AI or Not Game: Guess Which Pictures a Machine Made

The image appears at full width with no caption and no context. Two buttons, real and AI, are the only inputs. As soon as you answer, the game shows the correct label together with a short explanation of the giveaway, so the feedback arrives while you still remember what you were looking at.

That timing is the point. A tutorial read on a Tuesday changes little on Friday. A decision made thirty seconds ago, immediately corrected, is what your memory keeps. Ten of those in two minutes is a short training session that does not feel like one.

Most detection advice hands you a dozen categories to juggle: check the hands, check the text, check the reflections, check the metadata. In a feed you never run twelve checks. You make one call: forward it or not, believe it or not, use it or not.

A binary choice also exposes your own bias, which is more useful than any single verdict. If you keep marking genuine photographs as AI, you lean sceptical. If you keep missing generated images, you lean credulous. Ten rounds tell you which of the two you are, and the score is honest about it.

The generated images come from current models and show scenes people genuinely photograph: kitchens, streets at night, dogs, plates of food, portraits in daylight. The real photographs were chosen for their ordinary look, with mixed light, motion blur, awkward framing and noise in the shadows. Neither side is a trick collection.

What no pool can teach you is certainty. Every model release removes the mistakes of the previous one, and heavy compression destroys the fine detail clues usually live in. Your score measures how you read the images in front of you, not a permanent ability to detect synthetic media.

Points depend on both accuracy and speed, so a fast correct answer is worth more than a slow one. Four levels from easy to expert change how obvious the generated images are, and the harder levels also shorten the time you get per image.

A daily challenge adds one fixed set of images that everyone sees on the same day, plus a leaderboard that fills up over the day. It is the simplest way to build a habit: one round in the morning, one score to compare.

Because a round is short, the same rules work on a sofa with friends. Everyone calls real or AI at the same moment, then you reveal. Ties are common on the easy level and almost never happen on expert, which makes it a fair way to settle who actually looks at images carefully.

The multiplayer mode keeps that idea online: a shared code puts everyone in one room with the same ten images, and the scoreboard updates as people answer.