GeoAI Detective
How it works: human vs AI photo geolocation
GeoAI Detective is a free daily browser game where you and an AI both try to guess where a single mystery photo or artwork was taken. You reveal evidence clues one at a time, drop a pin on a world map, and see whether your deduction beats the AI detective — including the days it is confidently wrong by thousands of kilometres. No signup, no install, a new case every day. Play today’s case →
How a round works
- You see one mystery image. The location and date are hidden.
- Optional evidence cards (architecture, signage, vegetation, climate, road markings, era markers) can be revealed one at a time — but each clue you reveal is also given to the AI.
- You guess where (and, for some cases, when) by dropping a pin on the map.
- Both guesses are scored by distance, and you watch how the AI reasoned its way to its answer.
How AI photo geolocation actually works
A vision model doesn’t “know” where a photo is — it pattern-matches surface cues it learned from millions of captioned images: the side of the road people drive on, bollard and road-line styles, license-plate shapes, architectural vernacular, vegetation and climate, sun angle, written scripts and languages, even utility poles. From those it infers a region the way an experienced human player does.
Why the AI is sometimes confidently wrong
Its error is bimodal. On a famous landmark it can place the photo within a block from memory. But on a generic field, a back-alley, or a place built to look like somewhere else (a European-style replica town, an immigrant neighbourhood with another country’s signage), the surface cues point the wrong way — and because the model has no grounded world-model, it commits to a confident answer that can be thousands of kilometres off. Surfacing exactly those failures is the point of the game.
About the images
Every image is human-curated and open-license (e.g. Wikimedia Commons), with attribution shown on reveal. None are AI-generated. Browse solved cases in the archive.