This thesis explores the effectiveness of three different machine translation (MT) engines—Google Translate, ChatGPT, and Baidu—in translating English phrasal verbs into Mandarin. The study focuses...Show moreThis thesis explores the effectiveness of three different machine translation (MT) engines—Google Translate, ChatGPT, and Baidu—in translating English phrasal verbs into Mandarin. The study focuses on three categories of phrasal verbs: literal, aspectual, and idiomatic. A corpus of sentences containing various phrasal verbs was translated by each MT engine and evaluated for general translation quality using automatic metrics and for the accuracy of phrasal verb translations via human assessment. Results indicate that while improvements have been made in MT, significant challenges remain, particularly with literal phrasal verbs. Among the MT engines, Baidu performed the best, likely due to its extensive training on Mandarin data. However, ChatGPT shows promise, proposing that with advanced prompting and fine-tuning, it could surpass Baidu. This research highlights the current strengths and weaknesses of MT engines in handling phrasal verbs and suggests directions for future improvements in MT technology.Show less
Binnen dit onderzoek is getracht om op een kwalitatieve wijze te onderzoeken of een nationaal algoritmeregister bijdraagt aan de publieke waarden 'transparantie' en 'verantwoording' over de inzet...Show moreBinnen dit onderzoek is getracht om op een kwalitatieve wijze te onderzoeken of een nationaal algoritmeregister bijdraagt aan de publieke waarden 'transparantie' en 'verantwoording' over de inzet van AI en algoritmen door de overheid. Hiertoe is een casestudie uitgevoerd naar het Algoritmeregister van de Nederlandse overheid. Daarbij is gebruik gemaakt van een combinatie van literatuuronderzoek, deskresearch en semigestructureerde interviews met deskundige praktijkmensen uit de publieke sector van Nederland. De bevindingen laten zien dat een nationaal algoritmeregister bijdraagt aan de transparantie en verantwoording over de inzet van AI en algoritmen door de overheid. Niettemin onderstrepen ze ook het belang van een wettelijke verplichting voor publieke organisaties om hun algoritmen écht te publiceren in het algoritmeregister.Show less
This research explored students' trust in ChatGPT, in answering challenging trivia questions on an online multiple-choice questionnaire. Despite extensive research on ChatGPT's capabilities, a...Show moreThis research explored students' trust in ChatGPT, in answering challenging trivia questions on an online multiple-choice questionnaire. Despite extensive research on ChatGPT's capabilities, a significant gap in the literature remains regarding human trust in this AI system, particularly in educational contexts. This study aimed to assess to what extent do university students trust this novel AI, uncovering biases towards or against AI-generated responses. Using a between-subjects design, we manipulated answer source transparency and correctness. Each question presented participants with one human-generated and one AI-generated answer option. The results were analysed through paired samples and independent two-sample t-tests, alongside Ordinary Least Square (OLS) regression models. The findings revealed a distinct preference for human-generated content, which persisted even when answers were incorrect. This indicated a lower level of trust in ChatGPT among our participants, implying that they did not over-rely on AI-generated content. Surprisingly, the trend leaned towards an over-reliance on human-generated responses.Show less
Due to recent advances within the field of machine learning and computing power becoming more readily available, the use of machine learning within the field of psychology has increased. However,...Show moreDue to recent advances within the field of machine learning and computing power becoming more readily available, the use of machine learning within the field of psychology has increased. However, potential remains for greater use of machine learning within the field of psychology. In this study the usability and performance of 3 machine learning models namely K-nearest neighbors, the Random forest, and the Support vector machine algorithms were assessed when predicting gender, marital status, and family size from Big 5 personality measures and the Holland Code Career Test. Repeated cross-validation was combined with grid search to ensure performance measure accuracy and to optimize model accuracy and F1-score. The performances of the 3 models were compared to the performance of logistic regression to assess whether these models could outperform a model regularly used within psychology. The 3 models consistently outperformed the logistic regression under almost all conditions and proved far superior for groups sizes over 500 even outperforming logistic regression by 10 percentage points under some conditions. However, caution was advised due to wide confidence intervals for small group sizes (n ≤ 200). Therefore, a study was proposed with the aim to enhance predictions for small group sizes, focusing on feedforward neural networks, known to be able to capture complex relationships even with limited data. Addressing these aspects could improve the usability of machine learning in psychology settings involving small group sizes.Show less
Is the term ‘hallucination’ as it is used within the context of generative AI merely a metaphor? An investigation how hallucinations arise in the human perceptual system and in modern generative AI
Bachelor thesis | Cultural Anthropology and Development Sociology (BSc)
closed access
This thesis expands on an arguably underexplored interdisciplinary collaboration within both social and data science disciplines, and their potential to reimagine algorithmic systems as more...Show moreThis thesis expands on an arguably underexplored interdisciplinary collaboration within both social and data science disciplines, and their potential to reimagine algorithmic systems as more culturally and human-centered. To do so, this thesis firstly departs from current narratives related to the dichotomy between humans and machines to analyze the novel role social scientists play in this digital reality and future. Secondly, this thesis addresses what an interdisciplinary methodology of social scientists applying computational techniques may look like, as well as helps to answer the part of the research question that deals with “incorporating” the two disciplines. This more technical section focused on Laura K. Nelson’s Computational Grounded Theory which sets out to incorporate computational tools for social scientists wanting to conduct content analysis. Finally, this thesis highlights viable solutions for social science research and more culturally-centered research is working towards creating more human-centered algorithms by exploring possible strategies for cultural interpretations of these existing systems by evaluating metrics of algorithmic systems.Show less
The Korean peninsula knows a long history of book culture, and attention to it has ranged from the earliest known texts to the recently growing interest in modern Korean literature that came with...Show moreThe Korean peninsula knows a long history of book culture, and attention to it has ranged from the earliest known texts to the recently growing interest in modern Korean literature that came with the ‘Korean wave’. Many of the now canonical works find their roots in the colonial era in Korea (1910-1945) and various aspects pertaining to textual production during this era have been researched by scholars. However, an often-overlooked history is that of the printers and printshops in colonial Korea who had a significant influence over book production. With this loss of primary source material directly stemming from these printshops, the only other source indicating the printer of a text is the included colophon. This colophon is often damaged or simply missing in older books, and to this date there is no efficient method to recover this lost information. Thereby preventing any sizeable quantitative study of printshops in colonial Korea. This MA Thesis will examine the possibility of using convolutional neural networks (ConvNet) to identify the printshop of a given text dating to colonial Korea in order to allow large- scale quantitative research into colonial Korean printshops, which has been impossible thus far It will do this through a case study approach that aims to successfully classify books of four colonial Korean printshops, namely, the Hansŏng Tosŏ Chusik Hoesa (漢城圖書株式會社), Taedong inswaeso (大東印刷所), Sinmungwan (新文館), and Chosŏn inswae chusik hoesa (朝鮮印刷株 式會社). The findings here can be applied to a more extensive set of printers, given enough data and time. Therefore, this research is of high importance to the field of Korean history, as it is an essential step in charting the history of colonial Korean printers. Additionally, the benefits gained from this study are also helpful in the field of digital humanities, as this study will not only focus on the production and performance of a model but also include dataset construction and model explainability. The latter is a vital part and often missing in other DH scholarly work related to ConvNets. Hence, this paper is highly relevant to Korean historical research, and its methodology can be used far beyond the context of colonial-era printshops.Show less
With the current trend of moving toward automation (replacing human workers with technology) based on AI technology, developers often highlight their motivation as one of the progress and necessity...Show moreWith the current trend of moving toward automation (replacing human workers with technology) based on AI technology, developers often highlight their motivation as one of the progress and necessity. However, they often forget the social implications of large-scale implementation of these technologies, such as potential discrimination and oppression from automated decision-making systems. This thesis takes the community-based art project, The Feminist Data Set by Caroline Sinders, as a source of interest in how its activism reflects on these concerns and reveals the power structures in data science and networks. The Feminist Data Set essentially recognizes its power in its capacity to act and propose alternative practices to large technology companies. Through the analysis of the text by Sinders about the project and workshops, this thesis tries to review what specific strategies Sinders offers to tackle the issue of biases within AI and how to make networks more democratic. The model that The Feminist Data Set emphasizes actively involves participants, where the education and access to the data about algorithms could benefit social networks and the internet in more democratization and accountability, which could bring society closer to the ideas of social justice within AI-based systems.Show less
The ever-new emerging technological possibilities provide public authorities with new opportunities to improve service provision and interaction with citizens and optimise their administrative...Show moreThe ever-new emerging technological possibilities provide public authorities with new opportunities to improve service provision and interaction with citizens and optimise their administrative processes. Artificial Intelligence (AI) technologies can support public administrations and especially citizen service offices on this journey. Citizen service offices are the local governments' point of contact for citizens’ service provision in Germany. To improve service provision and interaction between the public institutions and citizens, AI-based chatbots have found their ways into citizens service offices. However, this process is certainly not without challenges. Barriers to interoperability refer to challenges that can impact the administrative ability to collaborate across departments and agencies to provide seamless online services to citizens. Regarding the adoption of AI-based chatbots, barriers to interoperability are especially relevant as chatbots need to be able to obtain data from different departments. This study aimed to examine to what extent barriers to interoperability impact the adoption of AI-based chatbots in German citizen service offices. This question has been explored through semi-structured expert interviews conducted with representatives of the German cities Berlin, Hamburg, Heidenheim an der Brenz, and Munich. The results suggest that political and organisational barriers are impacting the adoption of AI-based chatbots greatly. Challenges have been reported but mostly only by one, or a maximum of two, participants and most of these barriers have been overcome in the early stages of the adoption. Legal and regulatory, as well as technological barriers to interoperability, on the other hand, were found to be the most impacting ones regarding the AI-based chatbot adoption in German citizen service offices.Show less
Artificial Intelligence is a sophisticated emerging technology that cybercriminals have increasingly been using maliciously to facilitate their attacks. As the private and public sector are key...Show moreArtificial Intelligence is a sophisticated emerging technology that cybercriminals have increasingly been using maliciously to facilitate their attacks. As the private and public sector are key targets of these attacks, collaboration is called for. This study dissects the divergent perceptions of cybercrime in both sectors which have inhibited cooperation in the past during the growth of the Internet, then aligns these scattered views to build common ground to establish a cooperation for this up-and-coming threat. The societal aim is to avoid repeating previous mistakes encountered at the birth of cybercrime. This thesis asks the following question: To what extent does the perception of risk of malicious use of AI by cybercriminals differ within the public and private sector? After conducting a qualitative analysis of nine semi-structured interviews, the thesis finds that distinct perceptions about cybercrime between sectors is what is at the heart of the stagnation. But neither public or private organisations can solve the issue of the upcoming AI cybercriminal threat and reach desired goals without each other as each sector lacks some elements complemented by the other. The differences found in this research can shape the basis of a cooperation against AI-Cybercrimes between the public and private sector.Show less
This paper considers the morality that comes with the actions of autonomous AI-systems (AAA). AAA are becoming more autonomous and more intelligent every day, and are already part of our daily...Show moreThis paper considers the morality that comes with the actions of autonomous AI-systems (AAA). AAA are becoming more autonomous and more intelligent every day, and are already part of our daily lives in the form of decision support tools, all sorts of apps, or even self-driving cars and that is why they need a moral compass. Currently, implementing a moral compass into AAA is problematic since the current strategies, laws, principles, guidelines and moral codes do not suffice for solving the moral challenges that surround AAA and impact human lives. Two strategies are considered: Top-Down (TD) and Bottom-Up (BU) strategies. TD strategies use explicit rules that are programmed into AAA, and AAA are designed to follow these rules. BU strategies allow AAA to teach themselves the correct behaviour using feedback from the evaluation of their actions. The problems with TD strategies are that situational dilemmas require almost infinite lists of rules which account for every possible factor, and that a shorter set of rules will result in definitions that are too vague to be eligible for implementation in practise. The problems with BU strategies are that they provide limited transparency of the actions and reasoning behind these actions, and that it becomes harder to evaluate AAA’s moral compass. Furthermore, additional problems come with the moral compass of AAA. Understanding how to emulate the relevant human factors in the moral compass of AAA is difficult, and the moral compass of AAA should adapt to the changing and subjective moral compass of humans over time. This paper concludes that, currently, both TD and BU strategies for implementing a moral compass into AAA prove to be problematic, and that proper evaluation of AAA is difficult. Additionally, the moral compass of AAA becomes more problematic because the programmed moral compass differs from human reasoning and actions.Show less
