Artificial intelligence at Thomson Reuters

In vertical industries, AI-powered applications often require three key ingredients: data, subject matter expertise, and technology.

• Data or Content is key because it is often the input and the output of the process — it is the oil that makes everything run. Data is necessary to train machine-learning algorithms and — in our case — it’s often the reason our customers come to us for in the first place. By data, we mean large quantities of content that is current, accurate, comprehensive, and enhanced. At Thomson Reuters, data takes on even a more central role because we operate in data-driven industries, such as the law. Our common law system is, by definition, data-driven — it is a collection of statutes, regulations, case law and other legal and administrative opinions that collectively represent the data that attorneys and judges must research, analyze, interpret, and reason over. This is why, for more than 130 years, we have been collecting, organizing, and enhancing this legal data for our customers. For most of our history, we did this manually; but for the last quarter-century, we developed dozens of machine-learning capabilities to automated many of these tasks so our attorney editors and tax professionals can focus on those matters that require their keen insights and expertise.
• Subject Matter Expertise is crucial because it ensures we are solving the right problems, that we are able to capture the nuances of the domain in a way that we can understand and analyze. Subject matter experts drive the generation of training data and validate the performance of machine-learning algorithms. Equally important, subject matter experts play a critical role in error analysis; they help scientists and engineers understand the domain and data attributes that are responsible for errors, which drives subsequent iterations of algorithm-tuning and development.
• Technology drives solution design and development, including that in AI and machine learning. For most of our applications, developing a solution is much more than simply applying an algorithm to a problem or a data set. It often requires designing robust and complex solution architectures that are able to deliver the target functionality at scale. This is why successful AI scientists combine significant expertise in AI tools and technologies, with significant analytical and problem-solving skills as well as a solid understanding of the target domains.