Mishcon de Reya has launched a new PhD fellowship with UCL's Department of Computer Science which aims to accelerate the use of artificial intelligence in the legal market.
The Mishcon de Reya Fellowship in AI has been awarded to Yao Lu, whose research will focus on Natural Language Processing and possible applications for the legal sector.
The inaugural Mishcon de Reya fellowship in AI will provide four years of funding to the UK Research and Innovation (UKRI) Centre for Doctoral Training in Foundational AI at UCL. This is a significant extension of the research collaboration between Mishcon de Reya's MDR Research group and UCL's Centre for Artificial Intelligence.
The foundational research will play a key role in MDR Research's mission to expand the use of AI and other technologies in understanding legal reasoning, delivering legal services and improving access to justice and legal information.
MDR Research brings together lawyers, engineers, researchers, entrepreneurs and technologists, who are working side-by-side to advance the frontier of legal systems understanding.
Mishcon de Reya has founded MDR Research to work with academia to explore how recent advances in technology in a rapidly changing landscape can be used for legal applications and to deliver client value.
This includes how data is collected, interpreted and used to support the practice of law, across areas as diverse as AI and machine learning, distributed systems and graph databases.
Mishcon de Reya has also joined UCL's Centre for Artificial Intelligence's industrial steering board, joining companies like Google Deepmind and Cisco Systems to help steward the development of AI in the UK.
Yao's first paper to result from this research has been accepted as part of the 60th Annual Meeting of the Association for Computational Linguistics, to be held in Dublin in May 2022.
His work explores the use of large language models like GPT-3, where there has been rapid progress in recent years, and how to adapt their behaviour by altering the structure of prompts for pre-trained models. Pre-trained language models such as GPT-3 have shown competitive results when compared to fully-supervised, fine-tuned pretrained language models. Yao’s paper demonstrates that the order in which the samples are provided to the model can make the difference between near state-of-the-art and random guess performance and sets out solutions to alleviate the phenomenon.
Dr Alastair Moore, Head of Analytics and Machine Learning at Mishcon de Reya, said: "We are excited to support this fellowship and congratulate Yao on this first paper. This research is an important stepping stone in maximising the potential of technologies like AI to overcome challenges in how the legal sector currently operates. We also hope it will eventually help to remove barriers to open up legal knowledge and access to justice."
Yao Lu said: "I am honoured to receive this fellowship to support my PhD research. Our initial work shows that prompt-based learning in natural language processing has great potential to solve complex tasks which usually require human and manual expertise. The work reveals the order sensitivity issue of current prompt learning approaches, which is crucial for solving NLP tasks. With the support of Mishcon de Reya’s fellowship, I am confident to further explore this direction of research and produce new findings."
Professor Pontus Stenetorp, Deputy Director at the UCL Centre for Artificial Intelligence, said: “UKRI Centres for Doctoral Training are reliant on industry support to fund the very large number of highly talented overseas applicant, such as Yao, who apply each year. Thus we are immensely grateful for Mishcon de Reya’s decision to support the UKRI Centre for Doctoral Training in Foundational AI here at UCL. Their contribution towards foundational research into AI methods is applicable both to the legal domain specifically and to the growth of the AI sector in the UK.”
Over the course of the fellowship, Yao will work closely with Mishcon de Reya to develop new expertise in large language models and further their industrial application.
Paper link on arXiv.