Artificial Intelligence (AI) permeates various technologies including information and communication technology, significantly reshaping societal practices and individual routines. While this offers benefits, it also prompts concerns about the negative changes of AI, requiring careful governance. This paper focuses on mitigating these risks by offering recommendations for establishing coherent legal frameworks to safeguard innocent individuals from AI-related threats or potential harm, whether intentional or unintentional. We propose a framework that utilizes a large language model to assess the alignment between AI ethical guidelines and bills. A case study is conducted to analyze the Korean regulatory landscape, since Korea’s AI legal frameworks particularly lag behind amid its rapid advancement of technology. This study employs a range of methods - descriptive, correlation, cluster, and semantic analyses - to offer a comprehensive comparison of Korea’s legislative documents. Our findings reveal continuity in certain aspects but discontinuity in others between these two governance tools. Consequently, we suggest measures to enhance the consistency between these two realms for AI regulation, contributing to more robust and effective AI governance practices.

The advent of Large Language Models (LLMs) has potential efficiencies in information retrieval, notably in acquiring expert knowledge. This study aimed to compare the quality of responses derived from human expert interviews against those generated by ChatGPT in subfields within Industrial Engineering. Interview questions in fields of 'Technology Forecasting' and 'UI/UX' were asked to ChatGPT. Text similarity models were utilized to compare the responses made by human experts and ChatGPT, while the quality of contents were compared through qualitative analysis. Quantitative analysis revealed high similarity scores in both subfields, with UI/UX responses exhibiting slightly higher scores. Furthermore, responses to factual-based questions yielded higher similarity scores than those to personal queries. The findings suggest that LLMs can serve as viable alternatives for information retrieval, providing responses of comparable quality to those of human experts. 

The advent of AI models, particularly Large Language Models (LLMs) like ChatGPT, is being widely used in the educational sector, serving as an effective tool to perform routine tasks. This study was conducted to analyze the efficacy of ChatGPT plugins and extensions in Chrome browsers in automating routine academic tasks such as information retrieval, summarization, language translation, and text formatting, and to gauge the perceived efficiency gains by users. This research concentrated on the Human-Computer Interaction (HCI) facets of these tools. This study employed a methodology comprising usability tests, questionnaires, and interviews with participants of varying experience levels. The results proved that ChatGPT plugins and Chrome extensions could significantly automate routine tasks, with users perceiving notable efficiency gains and rating above-average scores on the System Usability Scale (SUS) for these tasks. There were no significant differences in performance between highly experienced and inexperienced participants in terms of time and error rates. However, disparity in the SUS scores and a divergence in the ethical perspectives on using AI tools in academia was noted, with inexperienced participants deeming it unethical, contrary to their experienced counterparts. This research has implications on the use of LLMs like ChatGPT in the academic fields.

The advent of large language models, such as ChatGPT, has revolutionized the way we interact with artificial intelligence. These models have the potential to save time and effort in obtaining information, including expert knowledge, through their ability to generate human-like responses. This study aims to compare the quality of interview responses generated by ChatGPT with those provided by human experts. A sample of interview prompts were collected to cover the topic of biology, and responses were collected from ChatGPT and human experts in each domain. The responses were then analyzed to measure the similarities in terms of validity and trustworthiness. Additionally, the efficiency and convenience of obtaining responses from ChatGPT were evaluated in comparison to traditional expert interviews. Preliminary findings suggest that ChatGPT can generate responses that are similar to those provided by human experts in terms of quality, validity, and trustworthiness. However, differences were also observed in terms of nuanced responses and expert insights, which may be attributed to the limitations of language models in capturing domain-specific knowledge and context. Despite these differences, the use of ChatGPT as an alternative source of information has the potential to save time and effort in obtaining expert insights, thereby increasing efficiency and convenience. The results of this study contribute to the growing body of literature on the use of large language models in obtaining expert knowledge and highlight their potential benefits and limitations. This research has implications for various fields, including information retrieval, data analysis, and decision-making, where the use of language models can provide a valuable alternative to traditional expert interviews.

Biodiversity acts as an important factor in protecting not only the ecosystem but also humans all over the world. However, currently due to indiscriminate development, biodiversity is being destroyed. In order to maintain biodiversity, it is crucial to protect endangered species. Even though both the United States and Korea have laws to protect endangered species, contrary to the outstanding achievements shown by the Endangered Species Protection Act (ESA) of the United States, Korea’s Wildlife Protection and Management Act (Wildlife Act) is not showing high effects. Hence, with the aim of promoting biodiversity, by analyzing the similarities and differences between the two Acts, this study determined what improvements can be made to Korea’s Wildlife Protection and Management Act. The similarities between the two Acts were regarded as the strengths of Korea’s Wildlife Protection and Management Act, which were 1) protection of habitat and successful execution of the law, 2) inclusion of both native and foreign endangered species, 3) adequate period of evaluation, and 4) clarity on the ministry and office concerned. On the other hand, differences were regarded as the limitations of Korea’s Wildlife Protection and Management Act, which were 1) lack of specificity and broad scope of aid, 2) lack of specific categorization, 3) lack of priorities, 4) lack of stability and long-term effects, and 5) lack of social consensus. Therefore, amendments such as specifically mentioning the method, contents, and standards of reward, adding a ‘candidate species’ category, prioritizing certain species, securing a sufficient amount of budget, creating a single specialized ministry that comprehensively manages matters related to endangered species, maintaining the Korean Wildlife Act for a while to increase its stability, and increasing public interest by clarifying the specific amount of rewards and penalties, ways of education provision regarding endangered species, and simple methods for protecting endangered species and so on are recommended to be made in Korea’s Wildlife Protection and Management Act. 

With the rapid development of artificial intelligence (AI) technology, AI policy plans are being suggested. However, no precise policies are established, nor are the evaluation criteria on AI policy proposals’ effectiveness adequate. Hence, this study presents a comprehensive framework for analyzing the legal status of three AI conversational agents to determine their necessary policies and evaluate the effectiveness of AI policies through comparison. By analyzing the reviews on conversational agents using Bidirectional Encoder Representations from Transformers (BERT) for legal status classification, the type of legal status and the degree to which a review aligns with that status was identified. Character AI had the highest level of legal status among the conversational agents, while differences in legal status intensities were mainly found depending on region. This study has significance for analyzing the mismatch between AI policy plans and policies needed for conversational agents, and suggesting the importance of AI policy improvements.

Due to the increased variation on the types of conversational agents, user’s emotions differ depending on the different characteristics of such agents. In this study, response data on 3 different types of conversational agents from various platforms were analyzed using Bidirectional Encoder Representations from Transformers (BERT) for emotion classification. The results provided emotional intensity scores for four basic emotions. Characteristics of reviewers were analyzed using a model to predict the demographic information and cluster reviewers into groups to analyze how different attributes influenced the type and intensity of emotion expressed towards such agents. Our findings indicated a fascinating pattern: Lee Luda elicited the highest intensity of emotion, primarily negative, such as anger and sadness. All three conversational agents had the lowest score for joy among the four emotions. We contribute to designing and regulating conversational agents in a way that is fairer and beneficial to everyone.

Routinized academic tasks are repetitive activities with uniform procedures. They are simple and timeconsuming but essential in research. The recent Large Language Models (LLMs) transform tools to assist academic works but they are criticized by their questionable credibility and efficiency. To clarify this, we conducted a usability study with a particular focus on GPTs and ChatGPT Chrome Extensions in academic writing. For GPTs, our customized GPT ‘Research AIde’ was used, while participants independently selected Chrome Extensions they deemed most effective for their tasks. Three experience-based groups of 30 participants shared user experience, functionality, ethical considerations, future applications in using AI in academia. We find that LLMs automates routinized tasks but has ethical problems. From this, we draw guidelines for responsible but effective use of AI in academia.

As Large Language Models (LLMs) develop and their accessibility increases, the opportunities for agents to make better decisions increase. In this study, we introduce a text-based game as a way to improve Reinforcement Learning (RL) agent's decision-making process. We explore how the agent acts in a text-based horror game and how the agent's persona influences its game-playing performance. We found that the non-instructional elements in the game can effectively train the agent and lead to more realistic behaviour. Furthermore, we utilized prompt engineering to make an LLM imitate the personal characteristics of human participants. Our result shows that the scores and success rates between the LLM and humans were statistically similar although there were significant differences in time. This underscores the possibility of LLMs mimicking human-like gameplay behaviour in text-based games. We conclude that there is a potential to advance both RL and AI-driven persona emulation within non-instructional contexts.