Background: As the landscape of the fashion industry is challenged by the emergence of big data and high sustainability demands, efficient solutions for product innovation and development are required. Artificial Intelligence (AI) is generating organizational shifts in various industries, but the fashion industry is still very early in its adoption. AI shows abilities to facilitate the challenges of the industry, and its application in creative design and product development processes is estimated to hold potential. Problem: As the fashion industry is characterized by creativity and human ideation, there is a need to evaluate if AI is compatible with the values of the industry. Management’s attitudes are proven to influence the adoption of digital technologies, leaving implications to study the attitudes of professionals in design and product development towards AI as well. Further, it is relevant to understand the possibilities and limitations of utilizing generative AI in creative processes, to ensure a successful implementation. Purpose: This thesis aims to investigate the implementation of AI in creative ideation and product development within the fashion industry, particularly exploring the attitudes of fashion professionals toward the relationship between human ideation and AI to determine the industry’s current position. Method: This study utilized qualitative research design by conducting 10 semi-structured interviews with professionals working in the fields of fashion design, product development, and AI. Conclusion: The results show that AI is currently not implemented within fashion, among the interviewees. The study identified determinants, such as awareness, attitudes, data, knowledge, objectives, and competencies that influence the adoption of AI, in the early stages. The attitudes toward AI are an essential factor in the early stages of adoption.

The marketing process has undergone significant changes over the years, due to new technologies. Among these advancements, artificial intelligence (AI) has been increasingly used to generate advertising messages and images. This study examines consumers' perceptions of AI-generated advertisements, with a focus on the influence of age. Participants were divided into two groups, with one group believing the ads were human-created and the other assuming they were AI-generated. Results indicated that perceived ad falsity led to more negative perceptions. However, younger individuals exhibited more favorable attitudes toward AI-generated ads compared to older individuals. Despite an overall negative bias towards AI-generated ads, one AI-generated ad was received as most positively across all age groups, supporting prior research that AI-generated content can be well-received. A theoretical model was built and tested to explore the relationship between age and ad perception, suggesting that prior experience with AI, attitudes towards AI, and AI credibility sequentially mediate this relationship. While the effect of machine heuristics was found to be nonsignificant, it did influence AI credibility, indicating potential avenues for future research.

Condition assessment of civil engineering infrastructure systems is of growing importance as they face aging and degradation due to both human-made activities and environmental factors. Nevertheless, challenges persist in data collection, leading to "data scarcity", and the need for frequent site visits in inspections, presenting significant obstacles in the assessment of the civil infrastructure systems. This dissertation aims to overcome these challenges by exploring the potential of two emerging technologies: Generative Artificial Intelligence (AI) and Virtual Reality (VR). In tackling the issue of data scarcity, the research question revolves around how Generative AI can be utilized to mitigate data collection-related constraints and increase data availability, thus facilitating health monitoring applications of infrastructure systems. For that, using various Generative AI models, the dissertation works on acceleration response data generation, including data augmentation and domain translation applications on different structures. In addressing the site visit challenge, the dissertation focuses on the use of VR to bring the infrastructure to the experts in a single collaborative immersive environment and investigate its impact on decision-making in inspections. For that, using VR technology, the dissertation develops a Virtual Meeting Environment (VME) integrated with the infrastructure data and models presented through novel immersive visualization techniques. The dissertation further investigates the impact of VME on decision-making in infrastructure inspections through experimentation with engineers. These investigations of the use of Generative AI and VR demonstrate various contributions. Generative AI effectively tackles the need for vast datasets in data-intensive damage detection applications. It also demonstrates its potential in estimating representative response data for various structural conditions across dissimilar infrastructures. VME, on the other hand, offers an increased understanding of the material along with a safer, practical, and cost-effective complementing alternative to traditional in-person site visits. It further reveals how VME improves decision-making in infrastructure inspections.

This thesis investigates the impact of generative AI technologies on deliberative democracy in Latin America, focusing on the cases of Peru and Brazil. Through qualitative interviews with experts from both countries, this research uses several techniques to process the data collected and intends to draw conclusions from it. The study found that unregulated generative AI is a source of concern for Peruvian and Brazilian participants. Another key finding was that these two countries are not keen on collaborating with other countries in the region to establish a regulatory framework on generative AI. Through the analysis of the five principles of deliberative democracy: citizen participation, inclusivity, accessibility, accountability, and quality of deliberation, this project was able to draw conclusions on which of these dimensions could be impacted by generative AI. The results of the study show that in the case of  Peru, generative AI technologies impact citizen participation and accountability the most.  In the case of Brazil generative AI technologies impact citizen participation, inclusivity, and transparency the most. The most important finding, however, is the commonality shared by Peru and Brazil which signifies that in both countries citizen participation will be impacted by generative AI. The question of how citizen participation will be impacted requires a deeper analysis of the responses of the interviewees as they have both positive and negative perspectives on the issue.

This thesis presents the design of an educational tool that enhances student engagement and interaction during group presentations in large classroom settings. Specifically, the study aimed to create a tool that streamlines the management of questions and participation, making the process more efficient and equitable for students and instructors. The research explored three primary questions: (1) How can educational software be designed to increase engagement and participation during student presentations? (2) How can AI be used to assist in tasks traditionally performed by professors, such as managing Q&A sessions? (3) How does the application of design thinking, particularly the empathy stage, influence the development of effective educational tools? Students provided ample feedback on improving the course and detailed explanations for their preferences. Qualitative methods including reflexive thematic analysis were used to process this volume of feedback. Descriptive statistics, confusion matrices, and Kappa scores were used to ensure the integrity of the analysis. An open-source large language model, Meta’s LLaMA, was implemented to automate the selection and clustering of questions during student-led Q&A sessions, with these results compared against instructor-selected questions. AI-driven question selection matched the effectiveness of instructor selections and enhanced efficiency, significantly reducing the logistical burden on educators while sustaining student engagement. Additionally, the research gathered extensive data on students’ experiences within the university classroom, with particular attention to issues such as anxiety, group dynamics, and disengagement. A paper prototype was developed to address these challenges, leveraging AI to foster interaction and improve peer-to-peer communication. These results have broader implications for educational technology, showing how AI could foster deeper student involvement and provide instructors with tools to manage participation effortlessly at scale, improving the overall learning experience. / Thesis / Master of Computer Science (MCS)

The importance of protecting artistic works and promoting the creation of new works has been well established since the inception of United States Constitution. Copyright protections were created by Congress to grant authors exclusive rights over how their works are used and any violation of these rights is copyright infringement. This paper outlines why the use of copyrighted material to train generative artificial intelligence (A.I.) systems is an infringement upon the rights of the author and not an exception under fair use. While no court decisions have been rendered on this legal issue, this paper utilizes previous court decisions in Authors Guild v. Google, Inc. (2015), Harper & Row Publishers, Inc. v. National Enterprises (1985), and Hachette v. Internet Archive (2015) to illustrate how the use of copyrighted material to train generative A.I. systems are an infringement and not an exception under fair use. This is a key point to establish as generative A.I. systems continue to be developed, and artists’ works are being used unfairly and without compensation. Creativity and innovation should be encouraged - but not at the expense of the rights of artists. Therefore, this paper suggests that laws be implemented that protect artists’ work while still allowing space for the development of generative A.I. programs.

As enrollment in Computer Science grows, traditional help-seeking opportunities for students, such as office hours and forums, become less effective due to rising student-to-teaching assistant ratios. To address this issue, research has investigated large language models (LLMs) to provide individualized help to students at scale. However, prior research primarily targets introductory computing courses, does not fully connect LLMs to course material, and does not expose relevant course material to students. As a result, existing approaches do not adapt well to advanced computing courses and limit opportunities for students to develop self-sufficiency. To address this, we present Disdoc, an LLM-based question and answer tool for students in advanced computing courses. Disdoc presents snippets of course material relevant to student questions and generates answers using an LLM. To include course-specific information in answers, we connect the LLM to all course material through retrieval-augmented generation (RAG). To ensure the RAG system retrieves the most relevant information, we organize course material into question categories. We evaluated Disdoc in a research study on a 340-student Computer Systems class at Virginia Tech, where we tracked student reviews, activity, and exit survey responses. Students indicated that Disdoc was helpful, particularly for questions about course assignments. Usage data revealed that students strongly preferred to see LLM-generated answers and rarely clicked on outgoing links, suggesting they were satisfied with the LLM-generated answers and snippets of relevant course material. / Master of Science / As enrollment in Computer Science grows, traditional ways for students to seek help, such as office hours and forums, become less effective due to rising student-to-teaching assistant ratios. To address this issue, research has investigated large language models (LLMs), a type of AI that generates responses to text-based prompts. Past research has used LLMs to provide individualized help to students, but has primarily focused on introductory computing courses and has not fully integrated course-specific materials. As a result, existing approaches do not adapt well to advanced computing courses and limit opportunities for students to develop self-sufficiency. To address this, we present Disdoc, an LLM-based question and answer tool for students in advanced computing courses. Disdoc presents snippets of course material relevant to student questions and generates answers using an LLM. To incorporate course-specific information in generated answers, the LLM references course materials through a process called retrieval-augmented generation (RAG). To ensure the RAG system retrieves the most relevant information, we organize course material into question categories. We evaluated Disdoc in a research study on a 340-student Computer Systems class at Virginia Tech, where we tracked student reviews, activity, and exit survey responses. Students indicated that Disdoc was helpful, particularly for questions about course assignments. Usage data revealed that students strongly preferred to see LLM-generated answers and rarely clicked on outgoing links, suggesting they were satisfied with the LLM-generated answers and snippets of relevant course material.

In the realm of Artificial Intelligence, Large Language Models (LLMs) have recently emerged as a new technology, rapidly gaining prominence across various domains due to their impressive capabilities. This paper investigates key concepts, potentials and obstacles associated with integrating LLMs into the product development process. The initial focus lies on clarifying the underlying mechanisms and capabilities of LLMs to provide a clear and practical understanding. Building upon this foundation, the exploration shifts to the potential applications of LLMs in product development. An assessment matrix evaluates the capabilities of LLMs with regards to engineering challenges, highlighting how these models could potentially improve key aspects of the development process. Additionally, the obstacles associated with implementation in a product development context are addressed.

This paper analyzes the current possible implementations and limitations of generative AI applications such as Chat-GPT, DALL-E, and Midjourney. The applications were used in a specific scenario to gauge whether it was able to effectively handle a potential request from a client, the scenario was to create a visual identity for a shoe company called WalkWise. The creations are then analyzed using Gestalt theories of perception and the machine-learning mechanisms that run these applications. To understand just how graphic designers may introduce these tools into their process, a process chart describing a typical graphic design process for a project has been created using data gathered from 8 professionals in the field who were interviewed. Using a thematic analysis, common occurring themes/activities were found and visualized in a process chart. The process was later analyzed using theories in process value analysis. Using all this information a conclusion was made that AI-generated art has several limitations that inhibit it from completely replacing human designers. These included: not understanding/generating vector graphics, not understanding objects in 3D space from less natural angles, often generating visual clichés, some potential copyright issues, and not being able to generate words. The implementation was therefore limited to a visual brainstorming tool which could aid graphic designers in quickly visualizing an idea or visualizing several different versions of one idea without having to sketch these differences, thereby making the idea-generating parts of the process more efficient.

[pt] No contexto de comércio eletrônico, descrições de produtos exercem grande influência na experiência de compra. Descrições bem feitas devem idealmente informar um potencial consumidor sobre detalhes relevantes do produto, esclarecendo potenciais dúvidas e facilitando a compra. Gerar boas descrições, entretanto, é uma atividade custosa, que tradicionalmente exige esforço humano. Ao mesmo tempo, existe uma grande quantidade de produtos sendo lançados a cada dia. Nesse contexto, este trabalho apresenta uma nova metodologia para a geração automatizada de descrições de produtos, usando as avaliações deixadas por usuários como fonte de informações. O método proposto é composto por três etapas: (i) a extração de sentenças adequadas para uma descrição a partir das avaliações (ii) a seleção de sentenças dentre as candidatas (iii) a geração da descrição de produto a partir das sentenças selecionadas usando um Large Language Model (LLM) de forma zero-shot. Avaliamos a qualidade das descrições geradas pelo nosso método comparando-as com descrições de produto reais postadas pelos próprios anunciantes. Nessa avaliação, contamos com a colaboração de 30 avaliadores, e verificamos que nossas descrições são preferidas mais vezes do que as descrições originais, sendo consideradas mais informativas, legíveis e relevantes. Além disso, nessa mesma avaliação replicamos um método da literatura recente e executamos um teste estatístico comparando seus resultados com o nosso método, e dessa comparação verificamos que nosso método gera descrições mais informativas e preferidas no geral. / [en] In the context of e-commerce, product descriptions have a great influence on the shopping experience. Well-made descriptions should ideally inform a potential consumer about relevant product details, clarifying potential doubt sand facilitating the purchase. Generating good descriptions, however, is a costly activity, which traditionally requires human effort. At the same time, there are a large number of products being launched every day. In this context, this work presents a new methodology for the automated generation of product descriptions, using reviews left by users as a source of information. The proposed method consists of three steps: (i) the extraction of suitable sentences for a description from the reviews (ii) the selection of sentences among the candidates (iii) the generation of the product description from the selected sentences using a Large Language Model (LLM) in a zero-shot way. We evaluate the quality of descriptions generated by our method by comparing them to real product descriptions posted by sellers themselves. In this evaluation, we had the collaboration of 30 evaluators, and we verified that our descriptions are preferred more often than the original descriptions, being considered more informative, readable and relevant. Furthermore, in this same evaluation we replicated a method from recent literature and performed a statistical test comparing its results with our method, and from this comparison we verified that our method generates more informative and preferred descriptions overall.