Will machines replace humans? That question has presented itself many times over the centuries. It has escaped no one that the current focus is around artificial intelligence. Can the machines become smarter than people? Is human labour over? Should that be celebrated or mourned?
Some call for a moratorium on AI research. Some ban it. EU regulates. Should the value of the input data for training the machine learning systems be recognised? Is “data dignity” the answer? I follow all these issues with great interest, but there is one aspect that I find underreported.
The jaw-dropping publicly accessible AI services such as text generator GPT4 or text-to-image machines like Dall-E or Midjourney have flooded our social media. Integration to existing services hints at great change, such as Midjourney integrated into chat service Discord. Or talking coffee machines. But the machine cannot create anything on its own. It uses predictive statistics (=math!) to generate these results. It needs to be trained on data before it can produce anything. A lot of data! In my recent video interview with AI scholar Stuart Russell, he takes the example of a giraffe: any child can see an image of a giraffe in a children’s book and then recognise every giraffe in any drawing, video, photo or zoo. It takes the machine thousands of correctly labelled giraffe images and it still doesn’t always get it right.
Those thousands of giraffe images must be prepared by humans. The quality of the input data decides the quality of the output. For an AI to generate an image of a giraffe in the style of Monet, it needs large numbers of giraffes and Monets – all correctly labelled. You help with this, when you click on motorcycles or traffic lights in order to convince the machine you are not one of them, you help train the AI to recognise motorcycles or traffic lights. But the bulk of the input data is made by… you guessed it – gig workers in the global South. Gig services like Fiverr or Sama, marketplaces like Defined or Appen… they connect the AI companies with the $2-an-hour-proletariat. Time Magazine has uncovered that Open AI (the company behind ChatGPT and GPT-4) paid such wages to Kenyan workers to make ChatGPT “less toxic”. AI has not taken people out of the equation. Much like the textile or manufacturing industries moving production to the cheapest workforce, AI companies do the same. Human labour on this side of the process has very little value, but comes in great quantity. (The AI engineers who develop the systems are of course far on the other end of the income spectrum).
A Centaur is not just a creature of ancient Greek mythology. In computer science, it refers to humans and AI working together. The term was coined after Deep Blue beat chess grand master Garry Kasparov in 1996-97, starting a new chapter in chess history. Today, the best chess is played by teams of humans and machines working together. Centaurs.
Turns out that generative AI is also a Centaur. Only most of the humans involved are invisible and far away on a different part of the planet. AI will not replace humans anytime soon. But perhaps there is a conversation to be had about the value of human labour. And the giraffe from the children’s book, turned out to be a Centaur from Kenya.
