AI σε κηδείες και μνημόσυνα

A digital memorial can present original records, generated material, or a combination of both. These categories should stay visible to the audience. Voice synthesis produces new audio that resembles a speaker; it does not recover an original recording of words that person never spoke. Research such as VALL-E demonstrates conditioning a pretrained speech model on an audio example, so voice imitation does not always require training a separate model on the individual. A conversational simulation, sometimes called a griefbot or deadbot, can use personal writings to imitate a style. Its generated answer remains a model output and may introduce information absent from the source records. Photo colorization has another uncertainty. The Colorful Image Colorization paper describes generating plausible colors from grayscale input. A convincing result is not evidence that the original clothing, room, or landscape had those colors. Preserve original material and identify alterations so viewers can tell an archive from a reconstruction. Ethical research also distinguishes the people involved. Hollanek and Nowaczyk-Basińska discuss the data donor, the person holding the donor's data, and the person interacting with a recreation. Their 2024 paper uses speculative scenarios to propose responsible design principles, including transparency and mutual consent. It does not establish that these tools improve grief outcomes. These distinctions matter when evaluating a proposed memorial: the holder of messages and the person expected to interact with a simulation may be different people. Neither an attractive demonstration nor a familiar voice resolves that difference.