Abstracted Away: Resisting Alienation and Ungrounded Abstraction in AI Research Communities

Imagine you are learning to build a giant, magical robot in a school workshop. The teachers tell you that to make the robot smart, you must first learn the art of "abstraction." In the world of computers, abstraction is like putting a box around a messy pile of parts and just labeling it "engine." You don't need to know how every single screw works to make the robot move; you just need to know what goes in and what comes out. This is a super useful trick that helps engineers build complex things without getting overwhelmed. But here is the catch: sometimes, when we put things in a box, we accidentally throw away the most important parts of the story. We might ignore the fact that the "engine" makes a weird noise only when it's raining, or that it gets angry if you speak to it in a certain language. If we only look at the box, we miss the reality of what's happening inside.

This paper is about what happens when that "boxing up" trick goes too far in the world of Artificial Intelligence (AI) research. The authors are three young scientists who felt like they didn't belong in their own school. They noticed that the way AI researchers talk about their work often ignores real people's feelings, histories, and struggles. They call this feeling "alienation," which is like being the only person at a party who speaks a different language, while everyone else keeps talking over you. The paper explores why this happens and suggests that the way researchers "clean" their data and ignore their own emotions is actually hurting the people those robots are supposed to help.

The Story of Three Researchers Who Felt Like Outsiders

The paper is written by three early-career researchers—Vyoma, Isabel, and Neha—who are studying AI at top universities in the US. They decided to tell their story not as a dry list of facts, but as a collection of three short scenes, or "vignettes," to show how they felt pushed out of their own research communities.

Scene 1: The "Clean" Data Problem
In their first story, the authors describe a class where they were learning how to predict insurance risks using medical records. The teacher told them to "clean" the data. This meant finding any person whose medical history looked weird or didn't fit the average pattern and deleting them. The teacher called these people "noise." But the authors realized something scary: the people being deleted were often the ones who were already struggling the most—people with rare diseases, disabilities, or different backgrounds. By throwing them away to make the math look "cleaner," the class was teaching them to ignore the very people who needed help the most. The authors felt like they were being asked to delete parts of themselves to make the robot work better.

Scene 2: The Double-Edged Sword
In the second scene, the authors talk about their excitement to start graduate school. They wanted to use AI to do good things. But then, they saw news reports showing how the same AI tools they were studying were being used by armies to hurt people in war zones. They tried to talk to their professors and peers about this, asking, "Should we be building tools that can be used for violence?" Instead of having a deep conversation, they were told to just focus on making the technology "more ethical" or "safer," as if the problem was just a small bug to be fixed. The authors felt silenced. They were told that asking big questions about whether the technology should exist at all was "simplistic." They felt like they were being forced to pretend that the bad stuff wasn't happening so they could keep their jobs and grades.

Scene 3: Finding a Safe Space
The third scene is the hopeful part. After feeling lonely and confused, the authors found a different group of people—a mix of computer scientists, teachers, and sociologists—who were willing to talk about the hard stuff. In this group, no one rushed. They took time to listen. They didn't try to "clean" the messy parts of the conversation; instead, they used those messy parts to understand the problem better. For the first time, the authors felt safe. They realized that their feelings of being an outsider weren't because they were the problem, but because the usual way of doing AI research was broken.

What They Found: The "Un-grounded" Box

The authors realized that the reason they felt so alienated was because of something they call "ungrounded abstraction."

Think of "grounded" research like a tree with deep roots. The roots are the real world: the people, the history, the pain, and the joy. "Ungrounded" research is like a tree that has been cut off from its roots and is just floating in the air. It looks like a tree, but it can't drink water or feel the wind.

The paper argues that AI research often cuts off these roots in four specific ways:

1. Testimony Abstraction: When a researcher says, "Hey, this hurts people," and the community says, "You're not a real expert, you're just emotional," and ignores them.
2. Purpose Abstraction: When the goal of a project changes. You start with "let's help people," but the way you measure success becomes "let's get the highest score on a test," and you forget the people entirely.
3. Position Abstraction: When the data treats everyone as if they are the same, ignoring that some people have more power and others have less. It's like pretending a rich person and a poor person have the exact same problems.
4. Affect Abstraction: This is the most personal one. It's when researchers are taught to numb their feelings. If you see something bad happening, you are supposed to feel nothing so you can keep working. The authors say this is dangerous because your feelings are actually a warning system telling you that something is wrong.

The Solution: Listening to Your Gut and Standing Together

The paper doesn't just complain; it offers a way to fix this. The authors suggest two main things:

First, Collective Action. You don't have to fight the system alone. If you feel like something is wrong, find other people who feel the same way. When you stand together, it's harder for the system to ignore you or make you feel crazy. The authors found their strength in a group of friends who talked openly about their fears.

Second, Affective Attunement. This is a fancy way of saying: "Listen to your feelings." If you feel uncomfortable, angry, or sad about a project, don't push those feelings down. Don't "abstract" them away. Instead, treat those feelings as important data. Your gut feeling might be the first sign that the research is hurting someone. By paying attention to your emotions, you can stop the "ungrounded" research before it causes real harm.

Why This Matters

The authors are careful to say they aren't saying "stop doing AI." They love technology. But they are saying that if we keep building AI without looking at the real world, without listening to the people who are hurt, and without paying attention to our own feelings, we are going to build things that are broken and dangerous.

They suggest that the way we teach and do research needs to change. We need to stop pretending that we can just "box up" the messy parts of life. We need to keep our roots in the real world. By doing this, we can make AI that actually helps everyone, not just the people who fit neatly into a box. The paper ends by offering their ideas as a tool for anyone else who feels like an outsider in the tech world, hoping that by sharing their story, others can find their own way to feel safe and make a difference.