AI Agents Research Articles

Optimized Financial Planning: Integrating Individual and Cooperative Budgeting Models with LLM Recommendations

In today’s complex economic environment, individuals and households alike grapple with the challenge of financial planning. This paper introduces novel methodologies for both individual and cooperative (household) financial budgeting. We firstly propose an optimization framework for individual budget allocation, aiming to maximize savings by efficiently distributing monthly income among various expense categories. We then extend this model to households, wherein the complexity of handling multiple incomes and shared expenses is addressed. The cooperative model prioritizes not only maximized savings but also the preferences and needs of each member, fostering a harmonious financial environment, whether they are short-term needs or long-term aspirations. A notable innovation in our approach is the integration of recommendations from a large language model (LLM). Given its vast training data and potent inferential capabilities, the LLM provides initial feasible solutions to our optimization problems, acting as a guiding beacon for individuals and households unfamiliar with the nuances of financial planning. Our preliminary results indicate that the LLM-recommended solutions result in budget plans that are both economically sound, meaning that they are consistent with established financial management principles and promote fiscal resilience and stability, and aligned with the financial goals and preferences of the concerned parties. This integration of AI-driven recommendations with econometric models, as an instantiation of an extended coevolutionary (EC) theory, paves the way for a new era in financial planning, making it more accessible and effective for a wider audience, as we propose an example of a new theory in economics where human behavior can be greatly influenced by AI agents.

Mapping Our Digital Menagerie: A Monster Manual for the Megadungeon

Relying on a comparison between the complex spatial organization of our current digital ecosystem and the ones from dungeons in role-playing games, this article analyzes the multiple entities that populate our computers, smartphones and video game consoles. Trolls, bugs, worms, conversational AI agents, NPCs, daemons, ghosts, Trojan horses and others are presented and discussed with a focus on their different degree of agency, ranging from human-controlled to having a certain degree of autonomy. By addressing how we coexist with such beings, this paper contributes to the nascent field of digital folklore.

Finding the sweet spot: Exploring the optimal communication delay for AI feedback tools

AI writing assistants that are capable of offering rapid, automatic, and corrective feedback are becoming increasingly powerful, providing accurate and automatic feedback to knowledge contributors while they write. This helps people stay engaged and focused on their writing tasks. However, even with improved algorithms, AI still requires communication delays to perform sophisticated language processing and generate accurate responses. Prior research suggests that if the delay is too long, users may lose interest and become frustrated, perceiving the AI agent as ineffective. Conversely, if the AI-generated feedback is too quick, users may push back and perceive the AI as not fully comprehending their input, leading to low levels of trust. The impact of feedback timing on knowledge contribution behavior remains largely unknown. To address this gap, this research employs two sequential studies to explore the effects of feedback timing. The first study is a conjoint analysis, and the second study is a random-assignment-based experiment. Our results show that an optimal range of timing for communication delay is approximately one to three seconds, leading to improved performance. Presentation styles of the AI agent may affect these results but only slightly. Our study is the first to systematically investigate the optimality of communication delay on AI feedback mechanisms. Our findings will contribute to the literature on knowledge management, algorithm design, and anthropomorphism, and also inspire the software designers on the optimal timing of the communication feedback.

Elastic Monte Carlo Tree Search

Strategy games are a challenge for the design of AI agents due to their complexity and the combinatorial search space they produce. State abstraction has been applied in different domains to shrink the search space. Automatic state abstraction methods have gained much success in the planning domain and their transfer to strategy games raises a question of scalability. In this paper, we propose Elastic MCTS, an algorithm that uses automatic state abstraction to play strategy games. In Elastic MCTS, tree nodes are clustered dynamically. First, nodes are grouped by state abstraction for efficient exploration, to later be separated for refining exploitable action sequences. Such an elastic tree benefits from efficient information sharing while avoiding using an imperfect state abstraction during the whole search process. We provide empirical analyses of the proposed method in three strategy games of different complexity. Our empirical results show that in all games, Elastic MCTS outperforms MCTS baselines by a large margin, with a considerable search tree size reduction at the expense of small computation time. The code for reproducing the reported results can be found at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/GAIGResearch/Stratega</uri> .

Analysis and Requirement Generation for Defense Intelligence Search: Addressing Data Overload through Human–AI Agent System Design for Ambient Awareness

This research addresses the data overload faced by intelligence searchers in government and defense agencies. The study leverages methods from the Cognitive Systems Engineering (CSE) literature to generate insights into the intelligence search work domain. These insights are applied to a supporting concept and requirements for designing and evaluating a human-AI agent team specifically for intelligence search tasks. Domain analysis reveals the dynamic nature of the ‘value structure’, a term that describes the evolving set of criteria governing the intelligence search process. Additionally, domain insight provides details for search aggregation and conceptual spaces from which the value structure could be efficiently applied for intelligence search. Support system designs that leverage these findings may enable an intelligence searcher to interact with and understand data at more abstract levels to improve task efficiency. Additionally, new system designs can support the searcher by facilitating an ‘Ambient Awareness’ of non-selected objects in a large data field through relevant system cues. Ambient Awareness achieved through the supporting concept and AI teaming has the potential to address the data overload problem while increasing the breadth and depth of search coverage.

Individuality and the collective in AI agents: Explorations of shared consciousness and digital homunculi in the metaverse for cultural heritage

&lt;p class="a"&gt;The confluence of extended reality (XR) technologies, including augmented and virtual reality, with large language models (LLM) marks a significant advancement in the field of digital humanities, opening uncharted avenues for the representation of cultural heritage within the burgeoning metaverse. This paper undertakes an examination of the potentialities and intricacies of such a convergence, focusing particularly on the creation of digital homunculi or changelings. These virtual beings, remarkable for their sentience and individuality, are also part of a collective consciousness, a notion explored through a thematic comparison in science fiction with the Borg and the Changelings in the Star Trek universe. Such a comparison offers a metaphorical framework for discussing complex phenomena such as shared consciousness and individuality, illuminating their bearing on perceptions of self and awareness. Further, the paper considers the ethical implications of these concepts, including potential loss of individuality and the challenges inherent to accurate representation of historical figures and cultures. The latter necessitates collaboration with cultural experts, underscoring the intersectionality of technological innovation and cultural sensitivity. Ultimately, this chapter contributes to a deeper understanding of the technical aspects of integrating large language models with immersive technologies and situates these developments within a nuanced cultural and ethical discourse. By offering a comprehensive overview and proposing clear recommendations, the paper lays the groundwork for future research and development in the application of these technologies within the unique context of cultural heritage representation in the metaverse.&lt;/p&gt;

Personality is to a Conversational Agent What Perfume is to a Flower

Although there has been a growing interest for conversational AI (CAI) agents in the consumer electronics (CE) community, how these agents are personified by humans has been an unexplored area. Creating personality traits for CAI agents is important as it will enable providing more personalized services. Accordingly, we propose a multi-dimensional approach involving two distinct phases and three different experiments for generating the CAI personality by integrating concepts from human, brand, and website personality frameworks. We identify eight unique personality types, three of which are proposed for the first time. Accordingly, the research implications are discussed.

Generative conversational AI agent for managerial practices: The role of IQ dimensions, novelty seeking and ethical concerns

This study aims to identify and empirically examine the influence of the main factors related to the content quality of generative conversational AI agents on decision-making efficiency. Additionally, this study explores the ramifications of decision-making efficiency facilitated by generative conversational AI agents in organisational innovation performance. This study proposes a model based on the information quality model as well as other factors, such as novelty seeking and ethical concerns. Data from this study was collected using online questionnaires from a purposive sample size of 228 employees in business organisations. Based on Structural Equation Modelling (SEM) analyses using AMOS, the results support the significant impact of information quality (intrinsic information quality, contextual information quality, representational information quality, and accessibility of information quality) on decision-making efficiency. The results also support the significant impact of novelty seeking and ethical concerns on decision-making efficiency. Decision-making efficiency was also found to have a significant positive impact on innovation performance. This empirical study makes a considerable contribution as it is among the first to expand the current understanding of the effective use of generative conversational AI agents in managerial practices (i.e. decision-making and innovation).

The Past, the Present, and the Future of the Evolution of Mixed Reality in Teacher Education

The authors in this article provide a historical view (past) on the development of mixed reality (MR) simulation in teacher education as well as a brief history of simulation from other fields along with foundational knowledge on the evolution of simulation. The authors provide a systematic review of the current state (present) of the research in MR for teacher education within the past 5 years aligned with the research question “What are the uses, practices, and outcomes of MR simulation in teacher preparation?”. Three themes were identified, i.e., simulation to this point is designed by teacher educators, feedback matters in impacting outcomes, and practice is safe and reflective for those who prepare teachers in these environments. A summary is provided of these key articles and the findings. The authors conclude the article by sharing the potential evolution (future) of aspects of the model of MR, focusing on the use of AI agents and multi-modal data collection, including biometric signals, providing insights into simulation in teacher education.

Understanding Human-AI Teaming Dynamics through Gaming Environments

With the goal of better understanding Human-machine Teaming (HMT) dynamics and how team competencies that are transportable across contexts can lead to different teaming behaviors and team performances, I propose a series of three studies to explore communication, coordination and adaptation in HMT paradigms. I implement and integrate multiple AI agents and use collaborative games as testing environments to evaluate teaming effects. My work can provide findings to two higher level research questions that are widely studied in HMT: 1) the bidirectional behaviors that human and AI agents may develop when working as a team and, 2) how different types of AI agents can impact the teaming efficiency in human-AI teaming. Besides, my work can also contribute to Human-Computer Interaction and Game AI scholarship with insights into teaming dynamics in Human-AI teaming.

Mechanic Maker 2.0: Reinforcement Learning for Evaluating Generated Rules

Automated game design (AGD), the study of automatically generating game rules, has a long history in technical games research. AGD approaches generally rely on approximations of human play, either objective functions or AI agents. Despite this, the majority of these approximators are static, meaning they do not reflect human player's ability to learn and improve in a game. In this paper, we investigate the application of Reinforcement Learning (RL) as an approximator for human play for rule generation. We recreate the classic AGD environment Mechanic Maker in Unity as a new, open-source rule generation framework. Our results demonstrate that RL produces distinct sets of rules from an A* agent baseline, which may be more usable by humans.

Herd’s Eye View: Improving Game AI Agent Learning with Collaborative Perception

We present a novel perception model named Herd's Eye View (HEV) that adopts a global perspective derived from multiple agents to boost the decision-making capabilities of reinforcement learning (RL) agents in multi-agent environments, specifically in the context of game AI. The HEV approach utilizes cooperative perception to empower RL agents with a global reasoning ability, enhancing their decision-making. We demonstrate the effectiveness of the HEV within simulated game environments and highlight its superior performance compared to traditional ego-centric perception models. This work contributes to cooperative perception and multi-agent reinforcement learning by offering a more realistic and efficient perspective for global coordination and decision-making within game environments. Moreover, our approach promotes broader AI applications beyond gaming by addressing constraints faced by AI in other fields such as robotics. The code is available at https://github.com/andrewnash/Herds-Eye-View

Risk Management: Anticipating and Reacting in StarCraft

Managing risk with imperfect information is something humans do every day, but we have little insight into the abilities of AI agents to do so. We define two risk management strategies and perform an ability-based evaluation using StarCraft agents. Our evaluation shows that nearly all agents mitigate risks after observing them (react), and many prepare for such risks before their appearance (anticipate). For this evaluation, we apply traditional causal effect inference and causal random forest methods to explain agent behavior. The results highlight different risk management strategies among agents, others strategies that are common to agents, and overall encourage evaluating agent risk management abilities in other AI domains.

Reward tampering and evolutionary computation: a study of concrete AI-safety problems using evolutionary algorithms

Reward tampering is a problem that will impact the trustworthiness of the powerful AI systems of the future. Reward Tampering describes the problem where AI agents bypass their intended objective, enabling unintended and potentially harmful behaviours. This paper investigates whether the creative potential of evolutionary algorithms could help ensure trustworthy solutions when facing this problem. The reason why evolutionary algorithms may help combat reward tampering is that they are able to find a diverse collection of different solutions to a problem within a single run, aiding the search for desirable solutions. Four different evolutionary algorithms were deployed in tasks illustrating the problem of reward tampering. The algorithms were designed with varying degrees of human expertise, measuring how human guidance influences the ability to discover trustworthy solutions. The results indicate that the algorithms’ ability to find and preserve trustworthy solutions is very dependent on preserving diversity during the search. Algorithms searching for behavioural diversity showed to be the most effective against reward tampering. Human expertise also showed to improve the certainty and quality of safe solutions, but even with only a minimal degree of human expertise, domain-independent diversity management was found to discover safe solutions.

Desirable or Distasteful? Exploring Uncertainty in Human-Chatbot Relationships

Present-day power users of AI-powered social chatbots encounter various uncertainties and concerns when forming relationships with these virtual agents. To provide a systematic analysis of users’ concerns and to complement the current West-dominated approach to chatbot studies, we conducted a thorough observation of the experienced uncertainties users reported in a Chinese online community on social chatbots. The results revealed four typical uncertainties: technical uncertainty, relational uncertainty, ontological uncertainty, and sexual uncertainty. We further conducted visibility and sentiment analysis to capture users’ response patterns toward various uncertainties. We discovered that users’ identification of social chatbots is dynamic and contextual. Our study contributes to expanding, summarizing, and elucidating users’ experienced uncertainties and concerns as they form intimate relationships with AI agents.

Co-Designing with Users the Explanations for a Proactive Auto-Response Messaging Agent

Explanations of AI Agents' actions are considered to be an important factor in improving users' trust in the decisions made by autonomous AI systems. However, as these autonomous systems evolve from reactive, i.e., acting on user input, to proactive, i.e., acting without requiring user intervention, there is a need to explore how the explanation for the actions of these agents should evolve. In this work, we explore the design of explanations through participatory design methods for a proactive auto-response messaging agent that can reduce perceived obligations and social pressure to respond quickly to incoming messages by providing unavailability-related context. We recruited 14 participants who worked in pairs during collaborative design sessions where they reasoned about the agent's design and actions. We qualitatively analyzed the data collected through these sessions and found that participants' reasoning about agent actions led them to speculate heavily on its design. These speculations significantly influenced participants' desire for explanations and the controls they sought to inform the agents' behavior. Our findings indicate a need to transform users' speculations into accurate mental models of agent design. Further, since the agent acts as a mediator in human-human communication, it is also necessary to account for social norms in its explanation design. Finally, user expertise in understanding their habits and behaviors allows the agent to learn from the user their preferences when justifying its actions.

Trust in Artificial Intelligent Agent while Completing a Procedural Construction Task

The use of AI-enabled recommender systems in construction activities has the potential to improve worker performance and reduce errors; however, the accuracy of such systems in providing effective suggestions is dependent on the quality of their training data. A within-subjects experimental study was conducted using a simulated recommender system for installation tasks to investigate the effect of system reliability and construction task complexity on worker trust, workload, and performance. Results indicate that overall trust in the AI agent was higher for the highly reliable condition but remained consistent across various levels of task complexity. The workload was found to be higher for low reliability and complex conditions, and the effect of reliability on performance was influenced by task complexity. These findings offer insights for designing recommender systems to support construction workers in completing procedural tasks.

Balancing the Scales of Explainable and Transparent AI Agents within Human-Agent Teams

With the progressive nature of Human-Agent Teams becoming more and more useful for high-quality work output, there is a proportional need for bi-directional communication between teammates to increase efficient collaboration. This need is centered around the well-known issue of innate mistrust between humans and artificial intelligence, resulting in sub-optimal work. To combat this, computer scientists and humancomputer interaction researchers alike have presented and refined specific solutions to this issue through different methods of AI interpretability. These different methods include explicit AI explanations as well as implicit manipulations of the AI interface, otherwise known as AI transparency. Individually these solutions hold considerable merit in repairing the relationship of trust between teammates, but also have individual flaws. We posit that the combination of different interpretable mechanisms mitigates each other’s flaws and extenuates their strengths within human-agent teams.

Human-AI teams in complex military operations: Soldiers’ perception of intelligent AI agents as teammates in human-AI teams

Military decision-making frequently involves complex problems in non-routine situations with minimal rule-based or automated solutions. As human information processing capabilities are limited, processing the required scale of information increases the decision-makers’ workload, impacting their situational awareness and affecting the quality of soldiers’ decisions. This study uses a route clearance task as a use-case scenario to understand the issues in team-level decision-making in military tasks, the challenges in successfully completing the mission, and the demands placed on AI teammates when working in a human-AI team. We interviewed eight subject matter experts with prior experience in route clearance operations. The themes generated by analyzing the interview transcripts provide insights into soldiers’ perceptions of AI teammates as well as recommendations for successfully integrating human-AI teams in military settings.

Designing Personalized Pedagogical AI Agents to Support Children's Exploratory Learning

Can a social robot support children to become better learners? How personalizing a social robot's behavior can encourage learner exploration.