It takes two to tango: a synergistic approach to human-machine decision making

Informazioni relative al progetto

TANGO

ID dell’accordo di sovvenzione: 101120763

DOI

10.3030/101120763

Data della firma CE 30 Giugno 2023

Data di avvio 1 Ottobre 2023

Data di completamento 30 Settembre 2027

Finanziato da

• Digital, Industry and Space

Costo totale

€ 7 008 798,75

Contributo UE

€ 7 008 798,75

Coordinato da

UNIVERSITA DEGLI STUDI DI TRENTO
Italy

Periodic Reporting for period 1 - TANGO (It takes two to tango: a synergistic approach to human-machine decision making)

Periodo di rendicontazione: 2023-10-01 al 2025-03-31

Sintesi del contesto e degli obiettivi generali del progetto

The TANGO project operates at the intersection of artificial intelligence (AI) and decision-making, addressing the increasing complexity of human-AI collaboration. TANGO aims to create a human-centric AI framework capable of supporting decision-making in critical real-world applications, ranging from healthcare to finance and policy. At its core, TANGO focuses on developing hybrid decision-making systems, blending human and machine intelligence for more accurate and transparent outcomes.

The project responds to the growing societal need for explainable, ethical, and trustworthy AI systems. The political context, such as the implementation of the EU Artificial Intelligence Act, emphasizes fairness, transparency, and accountability in AI systems, directly shaping TANGO's research objectives. Moreover, the strategic relevance of TANGO is underscored by its alignment with the EU’s push for human-centric AI leadership, as outlined in various Horizon Europe programs.


The TANGO project is structured around four key strategic objectives:

1) Cognitive Foundations for Hybrid Decision Making – This involves developing theories and models that explain how humans and machines can achieve mutual understanding. TANGO proposes an interactive approach to AI transparency, whereby AI systems can elucidate and defend their decisions based on ongoing human interaction, instead of requiring full transparency into internal processes.

2) Development of Computational Paradigms and Algorithms for Human-Machine Learning and Decision Making – TANGO is advancing machine learning methods that enable synergistic human-machine learning. This includes neuro-symbolic models and interactive learning systems that can incorporate human corrections and provide personalized, robust decision support.

3) Hybrid Decision Support Systems for Real-World Applications – The project aims to develop a software ecosystem to instantiate machine learning models capable of supporting a variety of use cases, from assisting surgical teams in complex procedures to providing transparent, ethical decision-making support in financial lending.

4) Strengthening European Leadership in Human-Centric AI – By fostering collaboration between various research institutes and industries, TANGO is creating a network of stakeholders committed to the ethical use of AI, with a focus on public trust, fairness, and regulatory compliance.

TANGO's long-term impact is expected to be significant across multiple sectors, including healthcare, finance, and public policy. The project is on track to deliver tools and methodologies that empower human decision-makers while ensuring fairness and transparency in AI-assisted decisions. By promoting ethical AI and aligning with European regulatory frameworks, TANGO contributes to maintaining Europe’s leadership in human-centric AI innovation.

Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti

Over the first 18 months, the TANGO project has made substantial progress across its work packages, laying the foundation for future development and validation of hybrid decision-making systems. The project focused on three key areas: the theoretical development of human-machine collaboration, the creation of computational paradigms, and the application of these technologies in real-world trial cases.

1) Development of Cognitive Foundations (WP1): The project explored cognitive mechanisms underpinning hybrid decision-making, focusing on developing theories of mutual understanding between humans and AI systems. The initial theory was formalized in a draft paper, which emphasizes interactive explainability as a core feature of hybrid systems. Additionally, early empirical studies suggest that human partners tend to trust AI decisions more when they are perceived as deliberate, setting the stage for future work in refining these trust mechanisms.

2) Creation of Computational Paradigms (WP2, WP3): TANGO leveraged insights from the cognitive findings in WP1 to revise cognition-aware explainability approaches for interactive machine learning (D2.1) and hybrid decision making (D3.1). These insights were instrumental in developing the first version of computational paradigms and algorithms for synergistic machine learning (D2.2) and hybrid decision making (D3.2). In particular, neuro-symbolic and human-in-the-loop solutions were developed to allow AI models to combine reliability and capacity to evolve through interaction with users. Additionally, the project investigated privacy-preserving AI frameworks and fairness in decision-making, providing critical mechanisms for transparent AI decision support.

3) System Design and Development (WP4): A major technical achievement was the definition of the architecture of the TANGO platform (Deliverable D4.1). This architecture defines the system’s modularity, scalability, and security features, enabling seamless integration of AI models. These technical foundations are pivotal for the deployment of hybrid decision-making tools across different domains, such as healthcare and policy.

4) Real-World Applications and Pilot Studies (WP5): TANGO’s real-world applications focus on four trial cases in healthcare, finance, and public policy. Key progress was made in preparing methodologies and data collection for these trials. Deliverable D5.1 the application guidelines for the Trial Handbooks, is the coordination tool guaranteeing coherence and alignment among case studies. The design of individual case studies is being done according to these guidelines.

5) Coordination and Risk Management (WP7): Effective project management, quality assurance, and risk mitigation strategies were implemented throughout the first year. The project team organized regular executive meetings, developed internal procedures, and successfully submitted all required deliverables. Notable deliverables include the Project Handbook (D7.1) and the First (D7.2) and Second (D7.3) versions of Data Management Plan, crucial for maintaining high standards of project execution.

These achievements underscore TANGO’s progress toward its objectives of developing human-centric AI systems that can support decision-makers in various sectors. The foundational work performed in the first 18 months provides a robust platform for subsequent project phases and the eventual deployment of hybrid decision-making systems.

Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)

The list of Key Exploitable Results is reported below:

- KER 1 Methods and Algorithms human-machine mutual understanding and hybrid decision making
- KER 2 TANGO Software Ecosystem
- KER 3 TANGO Personal Assistant for health and wellbeing
- KER 4 TANGO AI system for medical decision making
- KER 5 TANGO AI system for credit lending and financial applications
- KER 6 TANGO AI system for social policy making

Further details will be provided in the later stages of the project.

The TANGO methodology