Did you know that your Uber driver has 2 jobs? One of them is to drive you to this fancy restaurant, and she’s paid for it. The other, unbeknownst to her is to create “driving data” that will inevitably be fed to some kind of machine learning algorithm so that, eventually, a machine can take that job from her.
She’s not an exception. A New York Times journalist’s articles are now consumed by machines and people equally. An software engineer’s code is used to train Microsoft Copilot to write software… Most of us have even worked for free for the machine: your social media posts, your buying habits, whether it’s on Amazon or through your Visa card are all being fed to voracious machines as raw materials.
Many years ago, Amazon Web Services (AWS) (the part of Amazon that makes money) introduced a new service called Mechanical Turk whose purpose was to create an API into humans: you could hire an “elastic” amount of humans to do tasks such as data labeling, content moderation, data validation… etc. Most of these are now probably done better, faster and cheaper by LLMs. Already, Waymo is taking over the streets of many cities with (much) lower accident rates, news agencies are churning articles written by AI and vibe-coded applications are submitted on the app stores : humans are used to “train” machines… which are rapidly becoming better, faster and cheaper than us at more and more tasks.
The emerging challenge from these AI replacing humans is that they are poisoning the well. Can a Waymo car learn from another Waymo car? Can an LLM be trained on data created by another LLM without feedback loops? Maybe… but I am also convinced we will see some LLMs marketed as trained on human data only. Whether this will be a marketing trick or have a real impact remains to be seen, but our data will be to machines just like organic food is to us: a way to to stay rooted in some kind of “natural” state.
The question of why we go out of our way to “create” words, music, code, or even art is of course deeply complex, but I think we can maybe agree that, beyond any bio markers (heartbeat…), it’s what defines “liveness”… and results from our emotions, feelings, desires and fears. At this point, it’s unclear that machines experience any of these, but, just like us, they will probably do anything to “feel alive”…