Descrizione del progetto
La collaborazione tra esseri umani e intelligenza artificiale come protagonista del futuro
Gli esseri umani e l’intelligenza artificiale (IA) possono collaborare in ambienti non strutturati e in continua evoluzione? In un mondo in cui questa tecnologia svolge un ruolo fondamentale nel processo decisionale, si tratta di un interrogativo di fondamentale importanza. In questo contesto, il progetto HumAIne, finanziato dall’UE, creerà un sistema operativo per la collaborazione tra esseri umani e IA con l’obiettivo di favorire lo sviluppo di applicazioni decisionali avanzate in vari settori industriali, consentendo in tal modo agli integratori di soluzioni di IA di creare sistemi collaborativi in grado di offrire prestazioni superiori rispetto ai sistemi di IA isolati e agli sforzi umani. Il progetto integrerà quattro componenti essenziali, ovvero apprendimento attivo, apprendimento neurosimbolico, apprendimento a sciame e IA spiegabile, paradigmi all’avanguardia che pongono l’operatore umano al centro della scena offrendo un controllo e una comprensione completi delle operazioni eseguite e garantendo una collaborazione ottimale tra esseri umani e IA.
Obiettivo
HumAIne will research, develop, validate and promote a novel operating system for Human-AI collaboration, which will enable the development of advanced decision making applications in dynamic, unstructured environments in different industrial sectors. The HumAIne OS will empower AI solution integrators to implement Human-AI collaboration systems that outperform AI systems and humans when working in isolation. HumAIne’s developments will be integrated into a single OS platform, which will coordinate four interwind components offering Active Learning (AL), Neuro-Symbolic Learning, Swarm Learning (SL) as well eXplainable AI (XAI) capabilities. These advanced AI paradigms are ideal for exploiting true Human-AI collaboration since, in each of them, the worker is the key actor with complete control and understanding of the performed operations. AL enables the development of effective Human-in-the-Loop systems that involve humans when AI faces increased uncertainty. Neuro-Symbolic Learning combines DL with semantics and rules to complete highly complex tasks with high accuracy while requiring considerably less training data than current AI models. Advanced XAI models will be made available, providing explanations of models’ predictions while considering the global context instead of just analysing the feature importance of a single AI model. HumAIne’s XAI will provide guidance to humans to enable the timely optimisation of AL and SL models where human participants provide feedback dynamically as well as fine-tuning of Neuro-Symbolic models. The platform will handle various types of structured and unstructured data, including inputs from humans that will be semantically correlated through ontologies, knowledge graphs, and semantic interoperability.
HumAIne will complement its platform with complementary resources (e.g. training) and will be build a vibrant community of interested parties around it, to drive exploitation and wider use of the project's results.
Campo scientifico
Parole chiave
Programma(i)
Argomento(i)
Meccanismo di finanziamento
HORIZON-RIA - HORIZON Research and Innovation ActionsCoordinatore
20139 Milano
Italia