Mirella Lapata will teach machines to reason
Mirella Lapata was appointed one of the first Turing AI World-Leading Researcher Fellows to conduct ground-breaking work on Artificial Intelligence’s (AI) biggest challenges. The fellowship comes with £3.9 million from UKRI to set up Edinburgh Laboratory for Integrated Artificial Intelligence (ELIAI). The new research centre will work towards developing a new class of neural network models and a theoretical framework which explains what it means for neural network models to reason.
Problems with deep learning
Deep learning, which aims to mimic the human brain to process information and make decisions, has brought a revolution in the area of artificial intelligence (Al) and led to advances in computer vision, natural language processing, speech recognition, robotics, and clinical decision making. However, many problems at the core of artificial intelligence still need solving. One skill that machines still lack and humans are perfectly capable of is reasoning: combining different types of information from different sources to draw logical conclusions. Building a machine that can read a book or watch a movie, and answer simple questions about the plot and its characters, questions that a child would be able to answer is still beyond the reach of current artificial intelligence.
No brain like human brain
Human brains mastered the art of correlating and integrating different types of information from different sources and re-using previously acquired experience and expertise to transfer it to radically different challenges and domains. Imagine an airplane passenger sitting in an exit row, studying the emergency guide, which is often a combination of images and text. Their brain combines visual and textual information in order to infer the intended message - open the door in the unlikely event of an emergency.
A computer system seeing the same document would first employ an image recognition model to scan the image. An Optical Character Recognition (OCR) system would read the text, and a third system would correlate the image and text to understand the complete picture. Although the fundamental principles of analysing the world around us and the approach a machine takes to process complex information are both based on breaking down the data to its core elements, humans are better at integrating information to draw logical conclusions.
Current AI systems fail when exposed to data outside the information they were trained on, adhering only to superficial and potentially misleading associations instead of learning true causal relationships while also being unable to reason on an abstract level and provide us with a full understanding of how they came to a specific conclusion.
How to teach machines to reason
To address these AI deficiencies Professor Lapata’s project will develop a new class of neural network models. Rather than relying on a monolithic network structure, researchers will assemble a network from a collection of more specialised modules, making use of an explicit, modular reasoning process, which allows for differentiable training (with backpropagation) but without expert supervision of reasoning steps.
The project has the potential to improve machines processing and aggregating large amounts of data from different sources, as well as teaching them new skills such as: making generalisations when pre-training on all possible scenarios is impossible, dealing with changing situations and causality, being creative (e.g. writing poems) and – perhaps most importantly - being able to explain predictions and decisions.
This funding will provide a unique opportunity to build the Edinburgh Laboratory for Integrated Artificial Intelligence (ELIAI), an interdisciplinary research hub, whose aim will be to integrate expertise in different strands of AI within the School of Informatics. We will create an interdisciplinary environment where collaboration between different faculty members and researchers is encouraged, and work on projects that will shape the future of AI. Our first goal will be to develop AI models with reasoning abilities that go beyond pattern matching and show how these improve applications in the fields of machine learning, robotics, computer vision, and natural language processing.
To achieve such ambitious objectives drawing from expertise in different strands of AI will be crucial – and it will transform Al theory and practice. It will bridge a gap between the neural and symbolic views of Al and provide the means for developing a UK skill base in Al, and the fields of machine learning, computer vision, and natural language processing.
This is a wonderful recognition for Mirella who is at the forefront of AI research. The new Edinburgh Laboratory for Integrated Artificial Intelligence will offer new opportunities for her, but also for colleagues and students within the School of Informatics. Moreover, more broadly, the new Laboratory will serve as a beacon within the community, encouraging others to address the exciting challenges for integrative AI.
Currently much AI research is fragmented into sub-domains. The Edinburgh Laboratory for Integrated Artificial Intelligence will give us the opportunity to weave these strands together in order to change the AI landscape, and to build a sustainable lab supported by UKRI, along with commercial partners.
Professor Mirella Lapata, Personal Chair in Natural Language Processing in the School of Informatics, is a Fellow of the Royal Society of Edinburgh, Director of the UKRI Centre for Doctoral Training in Natural Language Processing, holds the Royal Society Wolfson Merit Award and was the first recipient of the Karen Spärck Jones Award.
The core funding for this project comes from the UKRI . Turing AI World-Leading Researcher Fellowship. ARM Ltd, BBC, British Library, Google DeepMind UK, Dyson Limited, Huawei, IBM, Naver Labs Europe, RAS Technologies GMBH, Scottish and Southern Energy SSE plc, Brainnwave Ltd, Wallscope, Amazon Research Cambridge are all partners in this project and will provide further funding.