NTT Develops AI Technology to Visualize Expert Knowledge from Dialogue Data

In a significant leap for knowledge transfer and operational efficiency, NTT has announced the development of a innovative AI technology that can visualize expert decision-making processes from unstructured dialogue data. This innovation, a world-first according to the company, promises to revolutionize how organizations train staff and automate complex tasks.

The new technology addresses a critical challenge faced by many businesses, particularly in fields like security incident response and call center operations: the difficulty of transferring tacit, expert knowledge. Highly skilled employees often rely on unique thought processes and unformalized experience, making their expertise challenging to document and pass on to less experienced personnel. NTT’s new AI system tackles this problem head-on by analyzing conversations and converting the abstract decision-making of experts into a clear, actionable flowchart.

Utilizing Large Language Models (LLMs), the technology works in a two-step process. First, it extracts and groups key questions and suggestions from a vast body of text-based dialogue. It then uses these lists to trace and structure the entire flow of a conversation, from initial query to final resolution. This process transforms a series of conversational exchanges into a “flowchart” that accurately represents the expert’s thought process.

NTT claims an impressive accuracy rate of approximately 90% in reproducing the question and suggestion structures from ground-truth flowcharts. This high level of fidelity means that the visualized knowledge is not just a general overview but a detailed and reliable representation of expert-level responses.

The implications of this technology are far-reaching. By making expert knowledge accessible and replicable, companies can significantly reduce the time and resources required for staff training. This is particularly crucial in an era of worsening labor shortages. Furthermore, NTT plans to integrate these extracted decision-making processes into future AI systems, paving the way for automated responses that are not only efficient but also grounded in real-world, expert operational wisdom. This could lead to a new generation of AI tools that can handle complex queries with a level of quality and consistency previously only achievable by human experts.