We bring the human touch to artificial intelligence

What to know about AI at Thomson Reuters

We have uniquely positioned ourselves to bring together effective AI solutions. Over time, we have identified three vital ingredients that distinguish the strength of our AI — technology, content, and human expertise.

With over a century of experience in curating and classifying rich datasets in legal, tax, and other professional fields, we recognized early on — beginning in the early 1990s — the distinctive capabilities of AI and machine learning to provide quick and reliable guidance, continually leveraging technological advances to enhance professional access to information. None of that is possible without human expertise.

Reliable AI is human-centered AI

Our experts are the critical underpinning to our best-in-class solutions. Our domain experts supply the industry background — bringing their knowledge and understanding of our customers’ workflows, nuances, and daily challenges into the core of our solutions.

In tandem, our technical experts do the heavy lifting of developing our AI-powered technology. These professionals train our data for machine-learning systems, develop search and question-and-answer systems, and create deep-learning modules to solve complex problems and improve decision-making. However, our distinction lies in the fact that many of our technical experts also possess domain expertise.

Because we invest in this combined expertise and research, we can ensure our solutions deliver accurate and thoughtful knowledge to our customers.

According to the latest Future of Professionals report, almost two-thirds of respondents believe a “human-in-the-loop” approach is critical for responsible AI use. Human-in-the-loop (HITL) refers to human involvement in machine-learning workflows to reduce errors and improve model performance.

Over time, our experts have developed AI systems that streamline searches and make them more valuable and intuitive. They can automate processes, find information traditional research may have missed, and gain data-driven insights.

The human expertise behind CoCounsel

The recent advancements in generative AI (GenAI) technology present an unparalleled opportunity to revolutionize how our products interact across the industries we serve. This development enables our customers to integrate the capabilities of GenAI into their daily workflows confidently.

Our professional-grade GenAI assistant, CoCounsel, is backed by 2,500 know-how experts. It provides a single, secure, human-centered chat interface equipped to conduct in-depth research, analyze extensive and complex data, and generate diverse types of content quickly.

CoCounsel’s process for ensuring reliable, trustworthy results involves three distinct levels of HITL testing.

Model testing

This process begins with the first phase — testing each new model discovered against a suite of tests created by experts. These tests ascertain the model's ability to reason and perform tasks based on a mix of both publicly available and Thomson Reuters legal benchmarks. These tests validate at the basic level how capable this model is of completing reasoning tasks.

Individual request testing

The second phase involves testing individual requests or tasks and evaluating their quality using prompt engineering. CoCounsel's prompt engineers and the Trust Team have worked together to create hundreds or even thousands of specific prompt-level test cases with known accurate answers to those tests. These teams validate that the model answers match against the known good answers.

Prompt engineers

“Our team of prompt engineers have a very unique background,” says Ryan Walker, Vice President of Technology at Thomson Reuters. “They are all attorneys who also have a background in software engineering, which gives them the expertise to test the prompts and assess the quality of the answers they receive. Even when you get to the engineering layers of CoCounsel, the domain expertise is embedded, which is critical to building this type of technology.” 

The Trust Team

The Trust Team comprises full-time attorneys focusing solely on the quality of CoCounsel’s answers. The partnership of prompt engineers with our Trust Team’s feedback is vital in ensuring the accuracy and reliability of the AI system .

Skill-level testing

The final testing phase involves end-to-end testing of skills against real use cases. Now that the microprompts have all been tested, the skill is tested holistically. At this stage, experts evaluate the overall quality of the answers generated and confirm that all the pieces come together correctly and the product is a high-quality, accurate response.

AI is not new to us — and neither is this model of domain expertise — it’s always been fundamental to our business and solutions.

While AI technology is remarkable, the results produced by AI are best when paired with human experience and critical thinking. We have been leveraging AI for over three decades to transform the way you get work done.