AI Ingenuity – Dr. Lisa Amini, Director, IBM Research Cambridge – The Future of Machine Learning and Natural Language Processing in AI-based Products and Structures

Dr. Lisa Amini is the director of IBM Research Cambridge, which includes the MIT-IBM Watson AI Lab.

Watson is a complex question-answering computer system that is capable of providing answers to questions that are directed in natural language; it was developed in IBM’s DeepQA project. Dr. Amini has extensive experience in deep information extraction, representation, and reasoning as it pertains to knowledge, semantic analysis, and decision science. Her past work includes the development of strategies for utilizing science and technology in intelligent urban and environmental systems, specifically centered on sustainability issues, resource management, transportation, and data.

Dr. Amini discusses how the Watson products focus on specific products that can be benefited by machine learning, natural language processing, computer vision, and image analysis. Beyond the interesting gaming aspect that gained Watson much public attention, Amini explains that the AI’s uses for business applications are quite diverse and multitudinous. She gives an insightful overview of the current AI algorithms that exist in the market and how they perform extremely well utilizing large volumes of data, training on that data, learning from patterns and responses, and image recognition, etc. As the AI expert details, the AI often needs volumes of data in order to make assessments or recognition of an image, and the innovation will come from pushing the AI to make inferences without the need for voluminous data. For as human knowledge comes from multiple sources and signals such as reading documents, recognition, talking to other informed individuals, etc.

, their learning is more broadly focused. And this multi-modal learning is essentially where Dr. Amini and her team are seeking to make improvements with AI technology.

The AI innovator discusses how her team is using advanced information theory techniques to understand how information flows within networks. By studying how models are trained regarding the neural network, researchers can build more scalable models that may provide better output. She explains that one valuable element of natural language processing is the analyzation of text, to understand the importance of relationships in order to bring that information to the user in a more qualitative manner.

Dr. Amini speaks about the questions that arise in human understanding, and how it is important to know ‘what causes what’ in terms of gathering a broader grasp of information. She discusses her hopes to see advances in machine learning applied to areas that would enable us to learn representations of causal structure and causal information, as well as in areas of security in order to provide more robust solutions. From healthcare to security and everything in between, Dr. Amini sees great potential for AI advances that will improve the ways we receive, use, and implement information.

Podcast: Play in new window | Download | Embed