DARKNITE: Dialogue Agents Relying on Knowledge-Neural hybrids for Interactive Training Environments / Agentes Conversacionales basados en Tecnología Simbólica y Neuronal para Entornos de Aprendizaje Interactivo

The overall objective of the DARK NITE project is to develop a mechanism for combining knowledge-based and neural solutions for the dynamic generation of narrative and dialogue in the context of interactive training. In the prototypes that will be developed in the project to achieve this objective, a human instructor will be able to iteratively refine a list of learning objectives and the system will rely on the generative functionalities for narrative and dialogue construction to construct a learning scenario. The interactive learning scenario will have intelligent characters and it will adapt to the users interactions. In addition, data from all interactions will be collected to analyse the performance of the users and provide them with valuable feedback.

The DARK NITE project intends to explore the combinations of knowledge-based and neural-driven artificial intelligence solutions to deliver an optimal experience to users interested in training skills and competences best acquired in interactive settings. Creating immersive, human-oriented interfaces through language, haptic devices, realistic audio or ambient music can improve the effectiveness of many learning processes. The increased engagement that these systems produce, together with natural interactivity, can decrease the entry barrier and promote free experimentation.

The project will consider a number of application case studies to identify whether particular domains may require specific combinations for optimal learning results. The potential domains of application include training in emergency response protocols, training in communication skills, and language learning among others, which have been historically addressed using non-neural AI techniques. Given the differences in the nature of skills and types of training across these different domains, such a broad scope of options is more likely to include specific modes of interaction that best exercise the different kinds of functionality that the hybrid AI solutions being explored in this project are likely to make possible. The methodology applied at this level will involve development of state-of-the-art training environments for each case, and then focus on applicability of hybrid AI solutions for domain specific problems that best exercise the new functionalities being developed.

Given that the set of interested users, the set of potential testers and most of the researchers are native speakers of Spanish, the ultimate aim of the project is to develop versions of all system functionality that operate in Spanish. However, as the state of the art in artificial intelligence technologies tends to be always more advanced for English, we will consider the development of versions in English of specific modules either to test feasibility or to acquire operational insights. Ideally, such versions in English will eventually be ported to equivalent versions in Spanish where required.

The main overall outcomes of the project will be: a) a solution for dynamically generating goal-oriented narratives; b) a solution for context specific construction of goal-oriented contributions to an ongoing dialogue; c) a mechanism for combining knowledge-based and neural solutions in aid of the above goals; and d) a significant reduction of the effort required to develop interactive training systems arising from the introduction of solutions a), b) and c), supported by empirical evaluation over working prototypes.

DARK NITE project web site