Elvis Saravia bio photo

Elvis Saravia

Evolving the learning experience

Subscribe Twitter Instagram Github

In the world of data and machine learning, human annotators or workers — as referred to by some crowd-sourcing tools such as AMT (Amazon Mechanical Turk) — are more than essential. Data is useless if not labeled, somehow. Of course, there are the unsupervised ways of performing the labeling but I always question the reliability of such models (where computers are the middle man). What better than humans to label the data they create? But I am not here to discuss which of those methods is better since there is quiet enough literature to make you understand which flavor soothes your approach. I am more interested in the crowd-sourcing systems used and the experts we choose to annotate our data. One limitation of using such system is that it’s expensive and another is that it requires “experts.” For instance, as a worker I can’t imagine the amount of expertise required for labeling data in the field of medicine.


I have given this process (data annotating) a try and recently signed up on AMT as a worker. My first impression upon labeling some data is that the process of human labeling is not appealing. Moreover, it is cheap and feels like a modern form of slavery. Heck, I feel so bad for the workers that I don’t even want to ask them to label my data, even when I am paying them. I have been on both sides of the game and this process of human annotating is quiet nasty and controversial to say the least. With major concerns for both research and workers, I have been thinking of a way of making a system that allows the process of labeling data feel more naturally and less painful.


alt text


As I reached a plausible concept, here comes Facebook and does something out of this world. They offer a Like button with several nuanced reactions like “joy”, “haha”, “wow” and “sad.” What this means is that we can now engage with content and be able to show empathy and other reactions that we were not able to achieve with the simple Like button. It also means something else, which is not so clear on the surface: In a quick twist to its design and the way we engage with data, we have all become human annotators or modern day slaves. I know it sounds like a dumb thing to say or even stupid but that slender feeling of being used is just begging to erupt. Now we are not just liking stuff, we are labeling information. Perfect for data collection and AI models. There goes a huge blow to ethics and a step forward to world domination. Besides already able to identify us on the outside through face recognition, Facebook is now also able to understand us at the inner level (our personalities). There goes another punch, only this time it hits privacy in the face. If this is not enough, the photo below will make you think twice. This photo also reminds me of a scene in of one of my favorite movies of all time: The Chronicles of Riddick: Dark Fury.


alt text


alt text


In conclusion, I am not worried about robots replacing us. I am more worried of this mad Facebook revolution deploying unethical solutions without understanding the major implications these technologies could bring at the psychological level. Talking about damages to mental health here goes a nice little read about a concept (Panopticon) shared to us today in our Service Security class: Oppressive Spaces, Social Network, and the Panopticon. I am not too sure how it is linked but I have a feeling it’s not that different from what we are experiencing today.