Introduction and Core Concepts

Introduction

The TANGO project aims to address the low adoption rate of AI-based decision support systems in high-stakes fields like healthcare, justice and public services. A primary barrier to adoption is a lack of trust, stemming from the difficulty in assessing the assumptions, limitations and capabilities of AI assistants, as well as the opacity of their intentions. TANGO argues for a fundamental rethinking of how AI systems are conceived, proposing a symbiotic relationship where humans and machines are aligned in values, goals and beliefs. The goal is to create a Hybrid Decision Support System (HDSS) where the AI enhances the awareness and autonomy of the human decision-maker, leading to a final decision that is better informed and less biased than one made by either the human or the AI alone.

Core concepts

To achieve this, TANGO's strategy is built upon three core concepts.

First is the development of a cognitive theory of mutual understanding, which seeks to transfer the mechanisms of human-human communication to human-machine interaction, allowing them to jointly infer objectives and interpret signals. This is complemented by a "folk theory of hybrid decision making" to understand how the perceived nature of a decision (e.g., intuitive vs. deliberate) affects trust in both human and AI partners.

The second concept is the creation of cognition-aware Explainable AI (XAI) paradigms. This approach views the problem of explanation as determining what crucial information a human needs to make an informed decision, while safely ignoring unnecessary algorithmic details. The optimal explanation may vary depending on the user—be it a domain expert, a layperson, or a data scientist.

The third pillar is a "Human-in-the-loop" co-evolution of decision-makers and machine learning models. In this model, the AI assistant and the human learn from each other through a bi-directional dialogue facilitated by explanations. The machine and user progressively understand each other's strengths and weaknesses, learning to complement one another to maximize the effectiveness of the joint system. This virtuous cycle of cognitive-computational research, where theoretical insights guide algorithmic development and real-world application provides feedback for refinement, forms the foundation of the TANGO framework.