start

At the UCU strikes at Goldsmiths there is a black banner with white, bold, capitalised, Ariel lettering which reads: “THE UNIVERSITY IS NOT A FACTORY”. This slogan is doing the rounds as “knowledge for knowledges sake” is dismantled by the “marketisation of education”. Running in parallel to this -insidiously and equally- is the opposite which is also true, the education of the market. Certainly factories are now universities of sorts. AI’s ingest bland, homogenous slabs of Big data, and learn. And now, urgently, we have a pedagogy of AI which needs considering.

AI pedagogy has two opposite injustices, in an ideal world they would be each others solution, but this is not the case:

1_Big data ain’t big enough:

Machine learning is a tool of averages. So difference is often placed in a blindspot, real people are represented vicariously through data, and when that data doesn’t include enough people like them you get things like; facial recognition cameras with a 19% success rate, being installed into an East London shopping centre, cameras which are especially bad at recognising POC and women.

2_Big data is too big:

We can comfortably assume that all of the products that Google sells are data collection points, little black holes which absorb the information we give off. This Data is passed on to a 3rd party which has a 3rd Party which has a 3rd party (and so on and so fourth) it exponentially duplicates and spreads outwards into unregulated territories. AI pedagogy (as it is at the moment) needs this IoT.

Google’s AIY robots (Google’s children) are small cardboard boxes with a blank DIY AI inside. We want to work with academics, researchers and the public to assemble datasets and training models which co-opt/hack/radicalise these bots, both internally and externally, digitally and analogally.

An example of an AI learning curriculum is the somewhat bizarre MNIST Fashion dataset (image attached). A catalogue of 60,000 28 x 28 grayscale images of fashion items (Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle Boot) is fed into nascent machine learning models in order to test whether they are functioning properly. These 9 categories are what the AI thinks of as fashion, no more, no less.

“We’re surrounded by invisible but powerful forces, monitoring us from devices scattered throughout our homes, even placed on our bodies, and those forces are busily compiling detailed dossiers of us. They pass the contents of these dossiers onto shadowy, unaccountable intermediaries, who use everything they learn to determine the structure of the opportunities extended to us - or, what may be worse, not extended. We’ll be offered jobs, or not; loans, or not; cures, or not. And the worst of it is that until the day we die we”ll never know which action or inaction of our own led to any of these outcomes” Radical technologies, Adam Greenfield