What is new with Artificial Intelligence? Human-agent interactions through the lens of social agency

The introduction of automated systems, and more broadly of Artificial Intelligence (AI), into many domains has profoundly changed the nature of human activity, as well as the subjective experience that agents have of their own actions and their consequences – an experience that is commonly referred to as "sense of agency" (SoA). In this review, we propose to examine the empirical evidence supporting this impact of automation on individuals’ sense of agency, and hence on measures as diverse as operator performance, system explicability and acceptability. Because of some of its key characteristics, AI occupies a special status in the artificial systems landscape. We suggest that this status prompts us to reconsider human-AI interactions in the light of human-human relations. We therefore approach the study of joint actions in human social interactions to deduce what are the key features necessary to develop a reliable SoA in a social context. We suggest that the study of social interactions and the development of SoA in joint actions can help determine the content of relevant explanations to be implemented in AI to make it "explainable". Finally, we propose possible directions to improve human-AI interactions and, in particular, to restore the SoA of human operators, improve their confidence in the decisions made by artificial agents, and increase the acceptability of such agents.

Gibsonian Information: an agent-based paradigm for quantitative information

We propose a new way to quantitatively characterize information: Gibsonian Information (GI). This framework is relevant to both the study of cognitive agents and single cell systems that exhibit cognitive behaviors. GI provides a means to characterize how agents extract information from direct perceptual signals. This differs from existing information theories in two ways. The first involves an emphasis on sensory processing, engagement in collective behaviors, and the dynamic evolution of such interactions. GI is useful for understanding information in terms of agent interactions with naturalistic contexts, and also allows us to distinguish the role of first-order sensory inputs in the context of higher-order representations. This allows us to extend GI to cybernetic and other types of symbolic systems representations. As an alternate way to look at information in nervous systems, GI also emphasizes the role of information content in the relationship between ecology and nervous systems. Along with direct sensory input and simple internal representations, statistical affordances are an important component of transforming environmental signals into GI. Statistical affordances, defined as clustered information that is spatiotemporally dependent perceptual input, facilitates the extraction of GI from the environment. Quantitatively accounting for perceptual information, GI provides a means to measure spatial concentration in addition to being a generalized indicator of nervous system input. To better understand GI as a measurable phenomenon, we characterize three model scenarios: disjoint distributions, contingent action, and coherent movement. All of these cases provide a means to create a differential system between both motion (information) and random noise/stasis (non-information) and between the active sensory channel and information derived from mental representations. By applying this framework to a variety of specific contexts, including a four-channel model of multisensory embodiment, we demonstrate how GI is essential to understanding the full scope of cognitive information processing.

A Comprehensive Modeling of the Discrete and Dynamic Problem of Berth Allocation in Maritime Terminals

In this study, the discrete and dynamic problem of berth allocation in maritime terminals, is investigated. The suggested resolution method relies on a paradigm of optimization with two techniques: heuristic and multi-agent. Indeed, a set of techniques such as the protocol of negotiation named contract net, the multi-agent interactions, and Worst-Fit arrangement technique, are involved. The main objective of the study is to propose a solution for attributing m parallel machines to a set of activities. The contribution of the study is to provide a detailed modeling of the discrete and dynamic berth allocation problem by establishing the corresponding models using a multi-agent methodology. A set of numerical experiments are detailed to prove the performance of the introduced multi-agent strategy compared with genetic algorithm and tabu search.

Enabling Game-Theoretical Analysis of Social Rules

In the field of normative multiagent systems, the relationship between a game structure and its underpinning agent interaction rules is hardly ever addressed in a systematic manner. In this work, we introduce the Action Situation Language (ASL), inspired by Elinor Ostrom’s Institutional Analysis and Development framework, to bridge the gap between games and rules. The ASL provides a syntax for the description of agent interactions, and is complemented by an engine that automatically provides semantics for them as extensive-form games. The resulting games can then be analysed using standard game-theoretical solution concepts, hence allowing any community of agents to automatically perform what-if analysis of potential new interaction rules.

Towards Applying Interactive POMDPs to Real-World Adversary Modeling

We examine the suitability of using decision processes to model real-world systems of intelligent adversaries. Decision processes have long been used to study cooperative multiagent interactions, but their practical applicability to adversarial problems has received minimal study. We address the pros and cons of applying sequential decision-making in this area, using the crime of money laundering as a specific example. Motivated by case studies, we abstract out a model of the money laundering process, using the framework of interactive partially observable Markov decision processes (I-POMDPs). We address why this framework is well suited for modeling adversarial interactions. Particle filtering and value iteration are used to solve the model, with the application of different pruning and look-ahead strategies to assess the tradeoffs between solution quality and algorithmic run time. Our results show that there is a large gap in the level of realism that can currently be achieved by such decision models, largely due to computational demands that limit the size of problems that can be solved. While these results represent solutions to a simplified model of money laundering, they illustrate nonetheless the kinds of agent interactions that cannot be captured by standard approaches such as anomaly detection. This implies that I-POMDP methods may be valuable in the future, when algorithmic capabilities have further evolved.

Little Computer People

Bolstered by a growing interest in simulating believable non-player characters (NPCs), work on NPC models has spanned topics such as planning, procedural storytelling, decision-making, and social dynamics. However, research groups work in isolation, designing and discussing their character models with disparate approaches, often using project-specific terminology. This makes it challenging to identify, classify, and accumulate existing knowledge. It is our position that since modelling of virtual characters has become an integral part of the scientific practice in our field, we must develop a common taxonomy to discuss these models. With this goal in mind, we conduct an in-depth analysis of a selection of projects, categorizing existing agent social interactions, and comparing results from research-based and commercial social simulation works in the entertainment domain. We conceptualize a taxonomy that classifies agent interactions by their social behaviours, inter-agent communication, knowledge flow, and the change in their relationships. We posit such a taxonomy would allow scientists to reproduce and evaluate existing models, collaborate on standards, share advances with other researchers and practitioners, allow for better communication and methodologies developed for new techniques, and allow for a more rigorous model-to-model analysis.

Investigating the Impact of Visual Environmental Uncertainty on Human-Agent Teaming

Uncertainty in Human-Agent interactions is often studied in terms of transparency and understandability of agent actions. Less work, however, has focused on how Visual Environmental Uncertainty (VEU) that restricts or occludes visual information affects the Human-Agent Teaming (HAT) in terms of trust, reliance, performance, cognitive load and situational awareness. We conducted a mixed-design experiment (n=96) where participants interacted with an agent during a collaborative aiming task under four different types of VEUs involving global and dynamic occlusions. Our results show that while environmental uncertainties led to increases in perceived trust, they induced differences in reliance and performance. Counter to intuition, when participants trusted the agent the most, they relied on the agent more, but performed worst. These findings highlight how trust in agents is also influenced by external environmental conditions and suggest that reported trust in HAT scenarios may not always generalize beyond the environmental factors in which they were studied.

Learning in Markets: Greed Leads to Chaos but Following the Price is Right

We study learning dynamics in distributed production economies such as blockchain mining, peer-to-peer file sharing and crowdsourcing. These economies can be modelled as multi-product Cournot competitions or all-pay auctions (Tullock contests) when individual firms have market power, or as Fisher markets with quasi-linear utilities when every firm has negligible influence on market outcomes. In the former case, we provide a formal proof that Gradient Ascent (GA) can be Li-Yorke chaotic for a step size as small as Θ(1/n), where n is the number of firms. In stark contrast, for the Fisher market case, we derive a Proportional Response (PR) protocol that converges to market equilibrium. The positive results on the convergence of the PR dynamics are obtained in full generality, in the sense that they hold for Fisher markets with any quasi-linear utility functions. Conversely, the chaos results for the GA dynamics are established even in the simplest possible setting of two firms and one good, and they hold for a wide range of price functions with different demand elasticities. Our findings suggest that by considering multi-agent interactions from a market rather than a game-theoretic perspective, we can formally derive natural learning protocols which are stable and converge to effective outcomes rather than being chaotic.

Identifying Norms from Observation Using MCMC Sampling

To promote efficient interactions in dynamic and multi-agent systems, there is much interest in techniques that allow agents to represent and reason about social norms that govern agent interactions. Much of this work assumes that norms are provided to agents, but some work has investigated how agents can identify the norms present in a society through observation and experience. However, the norm-identification techniques proposed in the literature often depend on a very specific and domain-specific representation of norms, or require that the possible norms can be enumerated in advance. This paper investigates the problem of identifying norm candidates from a normative language expressed as a probabilistic context-free grammar, using Markov Chain Monte Carlo (MCMC) search. We apply our technique to a simulated robot manipulator task and show that it allows effective identification of norms from observation.