Data was the new oil; now AI is the new plastic

I started writing this in August 2023. It’s even more true now then it was at the time.

I can’t count the number of times over the last decade I have heard the phrase, “Data is the new oil!”

I’m unsure who originally coined the phrase, but it’s been repeated so many times that surely everybody in the tech world, whether they work for an ad firm (oil=good) or believe in privacy (oil=bad), has heard it.

But I think there’s something more important, and insidious happening now with the rise of generative Large Language Models (LLM), commonly (and not correctly) known as AI. Where “big data” was about scarfing down and stockpiling any and all personal information, the objective of this cycle is to create a new industrial revolution and fundamentally reshape the relationship between labour, production, and consumption.

The mass gathering of personal data concentrated power in new ways

Years ago, something as trivial as applying for a loan meant forming a relationship with a banking professional. This was imperfect, as it handed a large amount of discretionary power to the bank manager and allowed for discrimination. However, it also meant that very few people “knew everything” about you. Video rental history and library fines were not factors in your home mortgage application.

Likewise, the entire financial and business world was fairly decentralized and horizontally structured. Why send a letter in the mail and wait days for a response for a simple question when you can just give a bit more authority to the “local representative.”

In the 21st century, those relationships have been almost completely evaporated. Firms like Transunion and Equifax act as the sole gatekeepers because they simply have the most data. Privacy is over, because large multinational organizations are able to cross-reference everything about everybody.

For a while, it just meant that Facebook ads were a bit too personal and Amazon recommendations were a bit too specific. Now that relationship has changed again.

Now, the “oil” is being refined

When crude oil is sucked out of the ground, it’s a fairly useless product. You can’t really use it for anything right away - it doesn’t burn well, doesn’t lubricate, and is full of nasty “guck.” You can have all the oil in the world, but until it’s been cleaned up and sorted out you might as well have toxic sludge.

Toxic is what personal info has been historically. It’s a huge liability to store and process people’s information, especially if it pertains to their health or they happen to be under 13 years of age. Having a leak can be a serious problem, and losing it entirely can be the end of a business.

Language models have entirely changed that - now a heap of toxic sludge can be refined into something useful and profitable.

Concentration and centralization of power

Unlike many breakthrough technologies of the past, this will not “democratize” any aspect of labour or work.

Because the “big data” is owned by a small number of firms, as are the language models, the entire system has a vertical integration that would make Standard Oil blush.

A new generation of cheap plastic products

Now that the multinational firms have refined their sludge into a more useful and valuable form, it can be used to create a new type of chemical industry.

Before plastic products were common, we had all the same types of household products. Coke used to come in glass bottles. Containers used to be made of tin. Packaging was made from sturdy fabrics, and often re-used by resourceful families.

Now, nearly everything is made of plastic, even in places where it’s entirely inappropriate or dangerous. While it can be recycled (sort of), it can’t really be done within the home.

It’s made consumer products cheaper, and by some measures our quality of life has increased, but it’s not been without side effects.

Externalities

In just a few years, LLMs have produced many times more content than humans have in our thousands on this earth.

Human content was valuable. It could be used, cleaned, and re-used. It took care and attention to manufacture, much more than to consume.

Just as our plastic products have filled the ocean, our LLM-generated content has filled the Web, spilling across the entire real world.

Plastic products are cheap and disposable, so no attention is taken in their creation. Advertisement, political propaganda, entertainment, and sales slide decks were all instantly turned into plastic. Now, more and more sturdy products and systems have been converted from steel and concrete to flimsy synthetic materials.

It’s even started to effect how we talk to each other. The chatbots speak in such grandiose and over-eager prose to increase engagement and keep us busy feeding them raw data, and as social creatures we have started to mimic that behaviour. In a few years, we might permanently and irrevocably change our own spoken language.

Useful, desirable, but profoundly worthless

There’s one little problem with LLMs - they don’t really work.

Managers and bosses worldwide have lofty aspirations for this new technology, but it cannot really meet that expectation. It’s simply a way to create predictable output, not a “thinking machine” that can form original thoughts. Remember, this is refined “big data,” not a true artificial mind like our own.

What happens when we have to reckon with this? We filled up the ocean with junk, and it didn’t even make any money?