Simon T. Perrault’s research while affiliated with Singapore University of Technology and Design and other places

... However, existing empirical evaluations of explainable fact-checking are almost exclusively directed at and executed with laypeople from a limited selection of Western countries, rather than expert fact-checkers with diverse and varied contexts and perspectives. One such study indicated that neither feature-attribution nor example-based explanations of automated veracity prediction had an effect on laypeople's perceptions of the veracity of a news story or their intent to share it, but increased their tendency to over-rely on the AI system when it provided incorrect predictions [75]. A separate study also found no effect of example-based explanations on people's accuracy in predicting the veracity of a claim [77]. ...

Reference:

Show Me the Work: Fact-Checkers' Requirements for Explainable Automated Fact-Checking

... As ION hinges on the performance of the detection of fallacies, we describe a technical evaluation of the LLM used for our study following prior work [58]. We did not evaluate the other prompts as most of them were for creative purposes (producing queries and questions) rather than decision-making (classifying fallacies). ...

Reference:

Iffy-Or-Not: Extending the Web to Support the Critical Evaluation of Fallacious Texts

... Design research primarily focuses on studying design processes and artefacts to propose generic methods (e.g., 6-3-5 sketch) and principles (e.g., function sharing) enabling different design approaches such as biomimicry [11], empathic design [12] etc. Research in innovation and management theorises the mechanisms that enable technological artefacts to create value and how organisations and innovation ecosystems co-ordinate to maximise such value [13][14][15]. Building upon the double-diamond process model of design innovation and leveraging the recent developments in data science, deep learning deep generative models etc., Luo [16] propose a model for Data-Driven Innovation (DDI), referred to as the "double-dump" model ( Figure 3.1). ...

Reference:

Data-Driven Innovation for Trustworthy AI

... Recent advances in LLMs have accelerated the development of new tools that assist researchers in tasks such as literature synthesis, idea generation, data analysis, and academic writing. These tools include open-source ones like ScholarQA [3], which helps researchers synthesize literature and identify research gaps and CollabCoder [16], which supports inductive collaborative qualitative analysis. An increasing number of new tools are being introduced. ...

Reference:

Adapting to LLMs: How Insiders and Outsiders Reshape Scientific Knowledge Production

... Given their strong capabilities in natural language generation, reasoning, and image understanding LLMs have become an increasingly popular topic of research in HCI and beyond [2,6,34,89,93]. ...

Reference:

Augmented Body Communicator: Enhancing daily body expression for people with upper limb limitations through LLM and a robotic arm

... Generative AI continues to be integrated into work practices to support knowledge workers' tasks with the goal of improving the quality and delivery time of their work. Large language models (LLMs) 1 can be effectively used to generate documents, code, or summaries, to analyze or transform data, to translate languages, to answer questions, to classify or categorize content, to brainstorm, or to plan and organize activities [12,15,17,29,73,75,79]. They can also be leveraged to evaluate the quality of both AI-generated and human-created content [20,26,42,80]. ...

Reference:

A Case Study Investigating the Role of Generative AI in Quality Evaluations of Epics in Agile Software Development

... Beyond reading, online deliberation also calls for being critical in writing content. Lightweight interface nudges have been shown to facilitate deeper deliberation [70] and reflection [129] to enhance the introspection and quality of arguments in the comments. Such nudges could work in tandem with ION when implemented in social platforms to provide a holistic end-to-end experience in engaging critically with online content. ...

Reference:

Iffy-Or-Not: Extending the Web to Support the Critical Evaluation of Fallacious Texts

... keep pace with rapidly evolving generative models [35,65,92,118]. Parallel efforts have explored the use of content warnings, especially in misinformation contexts [47,77], but disclosure design for AI-generated content (AIG) remains underdeveloped. As Generative AI tools become widespread across users of varying abilities, AIG is now more prevalent, thus disclosure has become both more necessary and more complicated. ...

Reference:

Signals of Provenance: Practices & Challenges of Navigating Indicators in AI-Generated Media for Sighted and Blind Individuals

... They also analyzed that human coders took much longer than GPT [14]. Similarly, Gao et al. developed a web-based tool using GPT-3 to make code suggestions from excerpts and discovered that employing this tool could reduce the workload of individual coding and improve mutual comprehension [25]. In another study, Xiao et al. found that GPT-3 using provided code definition and examples (i.e., few-shot) has better agreement with human experts [68]. ...

Reference:

LATA: A Pilot Study on LLM-Assisted Thematic Analysis of Online Social Network Data Generation Experiences

... Trust involves a certain degree of risk and uncertainty, and people rely on credibility cues to validate their choice of trusted communication sources. These credibility cues include experience, reliability, fairness, bias, and accuracy of the source, among others [26]. ...

Reference:

Influence of Fact-Checking Habits, Trust, and Collectivist Orientation on the Ability to Detect Fake News