The Schaubies' Story |||

We were interested in working with the probabilities that google vision API attaches to the label recognition of our images. What does it mean to recognize 0.674 Musician? 05.644 Lighting? 0.843 Art? What is then the rest of this percentage? Why isn’t it sure? Working with a tool of AI, we learn a lot about our human intelligence, creativity and artistic process. The way we approach object theater, in particular, is interesting in relation to how the image recognition systems are trained. How do we use our imagination for seeing something in an object that is not the ordinary result’ of what this object is? How do we make a common agreement - audience and performer - to see in objects something else/more? Are there different probabilities’ for transformation of objects? We created three games to visualize and represent these questions and values of probabilities/uncertainty of the AI:

game#1 - what is this?

game1game1 –> how to play: one person is presenting an object to the others, and asks what is this?”. The others say what they see. Different arrangements and constellations of objects are possible. (we found out that the audience’ then functions like an image recognition AI, giving labels to images. That’s a nice game that can be further developed and have more complicated rules).

game#2 - probability

game2game2 –> how to play: arrange values 0.0-0.8 on the floor, prepare different objects for use. Someone declares a label (ex. spaceship, dinosaur, elephant, castle…) and the others need to arrange the objects according to how probable they think this object fits the label, 0.0 being the lowest, 0.8 the highest value (we discovered that when played in a group, all participant start negotiate the probability of each object to fit this or that label, including demonstrating how and thus starting to manipulate the objects).

game#3 - objects&labels

game3game3game3 –> how to play: arrange different labels on the floor (we used a classical world - garden, castle, princess, prince, horse, sword, dragon) and prepare different objects for use. Start to converse’ with another person, matching an object to a label.

Beim dritten Spiel scheint besonders interessant, dass sich die Vorstellungen im Bezug auf die Objekte verändern, wenn sich die Relationen der Objekte zueinander verändern, bzw. einzelne Objekte ausgetauscht werden und so neue Relationen entstehen. Es ist als würde die Bedeutung im Zwischenraum, bzw. durch den Vergleich der Objekte entstehen. Das Spiel macht das irgendwie sehr gut sichtbar. (Es ist etwas anderes, wenn ein als Prinz gekennzeichneter Korken neben einem als Schwert gekennzeichneten Kugelschreiber steht, oder neben einem als Schwert gekennzeichneten Pürierstab.)

Up next 21282 sci-fi images (for neural-network training) This week I downloaded twenty-one thousand two-hundred and eighty-two images according to the following search terms: 934 asteroids 942 astronauts week#3
Latest posts The Last Schaubie Chain story Live with a computer vision software Baby steps with our own vision AI Telling worlds the technology of collaborative work, and other feelings Boxes, Boxes Everywhere Connecting the dots do sheep dream of fluffy robots? The cat and the musician week#3 games x3 21282 sci-fi images (for neural-network training) Narration Image Description and Restaging Trying the google vision API with drawings day#5 gcloud Vision API Example day#4 day#3 flag day#2 day#1