A Flexible Multi-Agent LLM-Human Framework for Fast Human Validated Tool Building

Ensuring that AI systems do what we, as humans, actually want them to do is one of the biggest open research challenges in AI alignment and safety. My research seeks to directly address this challenge by enabling AI systems to interact with humans to learn aligned and robust behaviors. The way robots and other AI systems behave is often the result of optimizing a reward function. However, manually designing good reward functions is highly challenging and error‐prone, even for domain experts. Although reward functions are often difficult to manually specify, human feedback in the form of demonstrations or preferences is often much easier to obtain but can be difficult to interpret due to ambiguity and noise. Thus, it is critical that AI systems take into account epistemic uncertainty over the human's true intent. As part of the AAAI New Faculty Highlight Program, I will give an overview of my research progress along the following fundamental research areas: (1) efficiently quantifying uncertainty over human intent, (2) directly optimizing behavior to be robust to uncertainty over human intent, and (3) actively querying for additional human input to reduce uncertainty over human intent.

Penetration testing is widely acknowledged as the foremost method for evaluating network security. However, three challenges impede the generation of strategies that align with human expectations. In this article, we present, for the first time, a method based on human feedback to enhance strategy generation. Our approach comprises two components: agent training and decision-making. During agent training, we establish a hierarchical framework to decompose tasks and a knowledge base to offer advice for improving data efficiency. We then impose constraints on the action space to mitigate ineffective exploration. Finally, we train a reward model based on human feedback and fine tune the model guided by this reward model. In decision-making, we process the model output to enhance decision accuracy. We crafted scenarios based on real-world networks, and the results demonstrate the effectiveness of our method in generating penetration testing strategies that align more closely with human intentions.

This paper critically evaluates the attempts to align Artificial Intelligence (AI) systems, especially Large Language Models (LLMs), with human values and intentions through Reinforcement Learning from Feedback methods, involving either human feedback (RLHF) or AI feedback (RLAIF). Specifically, we show the shortcomings of the broadly pursued alignment goals of honesty, harmlessness, and helpfulness. Through a multidisciplinary sociotechnical critique, we examine both the theoretical underpinnings and practical implementations of RLHF techniques, revealing significant limitations in their approach to capturing the complexities of human ethics, and contributing to AI safety. We highlight tensions inherent in the goals of RLHF, as captured in the HHH principle (helpful, harmless and honest). In addition, we discuss ethically-relevant issues that tend to be neglected in discussions about alignment and RLHF, among which the trade-offs between user-friendliness and deception, flexibility and interpretability, and system safety. We offer an alternative vision for AI safety and ethics which positions RLHF approaches within a broader context of comprehensive design across institutions, processes and technological systems, and suggest the establishment of AI safety as a sociotechnical discipline that is open to the normative and political dimensions of artificial intelligence.

Personalization is key to bring robots into our daily lives: people differ in goals, abilities, and preferences, yet most robot learning systems rely on fixed task objectives. My research vision is to develop efficient methods for personalized robot learning from human feedback. This extended abstract addresses: what types of human feedback are most useful, how can we use them efficiently, and how these insights enable personalized assistance for humans. We first study which modalities of feedback best capture human intent—demonstrations, language, or preference on trajectories—and how combining them increases clarity and data efficiency. We then investigate methods for augmenting limited human feedback to build robust and generalizable reward models. Finally, we extend these techniques to physical human-robot interaction, where data-efficient personalization is critical for safety and comfort. Together, these efforts aim to make personalization a practical capability for everyday robots.

We welcome you to the fourteenth annual International Computing Education Research Conference, ICER 2018, sponsored by the ACM Special Interest Group on Computer Science Education (SIGCSE). Espoo, Finland is the host city for this year's conference, with sessions taking place on the Aalto University campus. ICER stands as the premier ACM forum for dissemination and discussion of the latest findings in computing education research across the globe. ICER research papers represent significant, rigorous contributions to the field. One hundred twenty-five research papers were submitted, with twenty-eight papers accepted for publication (a 22% acceptance rate) in the conference proceedings in the ACM Digital Library. All papers were double-blind peer reviewed by three members of the program committee. In addition, each paper received a meta-review by a member of the senior program committee, with the two Program Co-Chairs making final acceptance decisions. In addition to the research paper presentations, ICER includes Lightning Talk and Poster sessions as a way for ICER attendees to present early results, gain feedback from conference attendees, find collaborators on a topic, and/or spark discussion among conference participants. The conference also serves a vital mentoring and advising role for upcoming discipline-based computing education researchers through the Doctoral Consortium. The Work in Progress workshop is a dedicated one-day workshop for ICER attendees to provide and receive friendly, constructive feedback on research during formative stages of development. Associated co-located workshops with external sponsorship at ICER 2018 include Building Your Team of Change Champions and Computing Science Educational Infrastructure: Crossing the Borders. We are honored to welcome Kirsti Lonka, the Professor of Educational Psychology in the Faculty of Education at the University of Helsinki, Finland to present the ICER 2018 keynote address. For the past 30 years, she has been investigating various aspects of learning in higher education, medical education, teacher education, and postgraduate education (PhD students). In the area of academic writing, her work has focused around conceptions of writing, note taking, process writing, portfolios and writing across curriculum (writing as a learning tool). Professor Lonka's keynote, Growing minds - 21st century competences and digitalization among Finnish youth?, discusses her recent research concerning the digital technology background among Finnish school children and its relation to their activities and attitudes. The keynote addresses the intensive reforms, both in research-based teacher education and at school that are going on in Finland. In addition to the new Finnish curricula, the latest developments in education, innovative pedagogical methods, technologies in education and new learning environments will be presented.

Artificial intelligence can be defined as one of the best gifts of science and technology to the mankind. These emerging technologies plays an important role in “revolutionizing the modern world”. Artificial intelligence has made life easy as it has been used and is still in use in different spheres of the evolving human life. Can AI be considered as a “legal person”? Currently, various countries including the UK and USA consider the fact that AI is still dependent on some amount of human input. The extensive usage of generative AI programs leaves us with the question of who, if anyone,may hold the copyright to content created using these programs, notwithstanding the fact that theAI's programmer as well as it's user, and the AI program itself all together play an important rolein the creation of these works. One of the main reasons why it's very challenging to copyright AI-generated works is the vagueness in relation to human involvement and intentions and various other questions regarding ownership. There are chances of a model creator to be accused of “vicarious liability” for the said infringement. Another issue is that, in International Journal of Scientific Research in Engineering and Management (IJSREM) Volume: 08 Issue: 01 | January - 2024 SJIF Rating: 8.176 ISSN: 2582-3930 © 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM27618 | Page 2 AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies. - Sam Altman jurisdictions, like the US, any output generated solely using a machine is not eligible for copyright protection, as most jurisdictions tend to give protection to only the original works, those having a “human author”. This research paper tends to give reasons on the questions of “whether AI generated works are eligible for copyright protection”. KEYWORDS: Artificial Intelligence, Intellectual property, innovation and technology, copyright law, dataprivacy, legal person, AI generated works, digital jurisdiction, liability of AI,copyright infringement.

Multi-tier retail organizations operate in highly dynamic environments where store performance, customer demand, and strategic positioning continuously evolve. As a result, store tier transitions are not static events but ongoing managerial decisions requiring coordination across merchandising, supply chain, and financial planning functions. This research paper presents a comprehensive managerial and analytical framework for Dynamic Tier Change Management (DTCM), emphasizing governance structures, decision-making processes, and system-level integration. The proposed framework systematically addresses the lifecycle of store tier transitions from trigger identification and cluster reassignment to assortment propagation and execution deployment. Unlike traditional approaches that rely heavily on manual interventions, this framework integrates rule-based decision systems, data-driven analytics, and structured planning workflows. The study highlights how organizations can significantly reduce operational latency, improve assortment accuracy, and enhance planner productivity through structured tier transition management

Formal and online education has come a long way and with changing times, new methods of teaching are being developed according to the need of the subject and the needs of the students. However certain age-old methods retain their charm with changing times and only emerge stronger. One such method is story telling. The purpose of this research paper is to work out how educators, students et al. round the world are using storytelling to support the digital educational process. This study provides a general framework about educational uses of storytelling in several subjects and describes the various perceptions of individuals in understanding storytelling. A survey was used to collect responses from a group of educators, students and others in educational settings to determine how they are using storytelling for online educational purposes. The results show the current situation of educational uses of storytelling and explore some of the benefits and challenges educators face in implementing storytelling in their institutions and teaching online classes. The paper also attempts to give solutions in bridge the digital divide through engaging storytelling. The research paper also explores reasons for storytelling's current popularity as a teaching and learning tool, then suggests how educators and students can work together to create a storytelling culture. To conclude, benefits for college kids who learn through storytelling are outlined.

The brain-computer interface (BCI) technique is a novel control interface to translate human intentions into appropriate motion commands for robotic systems. The aim of this study is to apply an asynchronous direct-control system for humanoid robot navigation using an electroencephalograph (EEG), based active BCI. The experimental procedures consist of offline training, online feedback testing, and real-time control sessions. The amplitude features from EEGs are extracted using power spectral analysis, while informative feature components are selected based on the Fisher ratio. The two classifiers are hierarchically structured to identify human intentions and trained to build an asynchronous BCI system. For the performance test, five healthy subjects controlled a humanoid robot navigation to reach a target goal in an indoor maze by using their EEGs based on real-time images obtained from a camera on the head of the robot. The experimental results showed that the subjects successfully controlled the humanoid robot in the indoor maze and reached the goal by using the proposed asynchronous EEG-based active BCI system.

A generic flexible and scalable framework for hierarchical parallelization of population-based metaheuristics

With the increase in popularity of smart devices for metering purposes, the concerns and the scope of the threats also gets bigger. In order to raise awareness and provide a simulated environment for information security training to smart meter manufacturers, we developed MeterGoat, which is a flexible framework for smart meter penetration testing. In this research paper, we describe the implementation of MeterGoat and present some sample lessons that explore vulnerabilities based on weak anti-tampering techniques, flawed authentication, and improper use of cryptography.

This article examines numerous aspects of using Chinese national music to enhance the effectiveness of dance education. Traditional musical works of China are not only intangible cultural heritage of ancient Chinese culture but also possess extensive potential within modern dance education. Their inherent rhythmic and melodic characteristics determine the significance of applying national music in the classroom: thoughtful listening to such compositions allows students to effectively develop comprehensive professional skills as dancers. This phenomenon has been thoroughly discussed in this research paper. Special attention has been paid to defining the role of Chinese national music in teaching dance techniques, strengthening cultural understanding among different peoples, and fostering cross-style artistic creativity. Within this study, a critical analysis of pedagogical and art literature, as well as the specifics of Chinese national music in the context of dance education, was conducted. Based on the research conducted, it is concluded that the systematic integration of national music into dance education can not only enhance the effectiveness of learning but also deepen the cultural context of the professional creativity of modern dancers as a whole. The study also showed that the rhythmic patterns of traditional Chinese music can effectively improve a dancer's coordination of movements. National music evokes an emotional and sensory response in learners, as well as raises their awareness of their own cultural identity. At the same time, ethnic music provides a structural foundation for free creative experimentation and the creation of hybrid dance styles, allowing dancers to incorporate modern techniques into their works while blending them with traditional compositions.

The Federal Institute of Materials Research and Testing (BAM) – 150 Years of Enabling Scientific and Technological Breakthrough

Machine learning plays a key role in a wide range of applications such as data mining, natural language processing and expert systems. It provides a solution in all domains for further development when large data is applied. Supervised learning is consist of mathematical algorithm to optimize the functions with given inputs. Machine learning solves problems that cannot be solved by numerical values. In this research paper, a model is developed to improve classification algorithm using anxiety of juvenile. Prediction and classification are made using data. A machine learning tool is used for pre-processing and first level of model is data preparation and ranking prototype used for filtration of data . Then Probabilistic estimation hypothesis is to find the hypothesis value based on statistical functions and classification of anxiety predictor model is used for prediction and classification. Comparison of Algorithm and experimental are done using machine learning software. According to the experiment, the model is more efficient and accurate compared with other classification algorithm as results shown.

Public procurement policies and legal frameworks play a crucial role in shaping the transparency, fairness, and integrity of government transactions. This paper examines and evaluates the effectiveness of public procurement policies and legal structures in addressing and mitigating corruption in selected regions within the Western Cape Province of South Africa. Drawing on a comprehensive review of relevant literature, case studies, and expert opinions, this research aims to provide insights into the existing framework’s strengths, weaknesses, and potential improvements. The main emphasis of this research paper is to assess public procurement policy design and the effectiveness of measures including regulations aimed at combating corruption in the Western Cape’s public sector in South Africa. Public procurement regulations are widely implemented in various governmental institutions around the world. Data were collected from twenty-five public sector organizations in the Western Cape province, South Africa, using three data collection methods: interviews, surveys, and document analysis of procurement policies and guidelines. The study’s key findings demonstrated that a fragmented legal and policy framework is complex and opens avenues of corrupt practices and maladministration. The paper fills the gap in scholarly research by exploring how corrupt practices can be reduced by implementing and operationalising robust, flexible, and outcome-based legal and policy frameworks. This paper contributed to the body of knowledge by making policymakers and implementers aware of the pitfalls of a fragmented and misaligned policy framework in public procurement.