Poster Presentation Abstracts

Explore the poster abstracts below, and be sure to visit each display to discover more in-depth insights!

Poster 1:

Development of Application-Specific Large Language Models to Facilitate Research Ethics Review

Sebastian Porsdam Mann, Joel Jiehao Seah, Stephen R. Latham, Julian Savulescu, Mateo Aboy and Brian D. Earp

Abstract

Institutional review boards (IRBs) play a crucial role in ensuring the ethical conduct of human subjects research, but face challenges including inconsistency, delays, and inefficiencies. We propose the development and implementation of application-specific large language models (LLMs) to facilitate IRB review processes. These IRB-specific LLMs would be fine-tuned on IRB-specific literature and institutional datasets, and equipped with retrieval capabilities to access up-to-date, context-relevant information. We outline potential applications, including pre-review screening, preliminary analysis, consistency checking, and decision support. While addressing concerns about accuracy, context sensitivity, and human oversight, we acknowledge remaining challenges such as over-reliance on AI and the need for transparency. By enhancing the efficiency and quality of ethical review while maintaining human judgment in critical decisions, IRB-specific LLMs offer a promising tool to improve research oversight. We call for pilot studies to evaluate the feasibility and impact of this approach.

Poster 2:

Bridging the Gap in Academic Research: A Multimodal System for Querying and Understanding Scholarly Texts

Shubham Jariwala

Abstract

Academic research is a complex and time-intensive process that involves extensive literature reviews, data analysis, and synthesizing insights across multiple disciplines. Traditional research methods rely heavily on manual effort, leading to inefficiencies in retrieving information, summarizing documents, and managing citations. With advancements in artificial intelligence, multimodal AI assistants—integrating text, speech, images, and structured data processing—offer a transformative solution to these challenges. This paper explores the potential of multimodal AI in enhancing research efficiency and effectiveness. Leveraging cutting-edge Natural Language Processing (NLP), speech recognition, computer vision, and knowledge graphs, these systems assist researchers in multiple stages of their workflow. By processing diverse inputs such as research papers, audio lectures, handwritten notes, and tabular data, multimodal AI enables seamless interaction and accelerated knowledge acquisition. We propose an intelligent research assistant that integrates key capabilities, including automatic document summarization, real-time speech-to-text transcription, research question answering, semantic search, visual content extraction, and automated citation generation. The system enhances literature reviews by providing concise summaries, highlighting key insights, and suggesting relevant studies based on user queries. Additionally, speech-enabled interactions allow researchers to dictate notes, retrieve contextual information, and engage in interactive question-answering, making the research process more intuitive and efficient. A comparative analysis between traditional and AI-assisted research workflows demonstrates significant improvements in efficiency, accuracy, and user engagement. By streamlining literature reviews and information synthesis, multimodal AI allows researchers to focus on critical thinking and innovation. Furthermore, these assistants enhance accessibility for multilingual and visually impaired researchers, democratizing knowledge acquisition. This paper also discusses ethical considerations such as bias in AI training data, misinformation risks, and dependency on proprietary models. Future research should focus on improving contextual understanding, integrating domain-specific expertise, and ensuring greater transparency. Ultimately, multimodal AI assistants have the potential to revolutionize academic research, fostering a more efficient, inclusive, and intelligent research ecosystem.

Poster 3:

Understanding AI Acceptance Anxiety: A Study of Psychological Effects of GenAI tool Usage Among NTU Undergraduate Students

Qu Yao & Loo Hui En

Abstract

The integration of Generative AI (GenAI) in higher education raises ethical concerns about academic integrity, authorship, and transparency. As AI shifts from information retrieval (Google Effect) to content co-creation (GenAI Effect), the boundaries between human and machine contributions become increasingly blurred, challenging traditional academic ethics frameworks. Key questions emerge, such as how to define plagiarism in AI-assisted work and who holds cognitive authority—the student or the AI. While institutions encourage students to disclose their use of AI, many hesitate to do so. Existing adoption models like the Technology Acceptance Model (TAM) and UTAUT2 do not account for ethical considerations such as fairness and transparency. Although emerging research on AI ghostwriting, institutional ambiguity, and psychological discomfort may provide explanations to non-disclosure behaviour, empirical evidence remains limited. This study explores the psychological and disciplinary factors influencing students' ethical decision-making regarding GenAI disclosure. Specifically, it examines why students avoid truthful disclosure despite institutional guidelines and how disciplinary norms shape their disclosure behaviors. Using an online survey distributed via snowball sampling, we will collect responses from 400 undergraduates at Nanyang Technological University (NTU). The survey will assess students’ academic backgrounds, GenAI usage, ethical perceptions, institutional guidelines, and emotional responses. Quantitative analysis will include descriptive statistics, chi-square tests, t-tests, and regression modelling to identify predictors of disclosure behaviours, while qualitative responses will undergo thematic analysis to uncover underlying motivations or concerns. The findings will contribute to both theoretical and practical discussions on GenAI ethics in higher education. The study aims to refine TAM by incorporating psychological barriers to GenAI adoption and provide evidence-based recommendations for discipline-specific AI policies. By addressing the ethical and psychological dimensions of AI use, this research will help institutions develop policies that balance technological innovation with academic integrity, ensuring transparency while supporting student well-being.

Poster 4:

Measuring the Singaporean Mental Lexicon: Lexical-Semantic Norms for Singapore English Words

Cynthia S. Q. Siew

Abstract

The goal of this project is to develop an integrated database of semantic, affective, and word association norms for a large set of Singapore English words. It is motivated by the observation that lexical-semantic word norms in psycholinguistic research are almost exclusively developed for dominant dialects of English, such as North American and British English. This raises questions about whether such linguistic norms capture and represent cultural and linguistic variation across people who speak diverse, non-dominant English dialects such as Singapore English. The data collected in this project will be made openly available so that the data can benefit local researchers who are interested in creating culturally relevant AI and computational approaches and models for linguistic and sentiment analysis of text produced by Singaporeans. By embracing open science practices and promoting citizen science initiatives, my work contributes significantly to advancing cross-cultural psycholinguistic research and promoting open science principles within the social sciences.

Poster 5:

Can understanding and empowering gender identities tackle job burnout while protecting the confidentiality and safety of vulnerable Singaporean workers?

Goh Zi An Galvyn

Abstract

Singaporean workers tend to experience a high prevalence and persistence of job burnout in a diverse range of professions, such as dentists, educators and healthcare workers (Aurora & Knight, 2022; Chen et al., 2023; Smith & Leng, 2003; Yang et al., 2015). At the same time, persons of minority gender identities reported having higher levels of burnout, even after adjusting for age, race, and ethnicity (Samuels et al., 2021; Watson, 2024). To the author’s best knowledge, this has not been investigated in the Singaporean context. This is important to study in Singapore due to the emergence and growth in diverse gender identities in Singapore (Chan et al., 2019).

There is a need to help workers/teams/leaders/organizations/policy makers understand the specific challenges of individual workers while protecting the person’s privacy and safety (Mone et al., 2018). A worker may not be willing to share his/her challenges if there is a risk that he/she can be identified and penalized (Detert & Edmondson, 2007). Persons with minority gender identities may face more challenges and discrimination at the workplace in specific job tasks, this can lead them to feel disengaged/exhausted/cynical, which then leads to burnout (Samuels et al., 2021; Watson, 2024. The research questions for this poster are: (1) Can we develop ways to identify burnout in workers in relation to gender identity?, (2) Can this be done in a way that enables feedback regarding burnout and gender identity from workers to bosses in a way that protects the workers’ privacy, and (3) how feedback given from workers to bosses can influence burn out management?

Previously, Goh & Poon (2022) carried out data collection by interviewing about 30 workers in white-collar and blue-collar occupations in Singapore using the Functional Job Analysis (FJA) using the framework of the Job Demands-Resources Model (Bakker at al., 2004; Bakker & Demerouti, 2007). The FJA is an approach that involves having workers rate their job tasks (work activities) based on how engaging or exhausting they are using a Likert scale of 1 (Strongly Disagree) to 7 (Strongly Agree). The tasks are taken from global and national task databases (ONET, 2025; SingStat; 2025). For the data analysis, the data was then converted into a one-page visualization for viewing by different stakeholders to address key pain points. The workers are also interviewed regarding their wellbeing. Thematic analysis was conducted to identify themes relating to worker wellbeing with reference to the Job Demands-Resources model (Bakker at al., 2004; Bakker & Demerouti, 2007) as the prior research was focused on worker wellbeing (Goh & Poon, 2022).

This present research builds on Goh & Poon (2022) to re-conceptualize how to apply this same method to investigate burnout in relation to gender identity issues at the granular level of tasks for future research. This lays the groundwork for future research to use this re-conceptualized improved method to communicate specific challenges a worker faces at the granular level of the job tasks to higher management in an aggregate fashion that protects the specific worker’s identity and safety. Importantly, this includes enabling persons with minority gender identities who may face gender identity discrimination at the workplace to give feedback while protecting their safety, privacy and confidentiality, and this can help address their burnout in relation to their job tasks in the long term.

Poster 6:

A degrading academic teaching and research culture: failures, struggles and reflections from an auto-ethnography study

Sohail Sanghab & Arlindo Silva

Abstract

Doctoral education in engineering is suffering from a spiraling feedback loop of publishing pressures, low-quality research, and lack of teaching rewards. First, faculty prioritizing research over undergraduate education has weakened the foundational skills of future graduate students and researchers. Second, this skill gap, combined with publication pressures, then pushes academics to use doctoral students as glorified technicians. Finally, the continued decline in literature quality (following from the previous two) results in an incomplete conceptual picture for doctoral students, increasing the risk of research ideas disconnected from societal value, ethics, and safety. Yet, graduation requirements force doctoral students to publish (eventually) mediocre results, thus perpetuating the cycle of declining literature quality.

A few global surveys, interviews, and reflective accounts corroborate aspects of this narrative. However, a thorough investigation of this situation requires deep ethnographic studies of the engineering research culture, with eventual consequences for the authors.

Lacking such data, we present an inward-looking autoethnographic and systems thinking perspective to situate ourselves within the academic complexity as beneficiaries, victims, perpetuators, and opposers. We inquire about our histories of successful and failed projects hoping to untangle our relation with impactful engineering work from struggles with the perceptions of academic success. Despite repeated failed attempts at materializing change within our systems, we question our motivations to continue being academics and recognize our acts of resistance against widespread neglectful teaching, degrading research methods, and sub-standard outputs that feed meaningless success metrics and fuel unrewarding career progress.

Here, we want to build a reflective space for individuals affected by neoliberal academic practices (that over-emphasize entrepreneurship and fast-paced publication/citation) to consider our collective role in perpetuating these norms. Our goal is to start a dialogue on altering our relationship with research, teaching, and the literature we produce, thus reshaping the conditions of our work.

Poster 7:

Implementation of a “Fast Lane” Approach for Data Transfer at NUHS

Daryl Tay, Stephanie Ruth Teo, Joshua Chan, Adrian Thian, Ke Wei Tan, Jiun-Yih Lin, Anushia Panchalingham, Valerie Foo, Jessamine Geraldine Goh, Yasotha Narendran, Bernard Nguang, Li Hui Goh, Sankari Ramanathan and Elizabeth Huiwen Tham

Abstract

In the rapidly evolving landscape of healthcare research, efficient data management and transfer processes are crucial for timely and impactful outcomes. National University Health System (NUHS) has implemented a "Fast Lane" approach, to streamline the transfer of data out of NUHS entities, significantly reducing approval times and enhancing efficiency.

The management of healthcare data needs to adhere to data protection laws, maintain patient privacy, and safeguard institutional reputation. Due to these requirements, traditional data transfer processes often involve lengthy approval procedures, potentially hindering research progress. The lead time for approvals typically takes over 6 months, and involve a large number of offices within the institution. To address this challenge, NUHS developed the "Fast Lane" approach, aiming to reduce the approval time for data transfers to just two weeks.

The "Fast Lane" approach is risked based, and allows low risk data transfers involving mainly data with minimal risks of re-identification to receive appropriate approvals within an expedited timeframe. This approach represents a significant step towards balancing the need for robust data protection with the demands of efficient research. By reducing bureaucratic hurdles while maintaining stringent privacy standards, NUHS has created a model to accelerate research without compromising data security. This enables NUHS to collaborate effectively and efficiently with global partners allowing us to bring innovative healthcare solutions to our patients.

Poster 8:

A Guide to AI Writing Tools: Opportunities and Challenges

Kevin Chong, Foo Ngee Chih and Ng Heok Hee

Abstract

Artificial Intelligence (AI) writing tools have significantly enhanced scientific writing by improving efficiency, precision, and clarity. Leveraging machine learning, natural language processing, and data analysis, these tools assist researchers in effectively communicating their findings. However, AI writing tools present both opportunities and challenges, necessitating careful consideration of their impact on research integrity.

This review examines AI writing tools in two primary modes: assistive and generative. In the assistive mode, these tools analyse text, identify key concepts, and provide suggestions to refine vocabulary, sentence structure, and coherence. In the generative mode, they facilitate the drafting process by creating outlines and producing initial content based on contextual parameters, thereby enhancing productivity.

Despite these advantages, AI writing tools raise concerns related to over-reliance on automation, potential biases, and ethical issues surrounding authorship and originality. Their use must be carefully managed to uphold academic integrity and ensure that AI-generated content aligns with rigorous scientific standards. By understanding both the benefits and limitations of AI writing tools, researchers can adopt responsible practices that enhance scientific communication while maintaining the highest standards of integrity and ethical authorship.

Poster 9:

Using Risk Assessments to evaluate (or is it identify) potential risks in Investigator Initiated Clinical Trials (IICTs) – the NUHS approach

Anushia Panchalingham, Jessamine Goh, Goh Lihui, Valerie Foo, Adrian Thian, Daryl Tay, Yasotha Narendran, Tan Ke Wei, Jeannie Lin, Stephanie Ruth Teo, Sankari Ramanathan and Elizabeth Tham

Abstract

With the increase in the number of Investigator Initiated Clinical Trials (IITs) in the National University Health Systems (NUHS), there is a greater burden on the institution to ensure that the trials conducted are of the highest ethical standards. There needs to be a robust monitoring and risk-mitigating steps in place to ensure that the participants’ safety and autonomy are ensured along with compliance of the protocol to regulatory, ethical, and institutional requirements.

The NUHS Research Office (RO), uses a Risk Assessment (RA) framework, in addition, the assessment aids NUHS RO to evaluate the impact on the participant’s safety, trial data integrity and if the proposed protocol is compliant with requirements.

Using an internally developed risk impact assessment matrix, NUHS RO can understand the complexity of the proposed IITs, identify and assess potential risks, and provide recommendations. By determining the overall trial risk (low, medium, or high), recommendations may be proposed to mitigate the identified risks. One such recommendation could be the formation of a Data Safety Monitoring Board (DSMB) to evaluate and manage the safety events and trial conduct. Another recommendation could be to conduct institutional monitoring, to ensure the study’s adherence to proper consenting measures, protecting participant confidentiality, minimizing risks to participants, and maintaining transparency in trial conduct.

The use of a RA framework provides NUHS as a sponsor institution a structured methodology to identify risks and reduce them to an acceptable level. The risk-stratified approach enables the institution to undertake a larger number of IITs and direct resources appropriately to ensure that there is pertinent governance and discharge of sponsor responsibilities.

Poster 10:

Optimizing Oversight and Sponsor Management of Investigator-Initiated Trials in a Hospital Setting

Adrian Thian, Daryl Tay, Ke Wei Tan, Jiun-Yih Lin, Stephanie Ruth Teo, Anushia Panchalingam, Valerie Foo, Jessamine Geraldine Goh, Yasotha Narendran, Li Hui Goh, Sankari Ramanathan and Elizabeth Huiwen Tham

Abstract

Effective oversight of investigator-initiated trials (IITs) within hospital institutions is critical to ensuring patient safety, regulatory compliance, and high-quality data integrity. National University Hospital (NUH) plays a key role in conducting and managing multiple IITs in collaboration with investigators and partners. Robust oversight and structured sponsor management are essential to streamline trial execution, mitigate risks, and enhance research outcomes.

NUH has implemented a structured framework to oversee IITs, ensuring adherence to Good Clinical Practice (GCP) and institutional ethics guidelines. This includes a centralized Research Office (RO) to monitor trial activities, facilitate regulatory submissions, and conduct internal audits. Additionally, we have established a collaborative approach in management of IITs, incorporating regular communication and education outreach, predefined responsibilities, and risk-based monitoring strategies to improve trial efficiency and compliance.

Our structured oversight has led to improved compliance with regulatory standards, and enhanced participant safety in ongoing clinical trials. Active engagements with investigators and study teams have also resulted in faster resolution of trial-related events, increased protocol adherence, and better alignment with study objectives. The implementation of risk-based monitoring has allowed efficient resource allocation while maintaining high-quality data integrity.

Building on these successes, NUHS aims to further optimize hospital oversight through digital trial management tools, such as automated reporting systems, together with enhanced training programs for investigators and site staff. Strengthening partnerships will drive more efficient and patient-centric research. Continued collaboration with stakeholders including pharmaceutical companies, regulators, and public hospitals will further position NUH as a leading hub for high-quality clinical trials in Singapore.

Poster 11:

Ensuring Ethical Excellence: A Robust Framework for Research Integrity in Universities

Lai Chunying, Lee Terence, Stanislaws Anna and Koh Willie

Abstract

Nanyang Technological University, Singapore (NTU), a research-intensive public university, houses a vibrant community of over 5,000 faculty, research staff, and graduate students engaged in diverse projects across various disciplines. Maintaining research integrity is integral to NTU's pursuit of excellence, ensuring the highest ethical standards in all research endeavours. This poster introduces a tailored framework designed for NTU, aiming to educate researchers on the importance of research integrity and promote responsible research practices.

To foster a culture of honesty, transparency, and accountability in research, NTU's Research Integrity framework focuses on three primary areas: Policy, Training, and Continuous Learning.

NTU’s Research Integrity Policy serves as the foundation, guiding faculty, staff and students in responsible research conduct and addressing allegations of research misconduct.

Raising awareness and understanding of research integrity within the NTU community involves implementing specialised educational programmes for individuals at different levels. Initiatives include a Research Integrity Declaration for new members and collaborations with Epiguem to develop online courses for faculty, research staff, and students. By equipping researchers with the knowledge and resources to navigate ethical challenges, the university empowers them to conduct research with integrity and confidence.

Learning about research integrity is an ongoing process, requiring continuous education and staying abreast of evolving trends. As such, the framework also prioritises regular educational activities such as research integrity conferences, workshops, and bulletins.

In conclusion, the Research Integrity framework in NTU has helped to cultivate a culture of ethical conduct and research integrity within its research communities.

Poster 12:

From Process to Culture: Building Research Integrity Through Design-Driven Compliance

Yu Lan and Wan Kah Fei

Abstract

In an increasingly complex and technology-driven research environment, upholding research integrity requires more than reactive compliance, it calls for systems and practices that enable researchers to do the right thing from the very start.

At EDDC, we’ve adopted a practical approach to build a responsible research culture. Our efforts focus on simplifying SOPs for research application, documentation, and data/sample use; tailoring modular training to specific research roles; and embedding risk-based oversight throughout the research lifecycle. These changes, though simple, made a meaningful impact.

We’ve learned that integrity doesn’t begin at audit, it begins with how we design the process. When systems are clear and supportive, compliance becomes intuitive. A checklist, a walkthrough, or a brief consult at the right moment can build confidence, reduce avoidable lapses, and foster trust. Researchers often view integrity as an obligation, and we reframe it as a shared responsibility, something embedded in the day-to-day, not imposed from the outside.

Looking ahead, we recognise that tools like AI, automation, and large-scale data use are starting to raise new questions for researchers around authorship and data handling. While these areas are still evolving, we see value in fostering awareness and creating space for open conversations, so teams can begin reflecting on these challenges in a practical and thoughtful way.

As Singapore transitions toward RIE2030 and a more innovation-driven research ecosystem, we believe that sustainable research integrity must be designed, not demanded. It is built, step by step, through thoughtful processes, inclusive training, and a shared commitment in doing what’s right.

Poster 13:

Leveraging Technology to Ensure Compliance Through Institutional Systems

Stanislaws Anna, Tan Ying Shi, Salim Muhammad Helmi, Lee Terence, Koh Willie

Abstract

Singapore’s NACLAR (National Advisory Committee for Laboratory Animal Research) Guidelines establish the foundation for the ethical use of animals in scientific research. The implementation of the 2nd Edition Guidelines in March 2023 introduced stringent new requirements on animal users to ensure continued competency for the ethical and responsible conduct of animal work. The new requirements specify a (i) mandatory theory refresher training every 5 years, and (ii) hands-on (practical) competency assessment every 2 years.

These new requirements present significant administrative challenges and necessitates a robust tracking framework to ensure compliance. There are at least 40 different common procedures performed on animals, which includes different anaesthesia, euthanasia and surgical techniques, which can differ from small to large animals. Nanyang Technological University, Singapore (NTU), is also a research-intensive institution with over 500 animal researchers. Given the complexity of monitoring multiple competencies per researcher, traditional tracking methods such as use of spreadsheets are inefficient, error-prone, and administratively burdensome. To address this, NTU has integrated technology-driven solutions via the use of the Ethics Review Management Portal (ERMP) and ServiceNow platforms to streamline compliance tracking.

The leveraging of these digital platforms enables the automation of this tracking process, seamless continuity through automated reminders, proper documentation, and continued monitoring of recertification requirements, thereby significantly reducing manual intervention and improving data accuracy. Furthermore, these platforms support report generation, in turn facilitating proactive training gap identifications and strategic resource allocation.

The use of digital platforms helps to ensure the competencies of animal users are kept up to date, thus ensuring the ethical and responsible conduct of animal research and reinforcing NTU’s commitment to research integrity. These also reduce administrative burden and reinforces a culture of accountability and ethical rigour.

Poster 14:

Strategies to Uphold Good Practices in Ethics Compliance at A*STAR Skin Research Labs: An Institutional Commitment to Research Excellence

Maria del Mar Alvarez, Sharon Chiang, Maureen Chia, Sophie Bellanger, Dinish U.S, Li Lin, Carine Bonnard and Thomas Dawson

Abstract

Ensuring ethics compliance is central to upholding research integrity and promoting institutional excellence. The A*STAR Skin Research Labs (A*STAR SRL) HBR Committee has implemented a set of strategies, grouped into four key categories, to reinforce institutional compliance and promote a culture of responsible research.

Standard Operating Procedures (SOPs): Clear SOPs have been developed and tailored to the institutional research context, in alignment with national and international guidelines. These provide researchers with a clear framework for ethical and consistent research conduct.

Internal Training and Staff Updates: All A*STAR SRL staff receive internal training, complemented by internal resources and regular briefings on good research practices. A*STAR SRL HBR Committee ensures timely communication of new institutional policies, regulatory changes and best practices through PI meetings, Town Halls and email communications.

Support Structure. Key members of the A*STAR SRL HBR Committee have been designated to assist researchers during the IRB application and submission process, provide guidance on research compliance matters and monitor research practices. Acting as liaisons with A*STAR HBR Office, the Committee streamlines and supports researchers through the ethics review journey. The Committee meets regularly to discuss emerging issues related to research ethics and compliance, and to offer coordinated support to researchers.

Internal Monitoring. An internal monitoring system has been implemented to review projects at different stages, with a focus on higher risk HBR studies. This helps to identify potential gaps, risks, or non-compliance issues, supports early intervention and promotes continuous improvement in research practices.

The A*STAR SRL HBR Committee aims to upholds research excellence through strong ethics governance and proactive support for the research community.

Poster 15:

Promotes IRB/HBRA awareness & compliance via Chatbot

Tham Yi Chuey, Cheryl Teo Wei Ling, Yeoh Wooi Gan, Gunady Ng, Lee Yu Kai, Janakiraman Prashanth, Zheng Tan Kye, Jiang Ridong and Quek Ming Kai Elton

Abstract

The challenges and our approach

To effectively build awareness on ethics & compliance in IRB/HBR regulated research, we look at the challenges faced by our researchers on getting timely, relevant and useful snippets of information on related processes, policies and guidelines. Often researchers feel overwhelmed by the vast information available on ethics, leading to uncertainty in compliance of certain policies.

We provide time-saving and easily accessible information for researchers anytime anywhere, via a chatbot. We have deployed the chatbot in May 2024, for access by all I²R staff.

Method

How does the chatbot work?

IRB/HBR knowledge base consists of question-answer pairs. It consists of deconstructed versions of the guides and policies, repackaged into bite-sized pieces, to deliver the right amount of information to address the queries. This helps to improve retention and understanding.

To provide a wholesome learning experience, we have included links to more in-depth online materials (where applicable) in the chatbot answers that users can access for additional reading.

The chatbot uses Natural Language Processing (NLP) to:

• Understand the questions and identify the correct response.

• Match to the closest approximate answers.

• Allow users to provide instant feedback with simple thumbs up/down or via online feedback to improve the NLP engine.

We monitor users’ feedback and usage patterns to improve on the knowledge base (e.g. analysis of response relevance and accuracy, analysis of nil responses provided by the chatbot, that guide us to design more targeted and comprehensive question-answer pairs).

Results

Easier & faster way to get information on IRB/HBR-related information.

Reduce admin overheads for I²R HBRO on answering simple queries.

Acknowledgement

The chatbot is developed by I²R’s Integrated Digital Product Centre and Aural & Language Intelligence teams.

Poster 16:

Promotes IRB/HBRA awareness & compliance via Chatbot

Jing Shi

Abstract

The integration of artificial intelligence (AI) into mental health research presents unprecedented opportunities for advancing diagnosis, prediction, and personalized interventions. However, it also introduces critical ethical complexities that demand nuanced consideration beyond general AI ethics frameworks. Increasingly, concerns are raised that AI may inadvertently reinforce hegemonic norms, exacerbate epistemic injustices, and further marginalize already vulnerable populations. Therefore, this poster explores how biases in research design, dominant discourse, and hegemonic norms in media and science risk entrenching privilege and diminishing non-normative experiences of mental illness when relying on AI models.

Mental health data are shaped by social, cultural, and clinical norms that inform our decision-making. Such data are often over-represented by populations who are engaged with clinical and research systems; data under-represent people who face systemic barriers, stigma, and other reasons for their exclusion. With only 1-4% of individuals who face mental health issues and addictions seeking formal help, this under-represented population is a vast majority of experiences and perspectives that are excluded from training AI systems.

AI systems trained on dominant narratives and often Western, English language ideologies raise the risk of misclassification, misrepresentation, and perpetuation of misinformation through AI outputs that would further reinforce dominant discourses, diagnostic inaccuracies, structural inequalities and power asymmetries. The lived experiences of unheard voices would be further diminished.

Researchers in mental health research should critically consider: 1) developing ethically grounded mental health methodologies, such as participatory-action designs, to generate data that could be used to highlight marginalized voices; 2) establishing dynamic consent when studying mental illness that involves transparency with how data are being interpreted, allowing participants to withdraw consent at any time; 3) validation of cross-cultural applicability and generalizability of AI findings. By being cognizant of AI ethics, researchers can help ensure AI tools in mental health serve to empower these marginalized populations.

Organization of this abstract was assisted by ChatGPT 4o.