At what cost?
Humans as Resources in the age of AI.
At what cost?
This is the question that has been running through my mind recently. The drive towards AGI (Artificial General Intelligence) sees a small number of highly influential individuals operating organisations to strive to be the first to discover AGI. OpenAI, Anthropic, Microsoft, Meta, Alphabet, X to name the most prominent. Huge amounts of money being ploughed into, at the moment, Generative AI and the drive to find uses for it.
At what cost?
This is about Humans as Resources. There are other costs, for example the environmental, but I’ll leave that for another day. I recently read (well listened to) the book Empire of AI by Karen Hao. A very comprehensive account of the reckless race for domination of “AI”.
One area of this race for domination is that of Data Annotation. For the Large Language Models (for that’s the main focus, at the moment for most of these organisations) huge amounts of data needs to be used to train the models. This data comes from a variety of sources - web scraping, dubious data sets (illegal collections of published books going against copyright laws) and other nefarious means. But what a lot of people are unaware of is that for the Models to be trained they need to be told what the data IS. The data needs categorising, it needs Annotating. This is a process where Humans spend time tagging content. The driver for this Annotation was first with driverless vehicle and computer vision but as Generative AI boomed so did the Data Annotation Industry.
At what cost?
Data is the new oil. And like the oil industry, the data industry is built on exploitation. Exploitation of poor folks in mostly the Global South for example in countries like Kenya and Venezuela. Silicon Valley led companies “recruit” workers across the poorest regions and set them to annotate. They get paid next to nothing, forced to compete against each other. Forced to use tools that are inadequate and also forced to look at some of the most depraved content. You see, part of the annotation work is to tag awful content so that the Models know good from bad. This is content that is deeply disturbing, illegal. The mental tole on these workers is immense. So many have suffered from life changing mental health issues due to the exposure of the worst of the worst content. And a lot of that content is also AI generated. In Feeding the Machine: The Hidden Human Labor Powering AI; Callum Cant, James Muldoon and Mark Graham go into more details on what this cost really is.
As the West strives for continuous growth (unsustainable growth). As governments push for more and more productivity the desire to harness “AI” is causing untoward suffering. The benefits of this “productivity” gains are mostly illusions, often being used to do the pointless work that people have always said was getting in the way of the “real work”. Or, the illusion of “speed” being “more productive” but often not being more effective. The tools making us move faster but often in the wrong direction. (2024 State of DevOps Report)
Again we see that Humans are one of the Resources to be exploited by rampant neoliberalism.
AI is broad. It doesn’t just mean Large Language Models and Generative AI, although that’s what most people now consider “AI”. The field of AI is being used for good. But due to the drive from big players a lot of the infrastructure, compute resources and funding is now out of reach for research.
At what cost?