Descrizione del progetto
L’IA antropocentrica che ci capirà meglio
L’Intelligenza artificiale (IA) ha già reso possibile la comunicazione tra le macchine e gli esseri umani, oltre ad aver offerto soluzioni a vari problemi. Ma l’IA potrebbe cogliere significati e motivazioni più profondi? Il progetto MUHAI, finanziato dall’UE, indagherà la questione esplorando un approccio radicalmente nuovo negli studi sulla tecnologia di IA antropocentrica che si concentrano sul significato e sulla comprensione. I ricercatori si riferiranno alla teoria della memoria dinamica personale per fare luce sul processo di ricostruzione dei significati delle esperienze. Sulla base dei recenti progressi nei componenti tecnici dell’IA, come le reti di apprendimento profondo, ne costruirà di nuovi, come la stimolazione mentale delle azioni utilizzando ambienti virtuali ludicizzati. Il nuovo approccio sarà testato in casi che richiedono senso comune e una comprensione dei fenomeni sociali, come per esempio la persistenza della disuguaglianza nella nostra società.
Obiettivo
The MUHAI project explores a radically new approach to push the envelope of human-centric AI technology to come to grips with meaning and understanding. Meanings are distinctions (categories) that are relevant for prediction, classification, communication, problem solving or other mental tasks. The meanings of an experience include what events, actors and entities play a role, the temporal, spatial and causal relations between events, intentions and motivations of the actors, and values that implicitly underly their behaviour. Understanding is the process of reconstructing these meanings and organizing them in terms of a coherent narrative that explains a new experience by linking it into a Personal Dynamic Memory.
The MUHAI project will advance our ability to capture more of the rich and complex understanding that humans are capable of. It critically relies on enormous recent advances in technical AI components (such as deep learning networks) and semantic resources (such as knowledge graphs). In addition, the project will build novel components, such as: mental simulation of actions using gamified virtual environments and a powerful context mechanism to deal with the complexity of accessing, expanding and managing a very big dynamic memory (> 10-100 billion of facts). All these components will be assembled in a library called CANVAS, made available through the AI4EU AI-on-demand platform.
The technical advances of the project will be developed and systematically tested in two challenging human-centric case studies emphasizing AI for the common good: (i) learning and using common sense every day knowledge on how to do things, in order to enhance well-being and empower common human activities, and (ii) learning about society through AI-supported analysis of historical sources and contemporary digital media streams, in order to understand social phenomena, specifically the origins and persistence of inequality in our society.
Campo scientifico
Parole chiave
Programma(i)
Invito a presentare proposte
Vedi altri progetti per questo bandoBando secondario
H2020-EIC-FETPROACT-2019
Meccanismo di finanziamento
RIA - Research and Innovation actionCoordinatore
28359 Bremen
Germania