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?
I acknowledge the unstable relationship between labels and images, and I somewhat believe the classification of an image dataset is problematic by nature. The passage mentions how the class of “person” was controversial and later got deleted. However, the problem of these datasets is not solved because they are still inside the computer system, been downloaded and maybe still being used, even if we cannot specify where they are stored. I notice that when AI-related projects have the potential to encounter such ethical/political criticisms, it is difficult to come up with a perfect solution to manage. I keep seeing reminders like “ChatGPT may produce inaccurate information about people, places, or facts” and “Bard may display inaccurate or offensive information that doesn’t represent Google’s views”. I feel like that people are pushing them away from whatever risks that may cause and leave the consequence to the machine/systems, which is quite irresponsible. After all, it is people behind who have the power and ownership of the images and perform trainings.
If the implicit assumption and problematic interpretation is inevitable, maybe we could at least find ways to mitigate their effects. From a user’s point of view, I would like to see a more transparent process about how the models or networks are labeled, trained, and used so that it gives a wider public discussion for people to identify the biases and decentralize the power in order to collectively supervise the models.
I trained a model with Teachable Machine with different directions.
I first tried to train the model with my hand pointing different directions 👆👈👇👉. However, it cannot classify well between UP and RIGHT and I don’t know why it is not responding when I opened the model in p5.js.
not responding 😭
Therefore, I tried to draw an arrow on a small piece of paper and trained the model by holding it to different directions. This time, the accuracy is pretty good; Yet, there are still some fluctuations and I think it varies with the way my hand holds the paper.
I wrote a simple game that the user will rotate the paper according to the random directions shown on the screen. Because of the stability issue, I set the terminating condition to "count==20" so the user could have a sense of how it works.
