Clean Up Your Act: The Invisible Hands Behind Big Data

The Help: The Invisible Hands Behind Big Data

We can have all the intelligence in the world, but if we don’t use it for good, for the betterment of society, what does it mean?

As someone who has extensively studied the sociology of knowledge and how knowledge is shared in our classrooms and communities, I have found that while knowledge can be liberating, institutions tend to deploy knowledge (like a tool) in ways that coercively alienate, exploit, and control. As such, it is important that we understand how knowledge is socially constructed in ways that reflect social relations of domination and resistance. This understanding becomes even more critical when we consider the development and deployment of large language models (LLMs) as a means of facilitating knowledge.

If LLMs are run by (i.e. cultural ethos) industry, then the information generated from them will reflect industrial needs. If LLMs are run by the people, then the information generated from them will reflect the humanity of the people.

Engaging in human-like conversations with an entity that possesses vast knowledge is undeniably attractive. The sheer volume of information LLMs hold is staggering — it literally would take hundreds of lifetimes to learn what they process at any given moment. Personally, I’ve found large LLMs to be invaluable for generating insights and ideas. I have used them to help develop research designs, outlines, scripts, lesson plans, workshop activities, and more. They have been profoundly helpful in ideating foundational insights and possibilities. In fact, I am of the mindset that if we don’t use them, we will be left behind. Yet, and still, I wonder about the people behind this vast body of knowledge. Who is cleaning and training this data?

As it turns out, communities of the Global South are training these models. According to Julia Zorthian of the New York Times, trainers in Nigeria received less than $2 an hour in exploitative conditions reviewing the content that powers these LLMs. And that these horrifying conditions influenced their worldviews in ways that traumatically damaged their mental health and their relationships with loved ones. Betsy Reed from The Guardian reported the following working conditions of these modern-day sweatshops:

The 27-year-old said he would view up to 700 text passages a day, many depicting graphic sexual violence. He recalls he started avoiding people after having read texts about rapists and found himself projecting paranoid narratives onto people around him. Then last year, his wife told him he was a changed man and left. She was pregnant at the time. “I lost my family,” he said.

LLM trainers (cleaners) spend approximately 9 hours a day tagging toxic illustrations of violence, child sexual abuse, rape, beastiality, murder, sex slavery, and more to clean the data that we access.

I am not saying that workers did not have the autonomy to quit or that Sama was forcing them to do this work, but I am wondering how we let this happen. We should all be wondering how we let this happen. And WHY this community? Why did they not care enough to think through the psychic trauma and violence this could do to this community? I mean, would you let your loved ones immerse themselves, their psyche, in this kind of discourse or rhetoric? Imagine the imagery and trauma this has caused to their worldview and imagination. I mean, there are very scientifically sound reasons for why we have movie ratings.

While I appreciate the efforts taken to ensure that these models are trained (cleaned) NOT to reflect the sick world we live in, I wonder who will carry the burden of healing it and at what cost? I’m asking us to examine how we treat those who clean up our digital mess. And so, I implore the AI community (UX researchers, designers, developers) to clean up our act!