The unseen labour of algorithm training

I’ve been thinking a lot lately about how labour is hidden by technology. This is true in a general sense - software can seem ethereal and non-physical to such an extent that we can forget it was built by people - but in particular in the case of artificial intelligence.

What’s most fascinating about human labour in the context of AI is the fact that much of the conversation around the topic is about automation and humans losing their jobs. Doing some quick research for this made this abundantly clear: foolish and lazy, I searched for ‘AI human labor’ only to be met by a wall of articles about how AI was or wasn’t going to replace humans.

Follow me on Twitter

There’s a further strand of this conversation that I find deeply ironic: the notion that AI is going to stop ‘boring jobs.’ Here are couple of pieces on that topic but there are probably hundreds, if not thousands swirling around on the web:

Now, the central premise isn’t incorrect. AI can minimize boring and repetitive tasks. But despite the benefits, there are plenty of examples of how AI is creating new boring tasks. There’s a whole labour force of people training or augmenting AI to understand language, identify images - pretty much everything.

“Crowdworking”

There’s a whole sub-sector of businesses that are using ‘crowdworking’ to train algorithms. In essence, this is the AI gig economy. This article on The New York Times explores what this labour looks like in practice.

In particular, I liked the description of the physical space in which these activities are happening:

“What I saw didn’t look very much like the future — or at least the automated one you might imagine. The offices could have been call centers or payment processing centers. One was a timeworn former apartment building in the middle of a low-income residential neighborhood in western Kolkata that teemed with pedestrians, auto rickshaws and street vendors.”

This raises some really important questions about AI and automation that we still haven’t really reckoned with: who benefits from automation? Who gets to remove the boredom from their work and who doesn’t?

Although the piece acknowledges that this work isn’t always low paid, for those in contract positions, the work is certainly precarious. But more importantly, it can also be arduous and even disturbing - some workers will have to label and categorize pornographic, violent, or surgical/medical images.

The numbers

I was shocked at the numbers behind this section of the tech industry:

“The market for data labeling passed $500 million in 2018 and it will reach $1.2 billion by 2023, according to the research firm Cognilytica. This kind of work, the study showed, accounted for 80 percent of the time spent building A.I. technology.”

There’s something grimly ironic about the material reality of artificial intelligence. The speed with which we can find answers, find recommendations, sort libraries and tweak photographs can feel irresistibly seamless and modern.

But the fact that these are built out of conditions that are at best dull and mindnumbing, and at worst traumatic highlights that we maybe shouldn’t worry so much about automation; we should instead focus on who it’s actually for.