AI Against Forced and Child Labour: Leveraging Large Language Model and AI in Supply Chains Due Diligence

Professor Dr Ser-Huang Poon, Professor of Finance at Alliance Manchester Business School will be discussing her research on Transparency in the Business Supply Chains.

This project is a collaboration with Quintin Lake (Fifty-Eight & Just Good Work), Philip Jobi Vallavanthra (UoM), Wyatt Siliang Wei (UoM), and Eghbal Rahimikia (UoM).

This talk explores the connection between Generative AI, Large Language Models (LLM), and a specific use case in Supply Chain Due Diligence. We will provide a brief overview of LLM and Generative AI, focusing on the Prompt Engineering technique. This technique is being used as the underlying technology for an Impact Study on the use of a WorkersBot to help resolve issues in global supply chains, which is an important part of Supply Chain Due Diligence.

Many workers who need assistance are unable to read and can only communicate in their native language. To improve their experience and engagement, our team is developing a solution to incorporate voice into AI systems. We are using existing resources for Speech-to-Text (STT) and Text-to-Speech (TTS) conversion. However, we have encountered a challenge with low-resource languages that do not have well-established STT/TTS libraries. To overcome this, we are working on creating a smaller, targeted audio library using a cutting-edge algorithm called Wav2Vec from the UnMute Toolkit, with support from Responsible AI UK.

The proposed WorkersBot project builds upon our previous research in Natural Language Processing (NLP). Specifically, we have conducted machine-topic modelling of corporate social responsibility reports produced by MSCI Europe constituent companies across 15 industrialized countries and multiple languages (Goloshchapova, Poon, Pritchard, and Reed, 2019). Our expertise lies in combining NLP with Machine Learning to create a trained Lexicon/Dictionary using techniques such as LASSO and word2vec word embedding with a neural network. More recently, we have been utilising a transformer-based algorithm.

We have processed a significant amount of data, including over 20,000 modern slavery reports, the entire Dow Jones Newswires database, and two and a half years of social media posts on Reddit WallStreetBets. Some of these contents together with many other business databases has been incorporated into our in-house LLM called FinText (fintext.ai), which has been widely downloaded since its launch, with over 3000 downloads.

Professor Dr Ser-Huang Poon

Dr Poon is internationally renowned for her volatility research which was cited as reference readings on Nobel website. She is the joint recipient of two best paper awards and has written three books and published widely in peer-reviewed journals.

Dr Poon was appointed as a Turing Fellow in 2018 and is an active member in Turing’s Trustworthy Digital Identity Interest Group since then. The UoM team has recently secured two UoM internal grants, viz., the Turing-Manchester Feasibility Project (£12K) and the AMBS Research Support Fund (£8K). These projects were collaborated with Graphcore, a leading British semiconductor company specialising in accelerators for AI and machine learning (ML), and the University of Oxford to create word embedding specialised in business finance and a set of golden standards for testing financial LLMs.

Dr Poon has secured a PhD funding and nine grants and awards in this area of research in the last five years. This includes £86k from the Turing and Gates Foundation for a project on “Resilience in Value Chain and Worker Vulnerability Reduction - Trusted digital identity and payments in the supply chain”, now appearing in the Journal of Data & Policy, £74k from Innovate UK for the invention “Proof of c19 credentials for care homes and post lock-down”, a UMIP Innovator Award, and was nominated as an ASPECT Social Science Innovation Fellow in 2021.

The research team are actively seeking relevant stakeholders. Businesses and NGOs interested in partnerships please contact ser-huang.poon@manchester.

CAIR Thematic Talks Series

CAIR Thematic talks brings you exciting research tackling real-life issues that can be solved by accounting and finance. We bring you the information and encourage you to share a pint with our speaker at the end.

This talk series has been designed to encourage cross-disciplinary collaboration in an informal setting.

This event will be in person only at Alliance MBS, please register via the link.