Funds for research into AI conversation training

Utrecht University and its spin-off, DialogueTrainer, have secured funding from Click.nl to elevate their virtual conversation trainer to the next level. By leveraging artificial intelligence, the platform aims to more closely mimic reality with avatars that respond to speech, intonation, emotion, facial expressions, and even body language.

The research, conducted under the banner of the AI & Education Lab, will span four years. “In the AI & Education Lab, scientists collaborate with tech companies that develop tools and solutions to improve education,” says Sergey Sosnovsky, coordinator of the AI & Education Lab. “In this time of AI breakthroughs, it is particularly important for these companies to future-proof their products and services. Our collaboration and knowledge exchange will contribute to the design, development, evaluation, and deployment of next-generation AI-enabled dialogue technologies for the domain of education.”

DialogueTrainer is a platform developed in 2013 at Utrecht University to train large groups of students in professional conversations. The initiative by Johan Jeuring, professor of Software Technology for Learning and Teaching, aims to enrich educational programmes and better prepare graduates for real-world practice. The platform offers various scenarios—such as giving feedback, coaching employees, and health interventions—where a virtual character in an online environment plays the role of customer, colleague, candidate, client, or patient. This virtual player, an avatar, responds to what the student or professional does in the conversation with speech, intonation, emotion, facial expressions, and body language.

Recognise and avoid pitfalls

With the €120,000 funding from CLICKNL, the top consortium for Knowledge and Innovation, Utrecht University and DialogueTrainer intend to explore how they can improve the quality of the conversation scenarios using generative AI. “The strength of AI lies in predicting text that fits well in a context,” says Jeuring. “The dialogue scenarios of DialogueTrainer share this characteristic: developed with experts and refined through testing with target groups, they resemble dialogues that are conducted in real life.”

“Our research focuses on training a large language model and enhancing the platform's responsiveness,” adds Michiel Hulsbergen, CEO of DialogueTrainer. “We are already able to offer AI-driven open speech simulations, and with this funding, we hope to incorporate body language. This will further enhance the sense of immersion.”

Realistic conversation models can make performance in conversations measurable, according to Hulsbergen. “Most people lack a systematic view of conversations. As a result, professionals only partially reflect on goals, and many digressions or missteps in conversations go unnoticed. With this platform, professionals and students can learn to recognise and avoid pitfalls. Thus, science, practice, and innovation beautifully come together.”