Sorry If I Broke Your Brain: Notes from a Data Annotator
Sorry If I Broke Your Brain”: An Annotator’s Notes
I do data annotation—and for a living. No gig. I observe grippers and zippers. Wonder about feasibility, when one starts an action or ends it. Do you think about how complex it might be to decide when an action starts—for example, waking up and preparing your bag? How many milliseconds are embedded in three minutes? Three minutes in the life of any human being is complex. Huge. Big time. Believe me, annotators know.
I do observe images and tell why they look artificial, and it’s not a matter of seeing six fingers instead of five. It’s a matter of knowing how people look into the camera or anything. The way you order coffee is a complex operation. Believe me, annotators know.
I also jailbreak at times, trying to make a model produce unsafe content. We need to check how harmful it can be, so we study the kinds of harm that are out there. There are many types; some of them it would be better never to learn about. But an annotator learns them, every one. You wouldn’t believe me, or maybe you would.
I do observe different languages and formats, such as JSON, and think about their structures, whether they are accurate. There are so many languages a human can use. And knowing a language is not enough because, to make a message useful, you need to maneuver the language. And naturally. And efficiently. Tone, mood. An annotator learns and knows.
When I go outside after a long day of work or when I am in the shower, I annotate the things around me. The drops on my skin. I get really surprised. Shocked by the nuances of droplets.
I look at the towels hung on the door, the curtain, every detail that was not prepared by the machine. I pay attention to my fingers. I am not AI-generated, indeed.
Yesterday, when I went to The Billionaire Problem (Walrus Talk) in a prestigious part of the city, in a theatre of a prestigious Canadian university, I annotated everything around me in my mind. Not because I needed or wanted to.
It was my brain’s behaviour, conditioned by so many hours of annotating. I do this for a living. No gig.
So much information to capture; the nuances and ambiguity were huge. I would consider it an edge case: The Billionaire Problem is an edge case. Let’s escalate the task!
There were too many humans, lots of people wearing their skin made of skin. There were no grippers made of non-skin things. The billionaire problem. The speakers started one by one, sharing their different perspectives and experiences. Many of them agreed that the super wealthy need to pay more and more taxes.
Many of them spoke in numbers to compare the land, the money, and the vulnerable people. The insecurities of our world. Our insecurities. As an annotator, I would have to stop at that part and calculate the numbers, check the sources. But I trusted that humans were doing their jobs as humans. How do we stop billionaires from being billionaires? Who does my work serve: the people or the billionaire? I grew uncomfortable in my chair. I annotated that: this anguish, perhaps.
A lot of people applauded the talk (I annotated).
I was indeed having a hard time understanding the entire concept of talking about the super rich and how taxing them would resolve lots of problems. And the problem was that, unlike an annotation task, this was a task I could not conclude. I wasn’t sure if the humans in that room could, in fact, create more fairness while at the same time some of them talking about their sponsors in a theatre. I believed their intentions, but I was confused about this task. What does intention create? I annotated.
Once, a quality leader told us, while explaining the ambiguities and complexity of guidelines, “I am sorry if I broke your brain!”
Breaking the brain of an annotator can happen after a long time spent overthinking the construction of a phrase and the momentum embedded in a gripper, a suitcase, and a zipper.
In my case, my brain breaks seeing humans. I feel at times that I can’t understand human incoherences: falling for money, being scared of death, while planning dinner at same time. Wanting immortality at times, but letting go of seconds of stillness and wonder. And I am amazed by our contradictions.
Big time, being human. The machine indeed learned from humans how not to be obviously flawed.
So it breaks my brain when, at times, I see things as they are in the human world, and I would just love for each human to understand this and be surprised by each second, trying to be less super and less millionaire. Trying to break their brains more. Be less like a machine.
If you don’t take advantage of what it means to be human, the machine is there, racing and wanting to be like you.
Then afterwards, I went to a bar in Koreatown, and while sitting on my own, I asked the waiter why the drink size was called “half cup” if it was served in an almost full cup. He agreed and said he didn’t know, and we laughed.