https://ojs.aaai.org/index.php/AAAI-SS <p>The AAAI Symposium Series, previously published as AAAI Technical Reports, are held three times a year (Spring, Summer, Fall) and are designed to bring colleagues together to share ideas and learn from each other’s artificial intelligence research. The series affords participants a smaller, more intimate setting where they can share ideas and learn from each other’s artificial intelligence research. Topics for the symposia change each year, and the limited seating capacity and relaxed atmosphere allow for workshop-like interaction. The format of the series allows participants to devote considerably more time to feedback and discussion than typical one-day workshops. It is an ideal venue for bringing together new communities in emerging fields.<br /><br />The AAAI Spring Symposium Series is typically held during spring break (generally in March) on the west coast. The AAAI Summer Symposium Series is the newest in the annual set of meetings run in parallel at a common site. The inaugural 2023 Summer Symposium Series was held July 17-19, 2023, in Singapore. The AAAI Fall Symposium series is usually held on the east coast during late October or early November.</p> en-US publications@aaai.org (Publications Manager) publications@aaai.org (Publications Manager) Mon, 22 Jan 2024 16:34:54 -0800 OJS 3.2.1.1 http://blogs.law.harvard.edu/tech/rss 60 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27641 Multi-agent path finding deals with finding collision free paths for a group of agents moving to given destinations. The off-line generated plan is assumed to be blindly executed on robots, which brings issues when something is not going according to the plan. This short paper discusses robustness as a way to prevent the issues with uncertainty, dynamicity, and possible involvement of other (uncontrolled) agents. Roman Barták Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27641 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27642 We present the effect of adapting to human preferences on trust in a human-robot teaming task. The team performs a task in which the robot acts as an action recommender to the human. It is assumed that the behavior of the human and the robot is based on some reward function they try to optimize. We use a new human trust-behavior model that enables the robot to learn and adapt to the human's preferences in real-time during their interaction using Bayesian Inverse Reinforcement Learning. We present three strategies for the robot to interact with a human: a non-learner strategy, in which the robot assumes that the human's reward function is the same as the robot's, a non-adaptive learner strategy that learns the human's reward function for performance estimation, but still optimizes its own reward function, and an adaptive-learner strategy that learns the human's reward function for performance estimation and also optimizes this learned reward function. Results show that adapting to the human's reward function results in the highest trust in the robot. Shreyas Bhat, Joseph B. Lyons, Cong Shi, X. Jessie Yang Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27642 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27643 Autonomous vessels have emerged as a prominent and accepted solution, particularly in the naval defence sector. However, achieving full autonomy for marine vessels demands the development of robust and reliable control and guidance systems that can handle various encounters with manned and unmanned vessels while operating effectively under diverse weather and sea conditions. A significant challenge in this pursuit is ensuring the autonomous vessels' compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). These regulations present a formidable hurdle for the human-level understanding by an autonomous systems as they were originally designed from common navigation practices created since the mid-19th century. Their ambiguous language assumes experienced sailors' interpretation and execution and, therefore, demands a high-level (cognitive) understanding of language and agent intentions that are beyond the capabilities of current state-of-the-art of intelligent system. This position paper highlights the critical requirement of a trustworthy control and guidance system and explores the complexities of adapting COLREGs for safe vessel-on-vessel encounters considering autonomous maritime technology competing and/or cooperating with manned vessels. Benoit Clement, Marie Dubromel, Paulo E. Santos, Karl Sammut, Michelle Oppert, Feras Dayoub Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27643 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27644 Defining and measuring trust in dynamic, multiagent teams is important in a range of contexts, particularly in defense and security domains. Team members should be trusted to work towards agreed goals and in accordance with shared values. In this paper, our concern is with the definition of goals and values such that it is possible to define ‘trust’ in a way that is interpretable, and hence usable, by both humans and robots. We argue that the outcome of team activity can be considered in terms of ‘goal’, 'individual/team values', and ‘legal principles’. We question whether alignment is possible at the level of 'individual/team values', or only at the 'goal' and ‘legal principles’ levels. We argue for a set of metrics to define trust in human-robot teams that are interpretable by human or robot team members, and consider an experiment that could demonstrate the notion of 'satisficing trust' over the course of a simulated mission. Edmund R. Hunt, Chris Baber, Mehdi Sobhani, Sanja Milivojevic, Sagir Yusuf, Mirco Musolesi, Patrick Waterson, Sally Maynard Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27644 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27645 When humans cooperate, they frequently coordinate their activity through both verbal communication and non-verbal actions, using this information to infer a shared goal and plan. How can we model this inferential ability? In this paper, we introduce a model of a cooperative team where one agent, the principal, may communicate natural language instructions about their shared plan to another agent, the assistant, using GPT-3 as a likelihood function for instruction utterances. We then show how a third person observer can infer the team’s goal via multi-modal Bayesian inverse planning from actions and instructions, computing the posterior distribution over goals under the assumption that agents will act and communicate rationally to achieve them. We evaluate this approach by comparing it with human goal inferences in a multi-agent gridworld, finding that our model’s inferences closely correlate with human judgments (R = 0.96). When compared to inference from actions alone, we find that instructions lead to more rapid and less uncertain goal inference, highlighting the importance of verbal communication for cooperative agents. Lance Ying, Tan Zhi-Xuan, Vikash Mansinghka, Joshua B. Tenenbaum Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27645 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27646 Rapid and continuous increase in greenhouse gas (GHG) emissions is warming our planet at unprecedented rates. Consumer products and services, including all aspects of the corresponding supply chain, contribute to more than 75% of these emissions. Attribution of GHG emissions to each product will drive awareness and change from individual consumers to large corporations that produce and own these products. However, accurate and standards-compliant accounting of carbon emissions for millions of products is challenging as it requires detailed manufacturing and supply chain data, and subject expertise in life cycle assessment (LCA). We posit that ideas from computer science and machine learning can alleviate bottlenecks in LCA, and that research contributions from this community will accelerate solutions for accurate carbon-footprint estimation as well as carbon-abatement strategies at scale. We present the principal components of an LCA study with a step-by-step walk-through. We elaborate upon the challenges to scale LCA, and identify the opportunities to innovate in this space with techniques such as information extraction, personalized recommendations, and decision-making under uncertainty. Bharathan Balaji, Geoffrey Guest, Venkata Sai Gargeya Vunnava, Jared Kramer, Aravind Srinivasan, Michael Taptich Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27646 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27647 In times of climate change and large machine learning models with petabytes of training data, the demand for responsible AI methodologies is more pressing than ever. This is in-particular true for the vast amount of remote sensing data. Its value to explore and inform about earth processes is of paramount importance, especially in times of global warming. Thus, it is nearly ironic that such applications can be the cause of substantial green-house-gas emissions, through energy demands from compute and communication systems. Thus, this study aims at reducing the data transfer between data centers while maintaining near-real time insights from remote sensing data. A recursive inference approach is introduced consisting of three steps: i) Data pyramid preparation in the host data center (sequence of upscaled raster data). ii) The transfer of low-resolution images to the service data center, where a deep-learning model performs a semantic segmentation task, including an uncertainty estimation. Images of higher resolution are then requested and segmented in a recursive fashion, in areas of high uncertainty only. iii) Finally, the merging of the predictions at different resolutions is performed to result in the final pixel-wise segmentation at scale. The method is demonstrated on synthetic and real-world data for a flood mapping task. A U-Net encoder-decoder model is used for the semantic segmentation task, using Monte-Carlo dropout to result in the uncertainty map. The proof-of-concept demonstrated a 35-38% performance gain per transferred pixel compared to high-resolution image segmentation only. Further, we perform a scaling study to estimate the true potential of the recursive inference approach, indicating the potential to reduce a data transfer by up to 98%, considering four hierarchy levels in the data pyramid. With this study, we hope to have contributed a small but important step towards sustainable machine learning. Thomas Brunschwiler, Tobia Claglüna, Michal Muszynski, Tobias Hölzer, Paolo Fraccaro, Maciel Zortea, Jonas Weiss Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27647 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27648 The convergence of Artificial Intelligence (AI) with climate science is a double-edged sword. AI-enhanced modeling has transformative potential for the field, but it comes with new vulnerabilities, especially from adversarial machine learning. Such adversarial tactics can distort AI-driven climate models, producing misleading projections on phenomena like sea-level changes and temperature predictions. Beyond just mod-eling, AI-enhanced systems in resource management, conserva-tion, and agriculture are at risk. Tampering with climate da-tasets can undermine decades of global research and erode trust, while adversarial misinformation campaigns can skew public discourse. Ethically, distorted data risks magnifying socio-environmental disparities. Addressing these challenges necessitates robust modeling using advanced techniques, data defense with cryptographic solutions, AI-driven infrastructure safeguards, and AI algorithms to detect and counter misinfor-mation. Simply put, securing AI in climate science is not just a technical challenge, but a global imperative. Sally Calengor, Sai Prathyush Katragadda, Joshua Steier Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27648 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27649 The electrification of transportation is a growing strategy to reduce mobile source emissions and air pollution globally. To encourage adoption of electric vehicles, there is a need for reliable evidence about pricing in pub-lic charging stations that can serve a greater number of communities. However, user-entered pricing information by thousands of charge point operators (CPOs) has created ambiguity for large-scale aggregation, increasing both the cost of analysis for researchers and search costs for consumers. In this paper, we use large language models to address standing challenges with price discovery in distributed digital data. We show that generative AI models can effectively extract pricing mechanisms from unstructured text with high accuracy, and at substantially lower cost of three to four orders of magnitude lower than human curation (USD 0.006 pennies per observation). We exploit the few-shot learning capabilities of GPT-4 with human-in-the-loop feedback—beating prior classification performance benchmarks with fewer training data. The most common pricing models include free, energy-based (per kWh), and time-based (per unit time), with tiered pricing (variable pricing based on usage) being the most prevalent among paid stations. Behavioral insights from a US nationally representative sample of 13,008 stations suggest that EV users are commonly frustrated with the slower than expected charging rates and the total cost of charging. This study uncovers additional consumer barriers to charging services concerning the need for better price standardization. Sarthak Chaturvedi, Edward W. Chen, Ila P. Sharma, Omar Isaac Asensio Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27649 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27650 Global Climate Models (GCMs) are the primary tool to simulate climate evolution and assess the impacts of climate change. However, they often operate at a coarse spatial resolution that limits their accuracy in reproducing local-scale phenomena. Statistical downscaling methods leveraging deep learning offer a solution to this problem by approximating local-scale climate fields from coarse variables, thus enabling regional GCM projections. Typically, climate fields of different variables of interest are downscaled independently, resulting in violations of fundamental physical properties across interconnected variables. This study investigates the scope of this problem and, through an application on temperature, lays the foundation for a framework introducing multi-variable hard constraints that guarantees physical relationships between groups of downscaled climate variables. Jose González-Abad, Álex Hernández-García, Paula Harder, David Rolnick, José Manuel Gutiérrez Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27650 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27651 Given the heightened global awareness and attention to the negative externalities of plastics use, many state and local governments are considering legislation that will limit single-use plastics for consumers and retailers under extended producer responsibility laws. Considering the growing momentum of these single-use plastics regulations globally, there is a need for reliable and cost-effective measures of the public response to this rulemaking for inference and prediction. Automated computational approaches such as generative AI could enable real-time discovery of consumer preferences for regulations but have yet to see broad adoption in this domain due to concerns about evaluation costs and reliability across large-scale social data. In this study, we leveraged the zero and few-shot learning capabilities of GPT-4 to classify public sentiment towards regulations with increasing complexity in expert prompting. With a zero-shot approach, we achieved a 92% F1 score (s.d. 1%) and 91% accuracy (s.d. 1%), which resulted in three orders of magnitude lower research evaluation cost at 0.138 pennies per observation. We then use this model to analyze 5,132 tweets related to the policy process of the California SB-54 bill, which mandates user fees and limits plastic packaging. The policy study reveals a 12.4% increase in opposing public sentiment immediately after the bill was enacted with no significant changes earlier in the policy process. These findings shed light on the dynamics of public engagement with lower cost models for research evaluation. We find that public opposition to single-use plastics regulations becomes evident in social data only when a bill is effectively enacted. Catharina Hollauer, Jorge Garcelán, Nikhita Ragam, Tia Vaish, Omar Isaac Asensio Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27651 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27652 Disaster response agencies have been shifting from a paradigm of climate forecasting towards one of anticipatory action: assessing not just what the climate will be, but how it will impact specific populations, thereby enabling proactive response and resource allocation. Machine learning models are becoming exceptionally powerful at climate forecasting, but methodological gaps remain in terms of facilitating anticipatory action. Here we provide an overview of anticipatory action, review relevant applications of machine learning, identify common challenges, and highlight areas where machine learning can uniquely contribute to advancing disaster response for populations most vulnerable to climate change. Benjamin Q. Huynh, Mathew V. Kiang Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27652 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27653 The melting of ice sheets significantly contributes to sea level rise, highlighting the crucial need to comprehend the structure of ice for climate benefits. The stratigraphy of ice sheets revealed through ice layer radargrams gives us a window into historical depth-age correlations and accumulation rates. Harnessing this knowledge is not only crucial for interpreting both past and present ice dynamics, especially concerning the Greenland ice sheet, but also for making informed decisions to mitigate the impacts of climate change. Ice layer tracing is prevalently conducted using manual or semi-automatic approaches, requiring significant time and expertise. This study aims to address the need for efficient and precise tracing methods in a two-step process. This is achieved by utilizing an unsupervised annotation method (i.e., ARESELP) to train deep learning models, thereby reducing the need for extensive and time-consuming manual annotations. Four prominent deep learning-based segmentation techniques, namely U-Net, U-Net+VGG19, U-Net+Inception, and Attention U-Net, are benchmarked. Additionally, various thresholding methods such as binary, Otsu, and CLAHE have been explored to achieve optimal enhancement for the true label annotation images. Our preliminary experiments indicate that the combination of attention U-Net with specific processing techniques yields the best performance in terms of the binary IoU metric. Atefeh Jebeli, Bayu Adhi Tama, Sanjay Purushotham, Vandana P. Janeja Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27653 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27654 Machine learning methods are increasingly adopted in communications problems, particularly those arising in next generation wireless settings. Though seen as a key climate mitigation and societal adaptation enabler, communications related energy consumption is high and is expected to grow in future networks in spite of anticipated efficiency gains in 6G due to exponential communications traffic growth. To make meaningful climate mitigation impact in the communications sector, a mindset shift away from maximizing throughput at all cost and towards prioritizing energy efficiency is needed. Moreover, this must be adopted in both existing (without incurring further embodied carbon costs through equipment replacement) and future network infrastructure, given the long development time of mobile generations. To that end, we present summaries of two such problems, from both current and next generation network specifications, where probabilistic inference methods were used to great effect: using Bayesian parameter tuning we are able to safely reduce the energy consumption of existing hardware on a live communications network by 11% whilst maintaining operator specified performance envelopes; through spatiotemporal Gaussian process surrogate modeling we reduce the overhead in a next generation hybrid beamforming system by over 60%, greatly improving the networks' ability to target highly mobile users such as autonomous vehicles. The Bayesian paradigm is itself helpful in terms of energy usage, since training a Bayesian optimization model can require much less computation than, say, training a deep neural network. A. Ryo Koblitz, Lorenzo Maggi, Matthew Andrews Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27654 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27655 Flood events have the potential to impact every aspect of life, economic loss and casualties can quickly be coupled with damages to agricultural land, infrastructure, and water quality. Creating flood susceptibility maps is an effective manner that equips communities with valuable information to help them prepare for and cope with the impacts of potential floods. Flood indexing and forecasting are nonetheless complex because multiple external parameters influence flooding. Accordingly, this study explores the potential of utilizing artificial intelligence (AI) techniques, including clustering and neural networks, to develop a flooding susceptibility index (namely, NeuralFlood) that considers multiple factors that are not generally considered otherwise. By comparing four different sub-indices, we aim to create a comprehensive index that captures unique characteristics not found in existing methods. The use of clustering algorithms, model tuning, and multiple neural layers produced insightful outcomes for county-level data. Overall, the four sub-indices’ models yielded accurate results for lower classes (accuracy of 0.87), but higher classes had reduced true positive rates (overall average accuracy of 0.68 for all classes). Our findings aid decision-makers in effectively allocating resources and identifying high-risk areas for mitigation. Justice Lin, Chhayly Sreng, Emma Oare, Feras A. Batarseh Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27655 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27656 Estimating the product carbon footprint (PCF) is crucial for sustainable consumption and supply chain decar-bonlization. The current life cycle assessment (LCA) methods frequently employed to evaluate PCFs often en-counter challenges, such as difficulties in determining the emission inventory and emission factors (EFs), as well as significant labor and time costs. To address these limitations, this paper presents AutoPCF, a novel auto-matic PCF estimation framework to conduct cradle-to-gate LCA for products. It utilizes deep learning models and large language models (LLMs) to automate and en-hance the estimation process. The framework comprises five stages: Emission Inventory Determination (EID), Activity Data Collection (ADC), Emission Factor Matching (EFM), Carbon Emission Estimation (CEE), and Estimation Verification and Evaluation (EVE). EID generates production processes and activity inventory, while ADC collects comprehensive activity data and EFM identifies accurate EFs. Emissions are then estimat-ed using the collected activity data and corresponding EFs. Experimental evaluations on steel, textile, and bat-tery products demonstrate the effectiveness of AutoPCF in improving the efficiency of PCF estimation. By auto-mating data collection and analysis, AutoPCF reduces re-liance on subjective decision-making and enhances the consistency and efficiency of carbon footprint assess-ments, advancing sustainable practices and supporting climate change mitigation efforts. Biao Luo, Jinjie Liu, Zhu Deng, Can Yuan, Qingrun Yang, Lei Xiao, Yucong Xie, Fanke Zhou, Wenwen Zhou, Zhu Liu Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27656 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27657 Climate change and its impact on global sustainability are critical challenges, demanding innovative solutions that combine cutting-edge technologies and scientific insights. Quantum machine learning (QML) has emerged as a promising paradigm that harnesses the power of quantum computing to address complex problems in various domains including climate change and sustainability. In this work, we survey existing literature that applies quantum machine learning to solve climate change and sustainability-related problems. We review promising QML methodologies that have the potential to accelerate decarbonization including energy systems, climate data forecasting, climate monitoring, and hazardous events predictions. We discuss the challenges and current limitations of quantum machine learning approaches and provide an overview of potential opportunities and future work to leverage QML-based methods in the important area of climate change research. Amal Nammouchi, Andreas Kassler, Andreas Theocharis Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27657 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27658 We explore the implementation of deep learning techniques for precise building damage assessment in the context of natural hazards, utilizing remote sensing data. The xBD dataset, comprising diverse disaster events from across the globe, serves as the primary focus, facilitating the evaluation of deep learning models. We tackle the challenges of generalization to novel disasters and regions while accounting for the influence of low-quality and noisy labels inherent in natural hazard data. Furthermore, our investigation quantitatively establishes that the minimum satellite imagery resolution essential for effective building damage detection is 3 meters and below 1 meter for classification using symmetric and asymmetric resolution perturbation analyses. To achieve robust and accurate evaluations of building damage detection and classification, we evaluated different deep learning models with residual, squeeze and excitation, and dual path network backbones, as well as ensemble techniques. Overall, the U-Net Siamese network ensemble with F-1 score of 0.812 performed the best against the xView2 challenge benchmark. Additionally, we evaluate a Universal model trained on all hazards against a flood expert model and investigate generalization gaps across events, and out of distribution from field data in the Ahr Valley. Our research findings showcase the potential and limitations of advanced AI solutions in enhancing the impact assessment of climate change-induced extreme weather events, such as floods and hurricanes. These insights have implications for disaster impact assessment in the face of escalating climate challenges. Maximilian Nitsche, S. Karthik Mukkavilli, Niklas Kühl, Thomas Brunschwiler Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27658 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27659 While carbon accounting plays a fundamental role in our fight against climate change, it is not without its challenges. We begin the paper with a critique of the conventional carbon accounting practices, after which we proceed to introduce the E-liability carbon accounting methodology and Emissions Liability Management (ELM) originally proposed by Kaplan and Ramanna, highlighting their strengths. Recognizing the immense value of this novel approach for real-world carbon accounting improvement, we introduce a novel data-driven integrative framework that leverages AI and computation, the E-Liability Knowledge Graph framework, to achieve real-world implementation of the E-liability carbon accounting methodology. In addition to providing a path-to-implementation, our proposed framework brings clarity to the complex environmental interactions within supply chains, thus enabling better informed and more responsible decision-making. We analyze the implementation aspects of this framework and conclude with a discourse on the role of this AI-aided knowledge graph in ensuring the transparency and decarbonization of global supply chains. Olamide Oladeji, Seyed Shahabeddin Mousavi, Marc Roston Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27659 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27660 Carbon accounting is a fundamental building block in our global path to emissions reduction and decarbonization, yet many challenges exist in achieving reliable and trusted carbon accounting measures. We motivate that carbon accounting not only needs to be more data-driven, but also more methodologically sound. We discuss the need for alternative, more diverse data sources that can play a significant role on our path to trusted carbon accounting procedures and elaborate on not only why, but how Artificial Intelligence (AI) in general and Natural Language Processing (NLP) in particular can unlock reasonable access to a treasure trove of alternative data sets in light of the recent advances in the field that better enable the utilization of unstructured data in this process. We present a case study of the recent developments on real-world data via an NLP-powered analysis using OpenAI's GPT API on financial and shipping data. We conclude the paper with a discussion on how these methods and approaches can be integrated into a broader framework for AI-enabled integrative carbon accounting. Olamide Oladeji, Seyed Shahabeddin Mousavi Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27660 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27661 Evolving hurricane patterns intensified by climate change are expected to exacerbate economic hardships on coastal communities. Climate resilience for these communities requires both the capability to recover rapidly from devastating storms, and the ability to develop an accurate and actionable understanding of vulnerabilities to reduce the impact of future storms. Available data from past storms can provide invaluable insight in addressing both these requirements. Post-disaster preliminary damage assessments (PDAs) are a crucial initial step toward a rapid recovery. They also provide the most accurate information on the performance of various types of dwellings after the storm. Traditional door-to-door inspection methods are time-consuming and can hinder efficient resource allocation by governments in the aftermath. To address this, researchers have proposed automated PDA frameworks, often utilizing data from satellites, combined with deep convolutional neural networks. However, before such frameworks can be adopted in practice, the accuracy and fidelity of predictions of damage level at the scale of an entire building must be comparable to human assessments. To bridge this gap, we present an innovative PDA framework that leverages Ultra-High-Resolution Aerial (UHRA) images alongside state-of-the-art transformer models for multi-class damage predictions across entire buildings. Our approach leverages vast amounts of unlabeled data to enhance accuracy of prediction and generalization capabilities. Through a series of experiments, we evaluate the influence of incorporating unlabeled data, and transformer models. By integrating UHRA images and semi-supervised transformer models, our findings indicate that this framework overcomes critical limitations associated with satellite imagery and traditional CNN models, achieving an 88% multiclass accuracy, ultimately leading to more precise, efficient, and reliable damage assessments that are a first step towards building more climate resilient societies. Deepank Kumar Singh, Vedhus Hoskere Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27661 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27662 Better forecasting of atmospheric composition is a critical aspect of environmental and climate monitoring. Among climate and weather numeric modeling, often ensembles are used to improve the forecasting power and to quantify the uncertainty of the model. However, the numerical simulation of atmospheric chemistry, critical for composition simulations, is computationally too expensive to generate numerical composition ensembles. One way to address this problem is to use deep learning to emulate the slow physical model. In this work we study the feasibility of two different deep learning methods and show how an emulator could be used to realistically estimate uncertainties of atmospheric composition forecasts, bypassing the need to run costly numerical ensemble simulations. One of the methods builds upon Fourier neural operators and the NVIDIA FourCastNet architecture and the second method builds on conditional Generative Adversarial Networks. We design the models to respond to perturbations to the most important drivers of air pollution, including meteorology and pollutant emissions. We apply this framework to the NASA GEOS Composition Forecast System (GEOS-CF), which produces daily global composition forecasts at approximately 25 km^2 horizontal resolution. Due to computational constraints, GEOS-CF currently has limited capability to produce probabilistic estimates or to optimally assimilate trace gas observations. We show how a deep learning emulator has the potential to improve composition forecasts produced by GEOS-CF or other, similar types of applications. These methods could be applied to other types of ensemble-based models, potentially providing a large speed-up in overall modeling time. Jennifer Sleeman, Christoph A. Keller, Christopher Ribaudo, David Chung, Mimi Szeto Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27662 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27663 In the context of urgent climate challenges and the pressing need for rapid technology development, Reinforcement Learning (RL) stands as a compelling data-driven method for controlling real-world physical systems. However, RL implementation often entails time-consuming and computationally intensive data collection and training processes, rendering them inefficient for real-time applications that lack non-real-time models. To address these limitations, real-time emulation techniques have emerged as valuable tools for the lab-scale rapid prototyping of intricate energy systems. While emulated systems offer a bridge between simulation and reality, they too face constraints, hindering comprehensive characterization, testing, and development. In this research, we construct a surrogate model using limited data from simulated systems, enabling an efficient and effective training process for a Double Deep Q-Network (DDQN) agent for future deployment. Our approach is illustrated through a hydropower application, demonstrating the practical impact of our approach on climate-related technology development. Alexander Stevenson, Mayank Panwar, Rob Hovsapian, Arif Sarwat Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27663 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27664 Satellite imagery can detect a wealth of ship tracks, temporary cloud trails created via cloud seeding by the emitted aerosols of large ships, a phenomenon that cannot be directly reproduced by global climate models. Ship tracks are satellite-observable examples of aerosol-cloud interactions, processes that constitute the largest uncertainty in climate forcing predictions, and when observed are also examples of Marine Cloud Brightening (MCB), a potential climate intervention strategy. Leveraging the large amount of observed ship track data to enhance understanding of aerosol-cloud interactions and the potentials of MCB is hindered by the computational infeasiblity of characterization from expensive physical models. In this paper, we focus on utilizing a cheaper physics-informed advection-diffusion surrogate to accurately emulate ship track behavior. As an indication of aerosol-cloud interaction behavior, we focus on learning the spreading behavior of ship tracks, neatly encoded in the emulator's spatio-temporal diffusion field. We train a convolutional LSTM to accurately learn the spreading behavior of simulated and satellite-masked ship tracks and discuss its potential in larger scale studies. Pierce Warburton, Kurtis Shuler, Lekha Patel Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27664 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27665 Task assignment and scheduling algorithms are powerful tools for autonomously coordinating large teams of robotic or AI agents. However, the decisions these system make often rely on components designed by domain experts, which can be difficult for non-technical end-users to understand or modify to their own ends. In this paper we propose a preliminary design for a flexible natural language interface for a task assignment system. The goal of our approach is both to grant users more control over a task assignment system's decision process, as well as render these decisions more transparent. Users can direct the task assignment system via natural language commands, which are applied as constraints to a mixed-integer linear program (MILP) using a large language model (LLM). Additionally, our proposed system can alert users to potential issues with their commands, and engage them in a corrective dialogue in order to find a viable solution. We conclude with a description of our planned user-evaluation in the simulated environment Overcooked and describe next steps towards developing a flexible and transparent task allocation system. Jake Brawer, Kayleigh Bishop, Bradley Hayes, Alessandro Roncone Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27665 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27666 We enable an autonomous agent sharing its artificial mind to its audiences like humans. This supports the autonomous human robot interactions relying on a cognitive architecture, LIDA, which explains and predicts how minds work and is used as the controllers of intelligent autonomous agents. We argue that LIDA’s cognitive representations and processes may serve as the source of the mind content its agent shares out, autonomously. We proposed a new description (sub) model into LIDA, letting its agent describing its conscious contents. Through this description, the agent’s mind is more observable so we can understand the agent’s entity and intelligence more directly. Also, this helps the agent explains its behaviors to its audiences so engage into its living society better. We built an initial LIDA agent embedding with this description model. The agent shares its conscious content autonomously, reasonably explaining its behaviors. Daqi Dong Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27666 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27667 Autonomous robots have the potential to supplement human capabilities while reducing cognitive and physical burden. However, deploying such systems in natural settings is currently a time-consuming process that revolves around a human’s ability to research, design, test, and evaluate the robot – thereby introducing unnecessary bottlenecks and significant delays to technology adoption. The current work in the field of human-robot interaction (HRI) has historically focused on robot use even though humans play a critical role during autonomous system design and deployment. We argue that the scope of HRI must be expanded beyond that of the current views within the scientific community, to include all phases of system development, deployment, and use. Furthermore, to facilitate the pursuit of this new expanded scope, we present eight opportunities for technological advances in HRI and autonomy using Generative AI that, if realized, could have transformational impact on the fielding of human-supervised autonomous robots. Broadly speaking, our identified opportunities relate to interaction and trustworthiness, collaboration and cooperation, robot motion, robot perception, synthetic scenario generation, testing and evaluation, failure detection, and robot design. Jason M. Gregory, Satyandra K. Gupta Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27667 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27668 Although large language models such as ChatGPT and GPT-4 have achieved superb performances in various natural language processing tasks, their dialogue performance is sometimes not very clear because the evaluation is often done on the utterance level where the quality of an utterance given context is the evaluation target. Our objective in this work is to conduct human evaluations of GPT-3.5 and GPT-4 to perform MultiWOZ and persona-based chat tasks in order to verify their dialogue-level performance in task-oriented and non-task-oriented dialogue systems. Our findings show that GPT-4 performs comparably with a carefully created rule-based system and has a significantly superior performance to other systems, including those based on GPT-3.5, in persona-based chat. Shinya Iizuka, Shota Mochizuki, Atsumoto Ohashi, Sanae Yamashita, Ao Guo, Ryuichiro Higashinaka Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27668 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27669 The rapid increase in the number of persons with Alzheimer’s Disease or related dementia has led many researchers to develop supplemental care to assist caregivers. One such form of care comes in the form of a quadruped robot that can interact with its environment to provide additional care. However, before such technology is fully implemented, researchers must understand how aware the public is of such technology for dementia care and how they perceive it. In this study, we surveyed 16 adults, all but one of which have been affected by dementia either directly or indirectly. We asked them questions regarding their attitude towards technology in healthcare and the quadruped robot that was demoed for them. It was found that people positively accept these robotic forms of dementia care, even if they do not have a comprehensive understanding of them. Furthermore, regarding the quadruped robot, people do perceive it positively but are not as confident in its ability to provide adequate care. They also have reservations about using robots to care for persons with dementia, mostly because of the lack of a “human touch,” and are afraid that robots might replace human caregivers altogether. From these results, researchers must do their best to not only develop the technology to be as robust as possible but keep the public informed of their research to bridge the gap between this revolutionary technology and its end users. Tyler Morris, Mengjun Wang, Yan Li, Songyan Liu, Shuai Li, Xiaopeng Zhao Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27669 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27670 It is crucial that users are empowered to use the functionalities of a robot to creatively solve problems on the fly. A user who has access to a Reinforcement Learning (RL) based robot may want to use the robot's autonomy and their knowledge of its behavior to complete new tasks. One way is for the user to take control of some of the robot's action space through teleoperation while the RL policy simultaneously controls the rest. However, an out-of-the-box RL policy may not readily facilitate this. For example, a user's control may bring the robot into a failure state from the policy's perspective, causing it to act in a way the user is not familiar with, hindering the success of the user's desired task. In this work, we formalize this problem and present Imaginary Out-of-Distribution Actions, IODA, an initial algorithm for addressing that problem and empowering user's to leverage their expectation of a robot's behavior to accomplish new tasks. Isaac Sheidlower, Reuben Aronson, Elaine Short Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27670 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27671 This paper formulates the problems of natural language generation and parsing as particular instances of the classical planning problem. It assumes the existence of a Categorial Grammar lexicon from which the preconditions and effects of available actions are obtained. A declarative formalization of heuristics for action selection is used to guide the search for solutions. Heuristics for mapping formulas in the description logic DL-Lite (R,and) into English sentences and backwards, and examples of application to Human Robot Interaction (HRI) are presented to illustrate the effectiveness of the approach. Josefina Sierra-Santibáñez Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27671 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27672 The focus of this research is to demonstrate how a platform composed of systems of artificial cognitive agents and social robotics can interact, teach and learn with students and teachers, through a pedagogical practice and methodological integration of the psychological concepts of the Theory of Multiple Intelligences, the educational foundations of the Dialectic methodology and the relationship explored in the Man-Machine Symbiosis. The objective of this article is to present the cognitive methodological concept of the project developed, grounding it from the theoretical principles, to the selection criteria for the models presented in the current discussions, taking as lines of thought that seek to define the strategy for creating personalized learning storytelling, the techniques for building social robotics in education and the interactivity between visual feedback in the challenging context of technological education. Rodrigo F. Souza, Walter T. Lima Jr Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27672 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27673 Imagine a service robot developed in the United States (US) being deployed in a public space in Israel. Due to the cultural differences, the robot from a ``contact-averse'' culture (i.e., the US) might find it difficult to find its way when navigating the crowd, as people from a ``contact-tolerant'' culture (i.e., Israel) - where a subtle touch between strangers is not uncommon - will always move closer to the robot than it would expect; conversely, an ``Israeli'' robot may be found too aggressive in US social spaces. Currently, these cultural differences hinder the ability to plug-and-play social robots in different cultures due to the requirement of extensive extra engineering effort. This paper presents a comparison of the results from an existing study conducted in the US with the same study design that was deployed in Israel. This comparison shows the clear, identifiable criteria that a socially aware robot will need to consider when navigating a new culture. More generally, the results from this paper offer a first step to identifying the cultural differences in social robot navigation so we can structure solutions to be compatible with these cultures and with novel ones, with minimum adaptation. Xuesu Xiao, Ross Mead, Reuth Mirsky Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27673 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27674 Socially assistive robots (SARs) have the potential to revolutionize educational experiences by providing safe, non-judgmental, and emotionally supportive environments for children's social development. The success of SARs relies on the synergy of different modalities, such as speech, gestures, and gaze, to maximize interactive experiences. This paper presents an approach for generating SAR behaviors that extend an upper ontology. The ontology may enable flexibility and scalability for adaptive behavior generation by defining key assistive intents, turn-taking, and input properties. We compare the generated behaviors with hand-coded behaviors that are validated through an experiment with young children. The results demonstrate that the automated approach covers the majority of manually developed behaviors while allowing for significant adaptations to specific circumstances. The technical framework holds the potential for broader interoperability in other assistive domains and facilitates the generation of context-dependent and socially appropriate robot behaviors. Yuqi Yang, Allison Langer, Lauren Howard, Peter J. Marshall, Jason R. Wilson Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27674 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27675 Automated Vehicles (AVs), acting as social robots, hold potential benefits for society. Prior research highlights how AV explanations can enhance passenger trust by clarifying the vehicle's reasoning and actions. However, an underexplored area is the impact of voice gender in AV explanations on this trust dynamic. To bridge this gap, our study, inspired by the gender-role congruity and similarity attraction theories, investigates the impacts of AV voice gender on user trust. The anticipated findings from our research are poised to play a critical role in designing AV explanations that enhance trust, thereby advancing the human-AV interaction landscape. Qiaoning Zhang, X. Jessie Yang, Lionel P. Robert Jr. Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27675 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27676 We present a new framework for AI alignment called Contractual AI, and apply it to the setting of dialogue agents chatting with humans. This framework incorporates and builds on previous approaches to alignment, such as Constitutional AI. We propose that fully aligned systems may need both a "think fast" and a "think slow" systems for approximating complex human judgements. Fast thinking (System 1) is computationally cheap but rigid and brittle in novel situations, while slow thinking (System 2) is more expensive but more flexible and robust. System 1 makes judgements by asking whether a rule or principle is violated. System 2 does the explicit reasoning that produces the rules, explicitly tallying costs and benefits for all stakeholders. Rule-based systems like Constitutional AI correspond roughly to System 1. Here, we implement a prototype of System 2, and lay out a road-map for enabling the system to make more thorough and accurate considerations for all stakeholder groups, including those underrepresented in the training data (e.g. racial minorities). For initial testing, we guided the decision process through the steps of: 1) identifying all stakeholders, 2) listing their individual concerns, 3) soliciting the projected opinions of various experts, and 4) combining the expert opinions into a final moral judgement. The resulting text was less generic, more aware of complex stakeholder needs, and ultimately more actionable. Christopher J. Bates, Ritwik Bose, Reagan G. Keeney, Vera A. Kazakova Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27676 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27677 MetroStar introduces "LabelUp," a transformative auto-labeling AI solution, custom-built for US government applications. Utilizing advanced transformer technology, LabelUp revolutionizes data labeling, significantly en-hancing operational AI models' efficiency for critical de-fense mechanisms. This innovative system promises over 800% workload reduction, facilitating rapid, precise la-beling with its intuitive low-code interface, featuring so-phisticated in-context-learning models. Compared to tra-ditional methods, LabelUp demonstrates staggering time and cost savings, redefining industry benchmarks. The paper further elucidates risk mitigation strategies, ensur-ing robust security and accuracy. In its prototype stage, LabelUp has shown significant potential, forecasting a breakthrough in image/video labeling processes. The white paper culminates with an invitation for a detailed government review and discussion. Joseph Early, Eric Kelly, Jesse Scearce Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27677 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27678 Mainstream machine learning methods lack interpretability, explainability, incrementality, and data-economy. We propose using logic programming (LP) to rectify these problems. We discuss the FOLD family of rule-based machine learning algorithms that learn models from relational datasets as a set of default rules. These models are competitive with state-of-the-art machine learning systems in terms of accuracy and execution efficiency. We also motivate how logic programming can be useful for theory revision and explanation based learning. Gopal Gupta, Huaduo Wang, Kinjal Basu, Farahad Shakerin, Parth Padalkar, Elmer Salazar, Sarat Chandra Varanasi, Sopam Dasgupta Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27678 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27679 Artificial intelligence (AI) is revolutionizing many industries, while at the same time facing challenges to safe and reliable use such as vulnerability to adversarial attacks and data drift. Although many AI test and evaluation (T&E) tools exist, integrating them is difficult. Under a program funded by the Chief Digital and AI Office (CDAO), we are developing a library to simplify the AI T&E process by providing user- and developer-friendly interfaces for composing T&E workflows. We illustrate the effectiveness of this approach with an example that compares clean and perturbed accuracy of two models on a computer vision dataset. Lei Hamilton, Garrett Botkin, Olivia Brown, Justin Goodwin, Michael Yee, Vincent Mancuso, Sanjeev Mohindra Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27679 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27680 Artificial Intelligence (AI) for decision support and diagnosis in pathology could provide immense value to society, improving patient outcomes and alleviating workload demands on pathologists. However, this potential cannot be realized until sufficient methods for testing and evaluation of such AI systems are developed and adopted. We present a novel metric for evaluation of multi-class classification algorithms for pathology, Error Severity Index (ESI), to address the needs of pathologists and pathology lab managers in evaluating AI systems. David Jin, Ariel Kapusta, Patrick A. Minot, Niels H. Olson, Joseph H. Rosenthal, Jansen N. Seheult, Michelle Stram Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27680 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27681 While deep learning models have seen widespread success in controlled environments, there are still barriers to their adoption in open-world settings. One critical task for safe deployment is the detection of anomalous or out-of-distribution samples that may require human intervention. In this work, we present a novel loss function and recipe for training networks with improved density-based out-of-distribution sensitivity. We demonstrate the effectiveness of our method on CIFAR-10, notably reducing the false-positive rate of the relative Mahalanobis distance method on far-OOD tasks by over 50%. Connor Mclaughlin, Jason Matterer, Michael Yee Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27681 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27682 We present an innovative approach to auto-annotate Expert Defined Linguistic Features (EDLFs) as subsequences in audio time series to improve audio deepfake discernment. In our prior work, these linguistic features – namely pitch, pause, breath, consonant release bursts, and overall audio quality, labeled by experts on the entire audio signal – have been shown to improve detection of audio deepfakes with AI algorithms. We now expand our approach to pilot a way to auto annotate subsequences in the time series that correspond to each EDLF. We developed an ensemble of discords, i.e. anomalies in time series, detected using matrix profiles across multiple discord lengths to identify multiple types of EDLFs. Working closely with linguistic experts, we evaluated where discords overlapped with EDLFs in the audio signal data. Our ensemble method to detect discords across multiple discord lengths achieves much higher accuracy than using individual discord lengths to detect EDLFs. With this approach and domain validation we establish the feasibility of using time series subsequences to capture EDLFs to supplement annotation by domain experts, for improved audio deepfake detection. Kifekachukwu Nwosu, Chloe Evered, Zahra Khanjani, Noshaba Bhalli, Lavon Davis, Christine Mallinson, Vandana P. Janeja Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27682 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27683 Knowledge engineering is an important task for creating and maintaining a knowledge base for cognitive models. It involves acquiring, representing, and organizing knowledge in a form that computers can use to make decisions and solve problems. However, this process can be a bottleneck for designing and using cognitive models. Knowledge engineering is a time-consuming and resource-intensive task that requires subject matter experts to provide information about a domain. In addition, models can acquire knowledge but require significant mechanisms to structure that information in a structured format appropriate for general use. Given the knowledge engineering bottleneck, we propose a solution that relies on natural language processing to extract key entities, relationships, and attributes to automatically generate chunks encoded as triples or chunks from unstructured text. Once generated, the knowledge can be used to create or add to a knowledge base within cognitive architectures to reduce knowledge engineering and task-specific models. Goonmeet Bajaj, Kate Pearce, Sean Kennedy, Othalia Larue, Alexander Hough, Jayde King, Christopher Myers, Srinivasan Parthasarathy Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27683 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27684 Recent popularity of generative models brought research on a variety of applications. We take a more architectural point of view, where we discuss ways in which generative AI techniques and cognitive architectures can benefit each other for a more capable overall integrated system. We use a cognitive architecture, ICARUS, as the framework for our discussion, but most of the discussed points should carry over to other architectures as well. Dongkyu Choi Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27684 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27685 What might the integration of cognitive architectures and generative models mean for sociocultural representations within both systems? Beyond just integration, we see this question as paramount to understanding the potential wider impact of integrations between these two types of computational systems. Generative models, though an imperfect representation of the world and various con-texts, nonetheless may be useful as general world knowledge with careful considerations of sociocultural representations provided therein, including the represented sociocultural systems or, as we explain, genres of the Human. Thus, such an integration gives an opportunity to develop cognitive models that represent from the physiological/biological time scale to the social timescale and that more accurately represent the effects of ongoing sociocultural systems and structures on behavior. In addition, integrating these systems should prove useful to audit and test many generative models under more realistic cognitive uses and conditions. That is, we can ask what it means that people will likely be using knowledge from such models as knowledge for their own behavior and actions. We further discuss these perspectives and focus these perspectives using ongoing and potential work with (primarily) the ACT-R cognitive architecture. We also discuss issues with using generative models as a system for integration. Christopher L. Dancy, Deja Workman Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27685 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27686 Recent developments in generative models have demonstrated that with the right data set, techniques, computational infrastructure, and network architectures, it is possible to generate seemingly intelligent outputs, without explicitly reckoning with underlying cognitive processes. The ability to generate novel, plausible behaviour could be a boon to cognitive modellers. However, insights for cognition are limited, given that generative models' blackbox nature does not provide readily interpretable hypotheses about underlying cognitive mechanisms. On the other hand, cognitive architectures make very strong hypotheses about the nature of cognition, explicitly describing the subjects and processes of reasoning. Unfortunately, the formal framings of cognitive architectures can make it difficult to generate novel or creative outputs. We propose to show that cognitive architectures that rely on certain Vector Symbolic Algebras (VSAs) are, in fact, naturally understood as generative models. We discuss how memories of VSA representations of data form distributions, which are necessary for constructing distributions used in generative models. Finally, we discuss the strengths, challenges, and future directions for this line of work. P. Michael Furlong, Chris Eliasmith Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27686 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27687 We present an approach to building systems that emulate human-like intelligence. Our approach uses machine learning technology (including generative AI systems) to extract knowledge from pictures, text, etc., and represents it as (pre-defined) predicates. Next, we use the s(CASP) automated commonsense reasoning system to check the consistency of this extracted knowledge and reason over it in a manner very similar to how a human would do it. We have used our approach for building systems for visual question answering, task-specific chatbots that can ``understand" human dialogs and interactively talk to them, and autonomous driving systems that rely on commonsense reasoning. Essentially, our approach emulates how humans process knowledge where they use sensing and pattern recognition to gain knowledge (Kahneman's System 1 thinking, akin to using a machine learning model), and then use reasoning to draw conclusions, generate response, or take actions (Kahneman's System 2 thinking, akin to automated reasoning). Gopal Gupta, Yankai Zeng, Abhiraman Rajasekaran, Parth Padalkar, Keegan Kimbrell, Kinjal Basu, Farahad Shakerin, Elmer Salazar, Joaquín Arias Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27687 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27688 In this paper, we provide a review of the current efforts to develop LLM agents, which are autonomous agents that leverage large language models. We examine the memory management approaches used in these agents. One crucial aspect of these agents is their long-term memory, which is often implemented using vector databases. We describe how vector databases are utilized to store and retrieve information in LLM agents. Moreover we highlight open problems, such as the separation of different types of memories and the management of memory over the agent's lifetime. Lastly, we propose several topics for future research to address these challenges and further enhance the capabilities of LLM agents, including the use of metadata in procedural and semantic memory and the integration of external knowledge sources with vector databases. Kostas Hatalis, Despina Christou, Joshua Myers, Steven Jones, Keith Lambert, Adam Amos-Binks, Zohreh Dannenhauer, Dustin Dannenhauer Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27688 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27689 A particular fusion of generative models and cognitive architectures is discussed with the help of the Soar and Sigma cognitive architectures. After a brief introduction to cognitive architecture concepts and Large Language Models as exemplar generative AI models, one approach towards their fusion is discussed. This is then analyzed with a summary of potential benefits and extensions needed to existing cognitive architecture that is closest to the proposal. Himanshu Joshi, Volkan Ustun Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27689 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27690 Large language models (LLMs) provide capabilities far beyond sentence completion, including question answering, summarization, and natural-language inference. While many of these capabilities have potential application to cognitive systems, our research is exploiting language models as a source of task knowledge for cognitive agents, that is, agents realized via a cognitive architecture. We identify challenges and opportunities for using language models as an external knowledge source for cognitive systems and possible ways to improve the effectiveness of knowledge extraction by integrating extraction with cognitive architecture capabilities, highlighting with examples from our recent work in this area. James R. Kirk, Robert E. Wray, John E. Laird Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27690 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27691 Large Language Models (LLMs) have shown impressive performance on a wide variety of tasks. However, apparent limitations hinder their performance, especially on tasks that require multiple steps of reasoning or compositionality. Arguably, the primary sources of these limitations are the decoding strategy and how the models are trained. We propose, and provide a general description of, an architecture that combines LLMs and cognitive architectures, called Language Model based Cognitive Architecture (LMCA), to overcome these limitations. We draw an analogy between this architecture and "fast" and "slow" thinking in human cognition. Kobe Knowles, Michael Witbrock, Gillian Dobbie, Vithya Yogarajan Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27691 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27692 We explore the potential integration of Transformers, trained online in real-time using an agent's ongoing experiences, as a learning and memory component of a cognitive architecture such as Soar. We identify key challenges and potential capabilities enabled by such an integration. John E. Laird, Robert E. Wray, Steven Jones, James R. Kirk, Peter Lindes Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27692 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27693 Cognitive architectures and generative models are two very different approaches for developing general embodied intelligence. This paper investigates their initial motivation, implementation ways, and the complementary strengths and weaknesses, and targets to fuse them into a general embodied intelligence so as to leverage strengths and complement weaknesses. Firstly, with analyzing their different application scenarios and the difficulties in further research and development, the potential synergy and possible integration strategies are explored between them. Then, by combining the strengths of cognitive architectures, which model human-like cognitive processes, and generative models, which excel in generating novel content based on learned patterns, it achieves the goal of creating embodied agents with enhanced overall capabilities. Finally, a comprehensive framework demonstrating the integration of cognitive architectures, generative models, and other AI methods to achieve general embodied intelligence is presented accompanied by an illustrative example. Yanfei Liu, Yuzhou Liu, Chao Shen Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27693 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27694 An evolutionary perspective on embodiment puts maintenance of physiology within a functional envelope as the brain’s base goal, with all other goals as refinements. Thus, all goals have physiological perturbation for their motivation and allostatic recovery as their signal of fulfillment. From this account, two entailments emerge. First, an object’s properties are not intrinsic to the object but a situated function of the morphology of the object and the affordances required by the goal. Second, categories do not exist without reference to some goal; they are constructed at the time of perception by blending prior conceptual knowledge to create an understanding of the perception with respect to the goal. Our thesis is that generative large language model (LLM) architectures are part of the solution to creating artificial organic-like cognitive architectures, but that LLMs as currently trained are generative only at a surface-level of behavior rather than deeper levels of cognition and, furthermore, that generative architectures must be coupled with an embodied cognitive agent architecture, which suggests both the additional levels at which generativity must operate and capabilities that the combined architecture must support. Spencer K. Lynn, Bryan Loyall, James Niehaus Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27694 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27695 The integration of cognitive and neural AI paradigms is a promising direction for overcoming the limitations of current deep learning models, but how to effect this integration is an open question. We propose that the key to this challenge lies in addressing the question of grounding. We adopt a cognitive perspective on grounding, and identify five types of grounding that are relevant for AI systems. We discuss ways that grounding in both cognitive and neural AI systems can facilitate the integration of these two paradigms, illustrating with examples in the domains of computational creativity and education. Because grounding is not only a technical problem but also a social and ethical one, requiring the collaboration and participation of multiple stakeholders, prosecuting such a research program is both timely and challenging. Mary Lou Maher, Dan Ventura, Brian Magerko Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27695 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27696 Instance-Based Learning Theory (IBLT) suggests that humans learn to engage in dynamic decision making tasks through the accumulation of experiences, represented by the decision task features, the actions performed, and the utility of decision outcomes. This theory has been applied to the design of Instance-Based Learning (IBL) models of human behavior in a variety of contexts. One key feature of all IBL model applications is the method of accumulating instance-based memory and performing recognition-based retrieval. In simple tasks with few features, this knowledge representation and retrieval could hypothetically be done using all relevant information. However, these methods do not scale well to complex tasks when exhaustive enumeration of features is unfeasible. This requires cognitive modelers to design task-specific representations of state features, as well as similarity metrics, which can be time consuming and fail to generalize to related tasks. To address this issue, we leverage recent advancements in Artificial Neural Networks, specifically generative models (GMs), to learn representations of complex dynamic decision making tasks without relying on domain knowledge. We evaluate a range of GMs in their usefulness in forming representations that can be used by IBL models to predict human behavior in a complex decision making task. This work connects generative and cognitive models by using GMs to form representations and determine similarity. Tyler Malloy, Yinuo Du, Fei Fang, Cleotilde Gonzalez Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27696 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27697 A prominent method to model human learning is through experiential learning, where decisions are influenced by the outcomes observed in previous actions. The decisions-from-experience approach often excludes other forms of learning in humans, such as learning from descriptive information. In humans, descriptive information can enhance learning by providing a denser signal, achieved through understanding the relationship between intermediate decisions and their future outcomes, instead of relying solely on observed outcomes. To account for experiential and descriptive information, we propose the use of large language models (LLMs) to convert descriptive information into dense signals that can be used by computational models that learn from experience. Building on past work in cognitive modeling, we utilize task instructions and prompt an LLM to define and quantify the critical actions an agent must take to succeed in the task. In an initial experiment, we test this approach using an Instance-Based Learning cognitive model of experiential decisions in a gridworld task. We demonstrate how the LLM can be prompted to provide a series of actions and relative values given the task instructions, then show how these values can be used in place of sparse outcome signals to improve the model’s learning of the task significantly. Chase McDonald, Tyler Malloy, Thuy Ngoc Nguyen, Cleotilde Gonzalez Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27697 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27698 Incorporating dynamic realistic human behaviors in population-scale computational models has been challenging. While some efforts have leveraged behavioral theories from social science, validated theories specifically applicable to Agent-based modeling remain limited. Existing approaches lack a comprehensive framework to model the situated, adaptive nature of human cognition and choice. To address these challenges, this paper proposes a novel framework, Psychologically-Valid Generative Agents. These agents consist of a Cognitive Architecture that provides data-driven and cognitively-constrained decision-making functionality, and a Large Language Model that generates human-like linguistic data. In addition, our framework benefits from Stance Detection, a Natural Language Processing technique, that allows highly personalize initialization of the agents, based on real-world data, within Agent-based modeling simulations. This combination provides a flexible yet structured approach to endogenously represent how people perceive, deliberate, and respond to social or other types of complex decision-making dynamics. Previous work has demonstrated promising results by using a subset of the components of our proposed architecture. Our approach has the potential to exhibit highly-realistic human behavior and can be used across a variety of domains (e.g., public health, group dynamics, social and psychological sciences, and financial markets). Konstantinos Mitsopoulos, Ritwik Bose, Brodie Mather, Archna Bhatia, Kevin Gluck, Bonnie Dorr, Christian Lebiere, Peter Pirolli Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27698 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27699 For generative AIs to be trustworthy, establishing transparent common grounding with humans is essential. As a preparation toward human-model common grounding, this study examines the process of model-model common grounding. In this context, common ground is defined as a cognitive framework shared among agents in communication, enabling the connection of symbols exchanged between agents to the meanings inherent in each agent. This connection is facilitated by a shared cognitive framework among the agents involved. In this research, we focus on the tangram naming task (TNT) as a testbed to examine the common-ground-building process. Unlike previous models designed for this task, our approach employs generative AIs to visualize the internal processes of the model. In this task, the sender constructs a metaphorical image of an abstract figure within the model and generates a detailed description based on this image. The receiver interprets the generated description from the partner by constructing another image and reconstructing the original abstract figure. Preliminary results from the study show an improvement in task performance beyond the chance level, indicating the effect of the common cognitive framework implemented in the models. Additionally, we observed that incremental backpropagations leveraging successful communication cases for a component of the model led to a statistically significant increase in performance. These results provide valuable insights into the mechanisms of common grounding made by generative AIs, improving human communication with the evolving intelligent machines in our future society. Junya Morita, Tatsuya Yui, Takeru Amaya, Ryuichiro Higashinaka, Yugo Takeuchi Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27699 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27700 The goal of the Companion cognitive architecture is to understand how to create human-like software social organisms. Thus natural language capabilities, both for reading and conversation, are essential. Recently we have begun experimenting with large language models as a component in the Companion architecture. This paper summarizes a case study indicating why we are currently using BERT with our symbolic natural language understanding system. It also describes some additional ways we are contemplating using large language models with Companions. Constantine Nakos, Kenneth D. Forbus Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27700 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27701 High-level reasoning can be defined as the capability to generalize over knowledge acquired via experience, and to exhibit robust behavior in novel situations. Such form of reasoning is a basic skill in humans, who seamlessly use it in a broad spectrum of tasks, from language communication to decision making in complex situations. When it manifests itself in understanding and manipulating the everyday world of objects and their interactions, we talk about common sense or commonsense reasoning. State-of-the-art AI systems don’t possess such capability: for instance, Large Language Models have recently become popular by demonstrating remarkable fluency in conversing with humans, but they still make trivial mistakes when probed for commonsense competence; on a different level, performance degradation outside training data prevents self-driving vehicles to safely adapt to unseen scenarios, a serious and unsolved problem that limits the adoption of such technology. In this paper we propose to enable high-level reasoning in AI systems by integrating cognitive architectures with external neuro-symbolic components. We illustrate a hybrid framework centered on ACT-R, and we discuss the role of generative models in recent and future applications. Alessandro Oltramari Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27701 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27702 Over the last few years, large neural generative models, capable of synthesizing semantically rich passages of text or producing complex images, have recently emerged as a popular representation of what has come to be known as ``generative artificial intelligence'' (generative AI). Beyond opening the door to new opportunities as well as challenges for the domain of statistical machine learning, the rising popularity of generative AI brings with it interesting questions for Cognitive Science, which seeks to discover the nature of the processes that underpin minds and brains as well as to understand how such functionality might be acquired and instantianted in biological (or artificial) substrate. With this goal in mind, we argue that a promising research program lies in the crafting of cognitive architectures, a long-standing tradition of the field, cast fundamentally in terms of neuro-mimetic generative building blocks. Concretely, we discuss the COGnitive Neural GENerative system, such an architecture that casts the Common Model of Cognition in terms of Hebbian adaptation operating in service of optimizing a variational free energy. Alexander G. Ororbia, Mary Alexandria Kelly Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27702 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27703 The paper describes a system that uses large language model (LLM) technology to support automatic learning of new entries in an intelligent agent’s semantic lexicon. The process is bootstrapped by an existing non-toy lexicon and a natural language generator that converts formal, ontologically-grounded representations of meaning into natural language sentences. The learning method involves a sequence of LLM requests and includes an automatic quality control step. To date, this learning method has been applied to learning multiword expressions whose meanings are equivalent to those of transitive verbs in the agent’s lexicon. The experiment demonstrates the benefits of a hybrid learning architecture that integrates knowledge-based methods and resources with both traditional data analytics and LLMs. Sanjay Oruganti, Sergei Nirenburg, Jesse English, Marjorie McShane Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27703 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27704 With the rapidly increasing use of AI and machine learning in recent years and the current generative AI revolution, it is no surprise that the malicious use of AI has begun to establish it- self in the realm of cybersecurity. At risk of being left behind in this “arms race”, it’s imperative that autonomous intelli- gent cybersecurity agent (AICAs) are developed to counter this emerging threat. Currently, a project at Argonne National Laboratory is using Soar as a starting point for developing a cognitive-architecture based AICA, but the utilization of Soar in this project has shortcomings, in particular the lack of modern AI principles to generate novel analysis in the face of novel situations. Generative AI has the potential to allow a Soar cognitive agent to consider a much broader range of con- textual information, and learn from past episodic knowledge in novel ways by using transformer architectures. This paper focuses on the theoretical integration of Generative AI into the Soar cognitive architecture from a cybersecurity stand- point, and discuss the advantages to doing so. Indelisio Prieto, Benjamin Blakely Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27704 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27705 Autonomous agents require the ability to identify and adapt to unexpected conditions. Real-world environments are rarely stationary, making it problematic for agents operating in such environments to learn efficient policies. There is therefore a need for a general framework capable of detecting when an agent has encountered novel conditions, and determining how it should adjust its actions. In this position paper we propose a framework that couples cognitive reasoning and generative algorithms by leveraging conflict detection to adjust to unexpected dynamics. Specifically, we propose that a metacognitive conflict resolution mechanism is necessary; such a mechanism would balance the use of commonsense and deliberative reasoning to allow the agent to navigate novel conditions. Anita Raja, Alisa Leshchenko, Jihie Kim Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27705 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27706 This paper explores the integration of two AI subdisciplines employed in the development of artificial agents that exhibit intelligent behavior: Large Language Models (LLMs) and Cognitive Architectures (CAs). We present three integration approaches, each grounded in theoretical models and supported by preliminary empirical evidence. The modular approach, which introduces four models with varying degrees of integration, makes use of chain-of-thought prompting, and draws inspiration from augmented LLMs, the Common Model of Cognition, and the simulation theory of cognition. The agency approach, motivated by the Society of Mind theory and the LIDA cognitive architecture, proposes the formation of agent collections that interact at micro and macro cognitive levels, driven by either LLMs or symbolic components. The neuro-symbolic approach, which takes inspiration from the CLARION cognitive architecture, proposes a model where bottom-up learning extracts symbolic representations from an LLM layer and top-down guidance utilizes symbolic representations to direct prompt engineering in the LLM layer. These approaches aim to harness the strengths of both LLMs and CAs, while mitigating their weaknesses, thereby advancing the development of more robust AI systems. We discuss the tradeoffs and challenges associated with each approach. Oscar J. Romero, John Zimmerman, Aaron Steinfeld, Anthony Tomasic Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27706 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27707 Dynamic Causal Modeling (DCM), a generative method for fitting large scale functional connectivity data, has provided a method of validating the architectural network structure proposed by the Common Model of Cognition (CMC). As the CMC expands, however, different methods will be required to handle the increased model complexity. While other generative methods exist that can deal with networks containing larger numbers of modules and connections, and even investigate the plausibility of different connections, a method for comparing these alternative structures will still be needed to make strong conclusions about the connection of the CMC and its variants to the true structure underlying human cognition. Catherine Sibert, Sonke Steffen Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27707 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27708 We present an updated position integrating cognitive architectures into workflow by utilizing the architecture for what it does most effectively: human-like few-shot learning integrating the vast amount of data stored by foundation models. By supplementing the language-generation capabilities with the constraints of cognitive-architectures guiding prompts, it should be possible to generate more relevant output and possibly even predict when the foundation model is hallucinating. Recent advances in few-shot learning capabilities of cognitive architectures in applied domains will be discussed with some parallel capabilities described by foundation models. Just as we use research from social psychology to 'nudge' people into making informed decisions, we should be able to use cognitive architectures to 'nudge' foundation models into developing more human-relevant content. Robert H. Thomson, Nathaniel D. Bastian Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27708 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27709 This article presents a theoretical framework for adapting the Common Model of Cognition to large generative network models within the field of artificial intelligence. This can be accomplished by restructuring modules within the Common Model into shadow production systems that are peripheral to a central production system, which handles higher-level reasoning based on the shadow productions’ output. Implementing this novel structure within the Common Model allows for a seamless connection between cognitive architectures and generative neural networks. Robert L. West, Spencer Eckler, Brendan Conway-Smith, Nico Turcas, Eilene Tomkins-Flanagan, Mary Alexandria Kelly Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27709 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27710 This paper presents experiments on using ChatGPT4 and Google Bard to create ACT-R and Soar models. The study involves two simulated cognitive tasks, where ChatGPT4 and Google Bard (Large Language Models, LLMs) serve as conversational interfaces within the ACT-R and Soar framework development environments. The first task involves creating an intelligent driving model using ACT-R with motor and perceptual behavior and can further interact with an unmodified interface. The second task evaluates the development of educational skills using Soar. Prompts were designed to represent cognitive operations and actions, including providing context, asking perception-related questions, decision-making scenarios, and evaluating the system's responses, and they were iteratively refined based on model behavior evaluation. Results demonstrate the potential of using LLMs to serve as interactive interfaces to develop ACT-R and Soar models within a human-in-the-loop model development process. We documented the mistakes LLMs made during this integration and provided corresponding resolutions when adopting this modeling approach. Furthermore, we presented a framework of prompt patterns that maximizes LLMs interaction for artificial cognitive architectures. Siyu Wu, Rodrigo F. Souza, Frank E. Ritter, Walter T. Lima Jr Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27710 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27711 The challenge of survival prediction is ubiquitous in medicine, but only a handful of methods are available for survival prediction based on time-varying data. Here we propose a novel method for this problem, using a random forest of survival trees for left-truncated and right-censored data. We demonstrate the advantage of our method on prediction of breast cancer and prostate gland cancer risk among healthy individuals by analyzing routine laboratory measurements, vital signs and age. We analyze electronic medical records of 20,317 healthy individuals who underwent routine checkups and identified those who later developed cancer. In cross-validation, our method predicted future prostate and breast cancers six months before diagnosis with an area under the ROC curve of 0.62±0.05 and 0.6±0.03 respectively, outperforming standard random forest, random survival forest, cox-regression model, dynamic deep-hit and a single survival tree. Our work proposes a new framework for survival risk prediction in time-varying data and our results suggest that computational analysis of data on healthy individuals can improve the detection of those at risk of future cancer development. Dan Coster, Eyal Fisher, Shani Shenhar-Tsarfaty, Tehillah Menes, Shlomo Berliner, Ori Rogowski, David Zeltser, Itzhak Shapira, Eran Halperin, Saharon Rosset, Malka Gorfine, Ron Shamir Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27711 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27712 Prognosis prediction is a pivotal aspect of survival analysis, particularly when considering competing risks. The contemporary landscape is enriched with an abundance of biobank data encompassing diverse risk factors like genetics, transcriptomics, and electronic health records, fueling efforts to enhance prognostic predictions. However, the resulting predictive models suffer from rare event rates, limited sample sizes, high dimensionality, and low signal-to-noise ratios. To address these challenges and amplify predictive capabilities, the integration of historical prediction models has emerged as a promising approach. Yet, prevailing integration methods often rest upon the premise of comparable underlying distributions across disparate data sources—a presumption that frequently diverges from reality. Disregarding the inherent heterogeneity among these information sources can inadvertently introduce substantial bias, underscoring the urgency of integrated competing risk analyses that systematically accommodate cohort heterogeneity. In response, we propose an original solution: a longitudinal multinomial relative entropy-based integration framework. This methodology incorporates the established prediction models from the literature, yielding refined prognostic predictions for newly acquired datasets. Qinmengge Li, Kevin He Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27712 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27713 Survival analysis is widely employed across medicine, business, and the social sciences. However, the absence of a unified and standardized software solution for evaluating survival analysis models impedes its broader application by researchers. In this research, we fill this gap by providing a comprehensive Python package, SurvivalEVAL, which implements seven evaluation metrics specific to survival analysis. The SurvivalEVAL package is designed to serve as a convenient and straightforward toolkit for individual survival distribution models. The package is publicly available on GitHub at https://github.com/shi-ang/SurvivalEVAL. Shi-ang Qi, Weijie Sun, Russell Greiner Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27713 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27714 Federated Survival Analysis (FSA) is an emerging technique for analyzing decentralized survival data while preserving data privacy and providing more generalized survival predictions. Existing FSA methods often rely on deep learning models, which can be computationally expensive and require substantial data and communication rounds. Recent research has demonstrated that ensemble-based approaches like random survival forests can achieve comparable performance to deep learning models with a single communication round within a federated learning (FL) framework, especially when dealing with small, decentralized datasets. However, these approaches have yet to address the challenges specific to FSA, such as data heterogeneity, non-uniform censoring, and competing risks. To address these challenges, we propose FedPRF, an FL framework for survival analysis based on Federated Pseudo Values (FPV) based random forests model, we call it RFpseudo. FedPRF introduces FPV to handle issues related to censoring and effectively address the unique challenges of FSA. FedPRF is computationally efficient, requiring only two communication rounds across clients: (i) computing FPV and (ii) aggregating a subset of trees trained locally at clients. Extensive experiments on distributed survival data with a single event and multiple competing events demonstrate that FedPRF achieves performance close to the gold-standard centralized training setting and outperforms deep learning-based FSA approaches. Importantly, FedPRF maintains the interpretability of centrally trained survival models. Furthermore, it is scalable to large-scale data and highly distributed settings with numerous clients. Md Mahmudur Rahman, Sanjay Purushotham Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27714 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27715 In this paper, we discuss a selection of tools from dynamical systems and order statistics, which are most often utilized separately, and combine them into an algorithm to estimate the parameters of mathematical models for infectious diseases in the case of small sample sizes and left censoring, which is relevant in the case of rapidly evolving infectious diseases and remote populations. The proposed method relies on the analogy between survival functions and the dynamics of the susceptible compartment in SIR-type models, which are both monotone decreasing in time and are both determined by a dual variable: the hazard function in survival prediction and the number of infected people in SIR-type models. We illustrate the methodology in the case of a continuous model in the presence of noisy measurements with different distributions (Normal, Poisson, Negative Binomial) and in a discrete model, reminiscent of the Ricker map, which admits chaotic dynamics. This estimation procedure shows stable results in experiments based on a popular benchmark dataset for SIR-type models and small samples. This manuscript illustrates how classical theoretical statistical methods and dynamical systems can be merged in interesting ways to study problems ranging from more fundamental small sample situations to more complex infectious disease and survival models, with the potential that this tools can be applied in the presence of a large number of covariates and different types of censored data. Alessandro Maria Selvitella, Kathleen Lois Foster Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27715 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27716 In the field of healthcare, individual survival prediction is important for personalized treatment planning. This study presents machine learning algorithms for predicting Individual Survival Distributions (ISD) using electrocardiography (ECG) data in two different formats. The models, which predict time until death, are developed and evaluated on a large, population-based cohort from Alberta, Canada. Our results demonstrate that models trained on raw ECG waveforms significantly outperform those trained on traditional ECG measurements in several metrics, including concordance index, hinge L1 loss, margin L1 loss, and margin truncated L1 loss. Additionally, the integration of predicted probabilities from wide-range diagnostic tasks not only enhances our ISD models' performance but also makes them significantly superior to other models across all evaluation metrics in individual survival prediction tasks. This innovative approach highlights the potential to leverage insights from diagnostic models for prognostic tasks, such as individual survival prediction. These findings could have far-reaching implications for the development of personalized treatment plans and open new avenues for future research in survival prediction using ECGs. Weijie Sun, Sunil Vasu Kalmady, Shi-ang Qi, Nariman Sepehrvand, Abram Hindle, Russell Greiner, Padma Kaul Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27716 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27717 Survival prediction via training deep neural networks with giga-pixel whole-slide images (WSIs) is challenging due to the lack of time annotation at the pixel level or patch (instance). Multiple instance learning (MIL), as a typical weakly supervised learning method, aims to resolve this challenge by using only the slide-level time. The attention-based MIL method leverages and enhances performance by weighting the instances based on their contribution to predicting the outcome. A WSI typically contains hundreds of thousands of image patches. Training a deep neural network with thousands of image patches per slide is computationally expensive and time-consuming. To tackle this issue, we propose an adaptive-learning strategy where we sample a subset of informative instances/patches more often to train the deep survival neural networks. We also present other sampling strategies and compare them with our proposed sampling strategy. Using both real-world and synthesized WSIs for survival, we show that sampling strategies significantly can significantly reduce computing time while result in no or negligible performance loss. We also discuss the benefits of each instance sampling strategy in different scenarios. Aliasghar Tarkhan, Trung Kien Nguyen, Noah Simon, Jian Dai Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27717 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27718 Classical planning representation languages based on first-order logic have preliminarily been used to model and solve robotic task planning problems. Wider adoption of these representation languages, however, is hindered by the limitations present when managing implicit world changes with concise action models. To address this problem, we propose an alternative approach to representing and managing updates to world states during planning. Based on the category-theoretic concepts of C-sets and double-pushout rewriting (DPO), our proposed representation can effectively handle structured knowledge about world states that support domain abstractions at all levels. It formalizes the semantics of predicates according to a user-provided ontology and preserves the semantics when transitioning between world states. This method provides a formal semantics for using knowledge graphs and relational databases to model world states and updates in planning. In this paper, we conceptually compare our category-theoretic representation with the classical planning representation. We show that our proposed representation has advantages over the classical representation in terms of handling implicit preconditions and effects, and provides a more structured framework in which to model and solve planning problems. Angeline Aguinaldo, Evan Patterson, James Fairbanks, William Regli, Jaime Ruiz Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27718 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27719 Human-robot interaction consists of a rich set of behaviors between humans and robots often requiring the exchange of personal and sensitive information between them. From a conceptual framework, this paper discusses whether a robot who self-discloses personal information when conversing with a user will prompt the user to reciprocate and self-dis- close personal and sensitive information to the robot. Additionally, the paper discusses various factors which may influence whether self-disclosure of personal information between human and robot occurs and briefly discusses aspects of a conceptual representational system necessary for HRI enabling the robot to self-disclose to a user. Jessica K. Barfield Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27719 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27720 Layered architectures have been widely used in robot systems. The majority of them implement planning and execution functions in separate layers. However, there still lacks a straightforward way to transit high-level tasks in the planning layer to the low-level motor commands in the execution layer. In order to tackle this challenge, we propose a novel approach to ground the manipulator primitive tasks to robot low-level actions using large language models (LLMs). We designed a program-function-like prompt based on the task frame formalism. In this way, we enable LLMs to generate position/force set-points for hybrid control. Evaluations over several state-of-the-art LLMs are provided. Yue Cao, C. S. George Lee Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27720 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27721 Large language models offer new ways of empowering people to program robot applications-namely, code generation via prompting. However, the code generated by LLMs is susceptible to errors. This work reports a preliminary exploration that empirically characterizes common errors produced by LLMs in robot programming. We categorize these errors into two phases: interpretation and execution. In this work, we focus on errors in execution and observe that they are caused by LLMs being “forgetful” of key information provided in user prompts. Based on this observation, we propose prompt engineering tactics designed to reduce errors in execution. We then demonstrate the effectiveness of these tactics with three language models: ChatGPT, Bard, and LLaMA-2. Finally, we discuss lessons learned from using LLMs in robot programming and call for the benchmarking of LLM-powered end-user development of robot applications. Juo-Tung Chen, Chien-Ming Huang Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27721 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27722 In this paper, we discuss the interplay between distributed multi-agent systems and the concept of "lag." We introduce a new representation of these systems to more clearly include time and the effects of latency. We also present real-world situations where communication delays can mean the difference between success and failure, between order and chaotic disarray. Sarah Dumnich, William Birmingham, Britton Wolfe Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27722 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27723 Reproducibility in robotics research requires capable, shared hardware platforms which can be used for a wide variety of research. We’ve seen the power of these sorts of shared platforms in more general machine learning research, where there is constant iteration on shared AI platforms like PyTorch. To be able to make rapid progress in robotics in the same way, we propose that we need: (1) shared real-world platforms which allow different teams to test and compare methods at low cost; (2) challenging simulations that reflect real-world environments and especially can drive perception and planning research; and (3) low-cost platforms with enough software to get started addressing all of these problems. To this end, we propose HomeRobot, a mobile manipulator software stack with associated benchmark in simulation, which is initially based on the low-cost, human-safe Hello Robot Stretch. Chris Paxton, Austin Wang, Binit Shah, Blaine Matulevich, Dhruv Shah, Karmesh Yadav, Santhosh Ramakrishnan, Sriram Yenamandra, Yonatan Bisk Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27723 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27724 We argue for the use of Petri nets as a modeling language for the iterative development process of interactive robotic systems. Petri nets, particularly Timed Colored Petri nets (TCPNs), have the potential to unify various phases of the development process-design, specification, simulation, validation, implementation, and deployment. We additionally discuss future directions for creating a domain-specific variant of TCPNs tailored specifically for HRI systems development. Pragathi Praveena, Andrew Schoen, Michael Gleicher, David Porfirio, Bilge Mutlu Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27724 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27725 As service robots become more capable of autonomous behaviors, it becomes increasingly important to consider how people will be able to communicate with a robot about what task it should perform and how to do the task. There has been a rise in attention to end-user development (EUD), where researchers create interfaces that enable non-roboticist end users to script tasks for autonomous robots to perform. Currently, state-of-the-art interfaces are largely constrained, often through simplified domains or restrictive end-user interaction. Motivated by our past qualitative design work exploring how to integrate a care robot in an assisted living community, we discuss challenges of EUD in this complex domain. One set of challenges stems from different user-facing representations, e.g., certain tasks may lend themselves better to a rule-based trigger-action representations, whereas other tasks may be easier to specify via a sequence of actions. The other stems from considering the needs of multiple stakeholders, e.g., caregivers and residents of the facility may all create tasks for the robot, but the robot may not be able to share information about all tasks with all residents due to privacy concerns. We present scenarios that illustrate these challenges and also discuss possible solutions. Laura Stegner, David Porfirio, Mark Roberts, Laura M. Hiatt Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27725 Mon, 22 Jan 2024 00:00:00 -0800 https://ojs.aaai.org/index.php/AAAI-SS/article/view/27726 This paper describes a general approach to integrating higher-level reasoning mechanisms including planning and scheduling methods with lower-level robotic control processes. We adopt a domain-independent task representation language TAEMS to describe the knowledge of tasks, resources, and their interrelationships. This TAEMS representation language serves as the input of the reasoning functions, which generate a schedule of executable methods to be executed by the robot in the physical world. In the execution process of this goal-directed plan, the robot also needs to attend to basic functions. The potential interactions between the plan and these basic functions would lead to interesting challenges that will be discussed. An integrated development platform with a simulator that supports real-world physics is also presented. Shelley Zhang, Alexander Moulton, Abhijot Bedi, Eugene Chabot Copyright (c) 2023 Association for the Advancement of Artificial Intelligence https://ojs.aaai.org/index.php/AAAI-SS/article/view/27726 Mon, 22 Jan 2024 00:00:00 -0800