Reimagining Singapore with AI

Read the campaign story

Reimagining Singapore used AI-assisted imagery and voice to invite social conversation about history and imagined alternatives. TBWA Singapore describes a campaign intended to challenge the idea of a museum as an unchanging archive. The One Show records collaboration with the National Museum of Singapore and curatorial input, including speculative versions of the present and perspectives outside conventional historical protagonists. The opened sources establish the work by the 2024 awards cycle, but not a clear launch date.

Editorial interpretation: a what-if question gives an audience an active role. Instead of receiving only a finished account, people are invited to consider how a place might have developed differently and what deserves attention. Generative imagery can make such a prompt tangible. That same vividness creates a serious interpretive responsibility, because a plausible image can be mistaken for evidence when removed from its caption.

This study does not treat generated material as historical documentation or repeat the agency’s claims of solidarity as independently measured outcomes. Curatorial involvement does not remove the need to label fiction, folklore, interpretation and established facts separately. The transferable lesson is to use speculation as an invitation back to reliable sources, not to fill genuine gaps in the record with images that appear to resolve them.

Write the factual boundary before writing the prompt

Start with a short record of what is known, what is uncertain and what the creative exercise deliberately invents. Ask a subject specialist to review that boundary. A what-if scenario should identify the changed assumption so audiences can follow the reasoning. Avoid presenting an imagined protagonist’s experience as recovered testimony. Keep a source trail for factual elements and explain where interpretation begins. The goal is not to make speculation less engaging, but to give people the tools to enjoy it without confusing a vivid possibility with a documented event or an authentic historical image.

Make labels survive ordinary sharing

A disclaimer on a campaign landing page may not accompany a screenshot. Put essential labels directly beside or within the content, and include a concise explanation in captions and accessible descriptions. Test how the work looks when cropped, reposted or viewed without sound. Do not use archival styling that contradicts the disclosure. If a voice is synthetic, avoid implying that it is a recovered recording or an actual person’s statement. The design should anticipate how social platforms detach images from context, rather than assuming that every viewer begins at the museum’s carefully organised entry point.

Turn comments into inquiry rather than a popularity contest

A speculative prompt can attract jokes, disagreements and misinformation. Plan moderation with curators and educators, including how to correct errors without shutting down reasonable interpretation. Offer links to relevant collection items or further reading. Do not measure success only by the volume of argument. Ask whether participants can distinguish evidence from imagination and whether they become interested in a real question about the past. A museum’s role is not to reward whichever invented story gets the most reactions. The conversation should preserve uncertainty while making reliable knowledge easier to find and discuss.

A bounded test for another brief

Proposed adaptation: a local history group could create a clearly fictional illustrated map asking what a neighbourhood might look like if a former public garden had remained in place. Use original drawings rather than photorealistic reconstructions, and show the documented historical map alongside the hypothetical one with permission. Have a historian verify the factual baseline and label every invented feature. Compare the prompt with a conventional short history panel in a small adult workshop. Ask participants to identify the changed assumption, separate known facts from speculation and name a source they would consult next. This tests participatory inquiry without manufacturing apparent archival evidence.

Measure correct separation of fact and fiction, source use, quality of historical questions, accessibility and misconceptions after sharing. Keep discussion volume distinct from learning, museum attendance or any claimed change in collective identity.

Do not publish generated material that can reasonably be mistaken for archival evidence, and pause if the team cannot correct consequential misconceptions or explain the source boundary.