Media Authenticity through Grounded and Interpretative AI - Visual semiotics and computational reasoning for contextual truthworthiness assessment of AI-mediated imagery -
ProgettoGenerative AI is transforming visual communication by enabling the creation of highly realistic synthetic content and facilitating sophisticated forms of image
manipulation. In this scenario, traditional distinctions such as authentic versus fake, manipulated versus pristine, or synthetic versus non-synthetic are
becoming insufficient to characterize the reliability and communicative role of visual information. Contemporary information disorders often emerge not from
the presence of manipulated content, but from the interaction between visual material, contextual framing, dissemination environment, and communicative
intent.
M.A.G.I.A. (Media Authenticity through Grounded and Interpretative AI) addresses this challenge by introducing a new paradigm for contextual
truthworthiness assessment. The project starts from the premise that authenticity should not be conceived as an intrinsic property of an image, but as a
contextual and relational condition emerging from the interaction between material evidence, communicative function, and dissemination context. By moving
beyond binary notions of authenticity and manipulation, M.A.G.I.A. aims to establish contextual truthworthiness assessment as a new research direction at
the intersection of multimedia forensics, multimodal AI, and visual semiotics.
To operationalize this vision, the project integrates multimedia forensics, visual semiotics, affective computing, and agentic AI within a unified assessment
framework. The proposed methodology is organized around three complementary phases: (i) contextual positioning, where images are characterized
according to their communicative role and evidentiary function; (ii) multiclue evidence gathering, combining forensic analysis, semiotic interpretation, and
affective cues; and (iii) truthworthiness reasoning, where agentic AI integrates heterogeneous evidence to produce explainable assessments grounded in
both material characteristics and contextual interpretation.
The project brings together six research units with complementary expertise in multimedia forensics, AI, affective computing and visual semiotics. Through
this interdisciplinary collaboration, M.A.G.I.A. will develop datasets, annotation resources, computational methodologies, and an interactive demonstrator
supporting explainable contextual truthworthiness assessment.
The expected outcomes will support journalists, fact-checkers, public institutions, and citizens in navigating increasingly complex information ecosystems
and in critically assessing AI-mediated visual content. More broadly, M.A.G.I.A. will contribute to the development of more transparent, explainable, and
trustworthy forms of visual communication in the age of generative AI.
manipulation. In this scenario, traditional distinctions such as authentic versus fake, manipulated versus pristine, or synthetic versus non-synthetic are
becoming insufficient to characterize the reliability and communicative role of visual information. Contemporary information disorders often emerge not from
the presence of manipulated content, but from the interaction between visual material, contextual framing, dissemination environment, and communicative
intent.
M.A.G.I.A. (Media Authenticity through Grounded and Interpretative AI) addresses this challenge by introducing a new paradigm for contextual
truthworthiness assessment. The project starts from the premise that authenticity should not be conceived as an intrinsic property of an image, but as a
contextual and relational condition emerging from the interaction between material evidence, communicative function, and dissemination context. By moving
beyond binary notions of authenticity and manipulation, M.A.G.I.A. aims to establish contextual truthworthiness assessment as a new research direction at
the intersection of multimedia forensics, multimodal AI, and visual semiotics.
To operationalize this vision, the project integrates multimedia forensics, visual semiotics, affective computing, and agentic AI within a unified assessment
framework. The proposed methodology is organized around three complementary phases: (i) contextual positioning, where images are characterized
according to their communicative role and evidentiary function; (ii) multiclue evidence gathering, combining forensic analysis, semiotic interpretation, and
affective cues; and (iii) truthworthiness reasoning, where agentic AI integrates heterogeneous evidence to produce explainable assessments grounded in
both material characteristics and contextual interpretation.
The project brings together six research units with complementary expertise in multimedia forensics, AI, affective computing and visual semiotics. Through
this interdisciplinary collaboration, M.A.G.I.A. will develop datasets, annotation resources, computational methodologies, and an interactive demonstrator
supporting explainable contextual truthworthiness assessment.
The expected outcomes will support journalists, fact-checkers, public institutions, and citizens in navigating increasingly complex information ecosystems
and in critically assessing AI-mediated visual content. More broadly, M.A.G.I.A. will contribute to the development of more transparent, explainable, and
trustworthy forms of visual communication in the age of generative AI.