intelligent agents in real-time search

Search is a fundamental problem-solving method in artificial intelligence. Traditional off-line search algorithms attempt to find an optimal solution whereas real-time search algorithms try to find a suboptimal solution more quickly than traditional algorithms to meet real-time constraints. In this work, a new multi-agent real-time search algorithm is developed and its effectiveness is illustrated on a sample domain, namely maze problems. Searching agents can see their environment with a specified visual depth and hence can partially observe their environment. An agent makes use of its partial observation to select a next move, instead of using only one-move-ahead information. Furthermore agents cooperate through a marking mechanism to be able to search different parts of the search space. When an agent selects its next move, it marks its direction of move before executing the move. When another agent comes to this position, it sees this mark and, if possible, moves in a different direction than the previously selected direction. In this way, marking helps agents coordinate their moves with other agents. Although coordination brings an overhead, from experiments we observe that this mechanism is effective in both search time and solution length in maze problems.     

The effects of animated pedagogical agents in an English-as-a-foreign-language learning environment

Animated pedagogical agents (APAs) have frequently been used as a powerful addition to learning environments, since APAs have been known to facilitate learning. APAs can present various features, such as voice, movements, gestures and pointing, and researchers have sought to verify specifically which features of agents effectively contribute to learning. Previous studies have studied these features by comparing different degrees of agent embodiment in the evaluation of the image effect (i.e., students learn more when learning systems have visual APAs), the embodied agent effect (i.e., fully embodied agents that deliver instruction aurally and use gestures to improve learning outcomes in text-only learning systems), the modality effect (i.e., oral instruction contributes to the learning process), and the expressiveness effect (i.e., fully embodied agents promote more effective learning than static ones). Some of these studies have investigated the image, embodied agent and modality effects in the same learning environment, but they were not the same studies that investigated the expressiveness effect. The expressiveness effect allows us to separate the movements of the agent from its other features, such as the agent's image, so investigating this effect is as important as investigating the other effects. We are not aware of any studies that investigated all of these four effects within the same learning system, nor that evaluated any of these effects in language learning environments. Accordingly, this paper describes the design, implementation, and analysis of an APA designed to evaluate the abovementioned effects. The APA was integrated into a computer-assisted language learning (CALL) system to teach English as a foreign language to Brazilian students. A total of 72 Brazilian undergraduate students were divided into four groups, each of which used a different version of the APA in the same CALL system: no agent, a voice-only agent, a static agent, or a fully embodied agent. We compared students’ gain scores (i.e., difference between pre- and posttest scores) across groups to evaluate each of the four effects. Though the outcomes of our study supported the presence of the embodied agent and modality effects, we were not able to demonstrate the image or expressiveness effects in the experiment. Our results indicate that the voice of the agent might contribute more positively to learning than movements, gestures and pointing.     

Interactive POMDPs with finite-state models of other agents

We consider an autonomous agent facing a stochastic, partially observable, multiagent environment. In order to compute an optimal plan, the agent must accurately predict the actions of the other agents, since they influence the state of the environment and ultimately the agent’s utility. To do so, we propose a special case of interactive partially observable Markov decision process, in which the agent does not explicitly model the other agents’ beliefs and preferences, and instead represents them as stochastic processes implemented by probabilistic deterministic finite state controllers (PDFCs). The agent maintains a probability distribution over the PDFC models of the other agents, and updates this belief using Bayesian inference. Since the number of nodes of these PDFCs is unknown and unbounded, the agent places a Bayesian nonparametric prior distribution over the infinitely dimensional set of PDFCs. This allows the size of the learned models to adapt to the complexity of the observed behavior. Deriving the posterior distribution is in this case too complex to be amenable to analytical computation; therefore, we provide a Markov chain Monte Carlo algorithm that approximates the posterior beliefs over the other agents’ PDFCs, given a sequence of (possibly imperfect) observations about their behavior. Experimental results show that the learned models converge behaviorally to the true ones. We consider two settings, one in which the agent first learns, then interacts with other agents, and one in which learning and planning are interleaved. We show that the agent’s performance increases as a result of learning in both situations. Moreover, we analyze the dynamics that ensue when two agents are simultaneously learning about each other while interacting, showing in an example environment that coordination emerges naturally from our approach. Furthermore, we demonstrate how an agent can exploit the learned models to perform indirect inference over the state of the environment via the modeled agent’s actions.     

Evolutionary asymmetric games for modeling systems of partially cooperative agents

Based on the tenet of Darwinism, we propose a general mechanism that guides agents (which can be partially cooperative) in selecting appropriate strategies in situations of complex interactions, in which agents do not have complete information about other agents. In the mechanism, each participating agent generates many instances of itself to help it find an appropriate strategy. The generated instances adopt alternative strategies from the agent's strategy set. While all instances generated by different agents meet randomly to complete a task, every instance adapts its strategy according to the difference between the average utilities of its current strategy and all its strategies. We give a complete analysis of the mechanism for the case with two agents when each agent has two strategies, and show that by the tenet of Darwinism, agents can find their appropriate strategies through evolution and adaptation: 1) if dominant strategies exist, then the proposed mechanism is guaranteed to find them; 2) if there are two or more strict Nash equilibrium strategies, the proposed mechanism is guaranteed to find them by using different initial strategy distributions; and 3) if there is no dominant strategy and no strict Nash equilibrium, then agents will oscillate periodically. Nevertheless, the mechanism allows agent designers to derive the appropriate strategies from the oscillation by integration. For cases with two agents when each agent has two or more strategies, it is shown that agents can reach a steady state where social welfare is optimum.     

A distributed multi-agent meeting scheduler

Electronic calendars are important tools that are used by consumers on a daily basis. However, scheduling a meeting that involves persons with different commitments and preferences remains a difficult task. Meeting scheduling is difficult because current calendaring applications cannot handle the responsibility of automatically and autonomously managing time slots. This paper presents a distributed multi-agent system architecture in which each person is represented by an agent. These agents automatically and autonomously work together to assist different users to book meetings on their behalf. Each agent has the capability to manage, negotiate and schedule tasks, meetings, events, appointments for its assigned user. In this multi-agent system, the agents coordinate their activities and negotiate on behalf of their associated users to find a solution that satisfies the users' meeting requirements and preferences. A prototype of this system is implemented to demonstrate how the agents can automatically book meetings.     

Networking embedded agents

Most of us will soon be managing an intranet in our homes, though we might not realize it. We might also be surprised at the devices that will be networked together. Just about every electrical device now contains one or more microprocessors. Designers typically find this a cost-effective way to provide device functionality, even when much of a processor's power is unnecessary or unused. For example, my coffee maker contains a processor, even though the appliance needn't be very smart and wastes most of its CPU cycles. Nevertheless, it is cheaper to include a general-purpose microprocessor than to incorporate custom logic devices. My kitchen, in fact, has at least six processors, in such appliances as the microwave, the dishwasher, and the toaster. These household devices are diverse and use their processors in quite different ways, but in the future they will share one important characteristic: Each will contain an agent. The agent will provide an intelligent interface to the device and, most importantly, will communicate with other devices in my home. At present, my devices are not very agent-like, and it is not useful to think, “My toaster knows when the toast is done” or “My coffee pot knows when the coffee is ready.” However, once the devices are interconnected so that they can communicate, they can arrange to have my coffee and toast ready at approximately the same time. Then I may think of them in anthropomorphic terms. For example, when I shut off my alarm clock, I can imagine it telling my kitchen devices to prepare my breakfast. When devices talk to each other, they begin to seem more like agents. At: this point my house becomes more than just a collection of processors-it becomes a multiagent system communicating over an intranet     

Communication between agents with heterogeneous perceptual capabilities

In real world applications robots and software agents often have to be equipped with higher level cognitive functions that enable them to reason, act and perceive in changing, incompletely known and unpredictable environments. One of the major tasks in such circumstances is to fuse information from various data sources. There are many levels of information fusion, ranging from the fusing of low level sensor signals to the fusing of high level, complex knowledge structures. In a dynamically changing environment even a single agent may have varying abilities to perceive its environment which are dependent on particular conditions. The situation becomes even more complex when different agents have different perceptual capabilities and need to communicate with each other.In this paper, we propose a framework that provides agents with the ability to fuse both low and high level approximate knowledge in the context of dynamically changing environments while taking account of heterogeneous and contextually limited perceptual capabilities.To model limitations on an agent’s perceptual capabilities we introduce the idea of partial tolerance spaces. We assume that each agent has one or more approximate databases where approximate relations are represented using lower and upper approximations on sets. Approximate relations are generalizations of rough sets.It is shown how sensory and other limitations can be taken into account when constructing and querying approximate databases for each respective agent. Complex relations inherit the approximativeness of primitive relations used in their definitions. Agents then query these databases and receive answers through the filters of their perceptual limitations as represented by (partial) tolerance spaces and approximate queries. The techniques used are all tractable.     

The impact of available information on negotiation results

Negotiation is the most famous tool for reaching an agreement between parties. Usually, the different parties can be modeled as a buyer and a seller, who negotiate about the price of a given item. In most cases, the parties have incomplete information about one another, but they can invest money and efforts in order to acquire information about each other. This leads to the question of how much each party will be willing to invest on information about its opponent, prior to the negotiation process. In this paper, we consider the profitability of automated negotiators acquiring information on their opponents. In our model, a buyer and a seller negotiate on the price of a given item. Time is costly, and incomplete information exists about the reservation price of both parties. The reservation price of the buyer is the maximum price it is willing to pay for an item or service, and the reservation price of the seller is the minimum price it is willing to receive in order to sell the item or service. Our research is based on Cramton’s symmetrical protocol of negotiation that provides the agents with stable and symmetric strategies, and involves a delay in proposing an offer for signaling. The parties in Cramton’s model delay their offers in order to signal their strength, and then an agreement is reached after one or two offers. We determine the Nash equilibrium for agents that prefer to purchase information. Then, in addition to the theoretical background, we used simulations to check which type of equilibrium will actually be obtained. We found that in most of the cases, each agent will prefer to purchase information only if its opponent does. The reason for these results lies in the fact that an agent that prefers to purchase information according to a one-side method, signals its weakness and thereby reduces its position in the negotiation. Our results demonstrate the efficiency of joint information acquisition by both agents, but they also show that one-sided information purchasing may be inefficient, if the acquisition activity is revealed by the opponent, which causes it to infer that the informed agent is relatively weak.     

Market-aware agents for a multiagent world

A computational market is any collection of software agents interacting through a price system. Markets can provide effective allocation of resources for a variety of distributed environments, and economic analysis is a powerful design tool for interaction mechanisms. The spread of computational markets puts a premium on market-aware agents, and presents a case for market awareness on the part of agent developers and AI researchers as well.     

Evaluating spoken dialogue agents with PARADISE: Two case studies

This paper presents PARADISE (PARAdigm for DIalogue System Evaluation), a general framework for evaluating and comparing the performance of spoken dialogue agents. The framework decouples task requirements from an agent's dialogue behaviours, supports comparisons among dialogue strategies, enables the calculation of performance over subdialogues and whole dialogues, specifies the relative contribution of various factors to performance, and makes it possible to compare agents performing different taks by normalizing for task complexity. After presenting PARADISE, we illustrate its application to two different spoken dialogue agents. We show how to derive a performance function for each agent and how to generalize results across agents. We then show that once such a performance function has been derived, it can be used both for making predictions about future versions of an agent, and as feedback to the agent so that the agent can learn to optimize its behaviour based on its experiences with users over time.     

Aborting,suspending, and resuming goals and plans in BDI agents

Intelligent agents designed to work in complex, dynamic environments such as e-commerce must respond robustly and flexibly to environmental and circumstantial changes, including the actions of other agents. An agent must have the capability to deliberate about appropriate courses of action, which may include reprioritising tasks—whether goals or associated plans—aborting or suspending tasks, or scheduling tasks in a particular order. In this article we study mechanisms to enable principled suspend, resuming, and aborting of goals and plans within a Belief-Desire-Intention (BDI) agent architecture. We give a formal and combined operational semantics for these actions in an abstract agent language (CAN), thus providing a general mechanism that can be incorporated into several BDI-based agent platforms. The abilities enabled by our semantics provides an agent designer greater flexibility to direct agent operation, offering a generic means to manage the status of goals. We demonstrate the reasoning abilities enabled on a document workflow scenario.     

Designing dependable agent systems for mobile wireless networks

A mobile ad hoc network (MANET) is a wireless network of mobile devices-such as PDAs, laptops, cell phones, and other lightweight, easily transportable computing devices-in which each node can act as a router for network traffic rather than relying on fixed networking infrastructure. As mobile computing becomes ubiquitous, MANETS becomes increasingly important. As a design paradigm, multiagent systems (MASs) can help facilitate and coordinate ad hoc-scenarios that might include security personnel, rescue workers, police officers, firefighters, and paramedics. On this network, mobile agents perform critical functions that include delivering messages, monitoring resource usage on constrained mobile devices, assessing network traffic patterns, analyzing host behaviors, and revoking access rights for suspicious hosts and agents. Agents can effectively operate in such environments if they are environment aware - if they can sense and reason about their complex and dynamic environments. Altogether, agents living on a MANET must be network, information, and performance aware. This article fleshes out how we apply this approach to populations of mobile agents on a live MANET.