Can Talking to an AI Version of a Loved One Help You Grieve? (www.scientificamerican.com)

Scientific American published an interview and upcoming feature by David Berreby exploring “griefbots”—generative-AI chatbots and avatars built to mimic deceased loved ones. Berreby tried several approaches himself, from bespoke startups that recreate voice and appearance to DIY prompts with large language models (e.g., ChatGPT). Building these models typically requires archival inputs (texts, photos, voice samples) plus the bereaved person’s subjective descriptions—because raw data alone won’t capture personality. The result is an interactive artifact (text or voice) that responds in real time rather than a private imagined conversation. The story matters because griefbots sit at the intersection of therapy, ethics and product design. Early reports and a small study suggest some users find griefbots helpful: they provide nonjudgmental space to process loss and can even increase social engagement rather than isolate users. But experts warn of risks—overreliance, false consolation, privacy and manipulation if companies optimize for engagement. Berreby’s verdict is cautious optimism: these tools can aid grieving people if transparently framed as artifacts (not resurrected persons), responsibly constrained by users’ memories, and designed with therapeutic safeguards instead of attention-maximizing tactics. The piece highlights both the technical mechanics and the broader social need these AIs are filling in death-averse societies.