Madhumita Murgia (Picador)
THE artificial intelligence revolution has overwhelmed us all since the release of OpenAI’s ChatGPT in November 2022. But AI is far more than that, as a new book by Madhumita Murgia, AI editor for the Financial Times, explains.
Murgia has long been fascinated by data, writing a story for tech magazine Wired a decade ago that outlined how eager tech companies are to collect and analyse information about their customers. But in Code Dependent: Living in the shadow of AI, she chooses to focus on the humans who help make the technology work, or whose lives are affected by its decisions.
Advertisement
There are no mentions of Sam Altman, OpenAI’s enigmatic CEO, in her book, but plenty of Sama, an outsourcing company used by many tech giants to label the data on which they train AI systems. Murgia trots the globe, but she studiously avoids the hotbed of the revolution. Her book is better for it.
She sites the action largely in the time before the release of ChatGPT, and this provides a useful primer for the world in which we now live. Her deftly told stories are those of the human annotators who, among other things, help train autonomous driving systems or pore through image classifiers and text, adding labels about what is and what isn’t appropriate content.
We also ride alongside delivery drivers whose livelihoods are taken from them in an instant due to misfiring algorithms. And we sit in a doctor’s surgery as patients are treated with the help of AI.
To judge by the meticulous detail in her book, Murgia has worked hard to win the trust of those she writes about in its 10 character-led chapters. She learns which flowers a beekeeper-turned-data worker (who escaped conflict in Iraq and ended up in a low-paid job training an algorithm) saw his insects feed upon. She eats alongside those she profiles, breaking bread with them while learning how they help power the tech we use daily.
Through these moments, we learn how precarious the work now being handled by algorithms is. One data labeller explains how a simple AI-guided change in shift patterns by her company lost her vast amounts of work – and how the bosses then banned her from working for them for a month for complaining.
We learn, too, how unreliable this data may be. Another data annotator works earnestly and hard. But he is labelling bones for medical AI without knowing human anatomy. He says that he enjoys labelling roads and traffic signs, hoping that, one day, he will be able to drive himself and so put that knowledge to use.
Other labellers admit that they sometimes have to guess at the least wrong answer, their decisions forever encoded into AI models.
Murgia also tells her own journey to bookend the narrative. In her introduction, she explains how she was a techno-optimist – perhaps understandably, given her career as tech journalist – but has since become more sceptical of the tech that is currently “altering the very experience of being human”, as she describes it.
However, she is no Luddite. To conclude, she offers a list of 10 questions designed to provoke thinking about how to better frame our relationship with AI, including who should be accountable for AI decisions in life-and-death, or life-altering, choices, and how people can opt out of being swept up in AI’s dragnet.
Given the topic’s ubiquity, it is refreshing when a new perspective comes along. And Code Dependent is just that, making it a must-read for those struggling to reckon with the AI revolution.
Chris Stokel-Walker is a writer based in Newcastle upon Tyne, UK
