Search for a command to run...
The communication of paleontological heritage to non-specialist audiences presents unique challenges: fossils are fragmentary, ancient environments are invisible, and the scientific reasoning connecting evidence to reconstruction is often opaque. This contribution examines how generative artificial intelligence and three-dimensional digital technologies are transforming science communication practice in paleontology while proposing an epistemological framework to ensure scientific integrity in public engagement.We present a four-paradigm classification distinguishing: (1) Empirical methods (photogrammetry, structured-light scanning, LiDAR) that produce metrically accurate digital surrogates of physical specimens; (2) Neural Scene Representation (Neural Radiance Fields, 3D Gaussian Splatting) that reconstruct scenes from sparse image sets through learned interpolation; (3) Generative AI (diffusion models, large language models, image-to-video synthesis) that create novel content based on pattern recognition rather than direct observation; and (4) Hybrid approaches that combine two or more methodologies. This framework addresses a fundamental question for science communicators: whether a given digital output constitutes a record, a representation, or a hypothesis—a distinction critical for maintaining public trust.We demonstrate applications ranging from constraint-based paleoartistic reconstruction to AI-generated video synthesis for museum exhibitions and educational programs using real-world workflows created at Centro Ciência Viva de Lagos, Portugal, as part of the PaNReD (ALG-07-527-FSE-000044) and SciTour (ALG-01-0145-FEDER-072585) projects. A key case study involves the digital reconstruction workflow for Cariocecus bocagei, a new hadrosauroid from the Lower Cretaceous of Portugal, illustrating the complete pipeline from photogrammetric capture of fossil specimens through AI-assisted life reconstruction and video generation. This process illustrates how empirical 3D models function as anatomical constraints for generative AI, guaranteeing that paleoart remains connected to physical evidence while simultaneously achieving the visual impact required for effective public engagement. We critically examine the phenomenon of “hallucinated heritage”—the risk that visually convincing AI outputs may inadvertently disseminate subtle biases or fabrications to public audiences who lack the expertise to distinguish evidence-based reconstruction from algorithmic speculation.The most challenging obstacle we have faced is the preservation of the distinction between what is known from fossil evidence and what is inferred or imagined, especially when AI-generated imagery attains a photorealistic quality that may imply false certainty. Our approach addresses this through explicit labeling of epistemological status, transparent documentation of AI prompts and constraints, and educational materials that use the reconstruction process itself as a teaching tool about scientific reasoning.We argue that these technologies do not diminish the role of the scientist-communicator but rather transform it from “guardian of the rock” to “authenticator of reality.” The emotional connection fostered by immersive 3D environments and lifelike paleoart reconstructions can enhance public engagement with deep time, provided that communication strategies explicitly address the epistemological status of digital outputs. This session contribution shares lessons learned from five years of integrating digital technologies into science centre programming, offering a framework for practitioners seeking to harness AI's communicative power while preserving scientific integrity.