Review Excavating AI: The Politics of Images in Machine Learning Training Sets) by Kate Crawford and Trevor Paglen. Consider the following excerpt from the conclusion:

The artist René Magritte completed a painting of a pipe and coupled it with the words “Ceci n’est pas une pipe.” Magritte called the painting La trahison des images, “The Treachery of Images.”

Magritte’s assumption was almost diametrically opposed: that images in and of themselves have, at best, a very unstable relationship to the things seem to represent, one that can be sculpted by whoever has the power to say what a particular image means. For Magritte, the meaning of images is relational, open to contestation. At first blush, Magritte’s painting might seem like a simple semiotic stunt, but the underlying dynamic Magritte underlines in the painting points to a much broader politics of representation and self-representation.

Reflect on the relationship between labels and images in a machine learning image classification dataset. Who has the power to label images and how do those labels and machine learning models trained on them impact society?

Train your own image or sound classifier using Teachable Machine and apply the model to an interactive p5.js sketch. Code examples
Document your reading reflections and your coding exercise in a blog post and add a link to the post and your p5 sketch on the Assignment 2 Wiki page. In your blog post, include visual documentation such as a recorded screen capture / video / GIFs of training the model and working in p5. You may include both in one post or write two separate posts if you prefer.

Reflection：

A single image can carry many possible labels, which means it can be understood in many different ways. But to train an image classification model, a dataset has to compress this ambiguity into fixed categories. As a result, it is hard for any dataset to cover all situations, and the labeling system becomes rigid and limited.

What feels more unsettling is that labels do not only define the image for the model, they can also quietly shape how humans understand the image. For example, I might think an object in a picture is an apple, but the label says it is a peach. Over time, people may start to accept the label as the “truth.” But in many cases, there isn’t a single ‘standard’ correct answer - labels are just one chosen interpretation.

In reality, many things are not naturally destined to look a certain way. A fan is often imagined as a cage with three blades, but that is simply a common human design choice. A fan could be built in other forms. Many standards and categories are human-made definitions, not absolute truths.

Also, “different people have different tastes.” Some may find someone in a photo beautiful, while others may not. Human perception is always slightly different. Yet datasets often force one “correct” label, flattening these differences.

I think the power to label images is usually held by a small group, such as dataset builders, companies, or research institutions. Crowdworkers often only execute the labeling tasks, while the taxonomy is decided from above. Models trained on these labels can even “brainwash” ordinary people. Someone who didn’t have bias at first may gradually develop bias and stereotypes because of these labels, starting to believe that this is simply how the world truly is. Then people begin to be treated unequally: they assume that someone who looks a certain way must behave a certain way. Human thinking is easily influenced and easily led off track, and this can seriously harm the fairness and equal development of society.