Open research programme extended as EU AI Act comes into force

03 April 2024, Galway: Insight SFI Research Centre for Data Analytics in Galway is scaling up its research partnership with Genesys, a global cloud leader in AI-powered experience orchestration, building on their collaboration around bias detection techniques in machine learning models that began in 2021. Genesys has signed a Phase 3 extension to its contract, committing to working with the Insight SFI Research Centre until 31st June 2025.

The new project; ‘Exploring gaps in Large Language Models for Trustworthy AI’; is focused on building AI systems that are trustworthy and ethical, adhering to the key principles of transparency and accountability. Together, the research teams are working with standardisation bodies to influence global standards in establishing frameworks for explainability and fairness in AI models and datasets.

This project builds on the bias detection research work completed in Phases 1 and 2. Phase 3, which kicked off in January 2024, will explore gaps in Large Language Models (LLMs) including domain-specific translations for under-resourced languages, prompt-based fine-tuning approaches and domain injection approaches.

Genesys and Insight SFI Research Centre have taken an open approach to this programme, as this research can benefit the wider technology sector in numerous ways, including: improving transparency on data and models, increasing awareness of the AI decision risks, contributing towards industry standards, and importantly, guiding AI systems’ adherence to upcoming regulations such as the EU Artificial Intelligence Act, which will come into force in the coming months.

Professor Edward Curry, Director of the Insight SFI Research Centre at University of Galway, added: ‘Building trustworthy AI through the development of systems with explainable processes is a core activity at Insight, Ireland’s national research centre on data and AI. Genesys is a global leader in AI-powered customer and employee experiences and our research partnership comes at a critical time as we move to a more regulated AI space. The open innovation approach of this research is of particular value to both partners, as we seek to support the wider mission of building more accountable AI systems in Europe and beyond.’

Project co-lead Dr Ihsan Ullah commented: “Bias exists at various stages of a machine learning pipeline. LLMs are trained on gigantic amounts of data from the web, hence, necessitating an investigation into potential biases affecting their decisions. This project will delve into automation of detection of bias at various levels using explainable AI techniques and our proposed linked dataspace, model, and data card schema. It will help in bringing fairness and standardising the machine learning pipeline to be adopted by academia and industry across Ireland and the EU.”

Joe Smyth, SVP of R&D, Digital and AI at Genesys, said: “Since embarking on this project with the Insight SFI Research Centre for Data Analytics, we’ve made tremendous progress to advance bias detection. As we continue, our ultimate objective is to help organisations ensure the trustworthiness of the LLMs they are using to fuel their AI capabilities and to further apply LLMs to underserved languages, so all cultures can reap the benefits of this revolutionary technology.”

Project co-lead, Andy Donald said, “This represents an exciting opportunity for Insight to contribute to a more transparent and trustworthy AI ecosystem. As we move to a more data-centric global society, it is imperative that we lead the charge on this important work, and we’re excited to collaborate with Genesys on this journey.”

Dr John McCrae project co-lead, added: “In this project we will investigate techniques to reduce the cost of training large language models for new domains and languages, enabling Genesys to further expand the reach of its products”.

The project was shortlisted for the ITAG Excellence Awards in 2023.