In 1950, Ray Bradbury wrote 'There Will Come Soft Rains,” a dystopian short story about a futuristic house—what today we’d call a smart home. The house continues to cook and clean and recite poetry, even though its inhabitants—and the world—have long since disappeared.

Bradbury set his story in the ‘distant future’: August 2026.

He may have not gotten everything right, but it’s undeniable that AI agents and machines can perform a number of tasks previously unconceivable—read poetry, drive cars, vacuum the house. But Bradbury wasn't prescient because he anticipated smart homes—he was prescient because he foresaw how the world might die around us. 

In 2016, an idea began to circulate widely on fringe online forums like 4chan: the “Dead Internet Theory.” Internet traffic and forum activity, users suggested, was becoming increasingly bot-generated, devoid of real human discourse. At its most conspiratorial, Dead Internet Theory posits that there is no human activity online anymore—instead, there is some tyrannical scheme to control humankind by keeping us isolated and divided.  

The situation is not that dire (yet). But now, as AI agents flood online platforms, academics, journalists, and influencers have taken a renewed interest in Dead Internet Theory, treating it as a measurable phenomenon instead of a fringe conspiracy. Reports have shown that 90% of all of the world’s data was created in the last two years, and a recent study by a cybersecurity firm showed that traffic generated by agents grows at nearly 8000% year after year. And bot activity is already leading to real-world consequences. Consider the Cracker Barrel brouhaha last year, when online outrage over a new logo led to a $100 million loss and the departure of their CEO. Later, independent analysis showed that at least 35% of the heightened online discourse was stoked by fake social media accounts.

Around the time that this theory was gaining traction online, Tom Boellstorff and Bill Maurer edited a pamphlet for Prickly Paradigm Press titled Data, Now Bigger and Better! (2015). The contributors took an anthropological and social science lens to technology, and raised questions about things often taken for granted in the tech world. The industry was becoming riddled with a slew of new vocabulary, including terms like 'big data' and 'data science.' The authors wanted to understand what terms like these really mean—and what they mask. They attacked the vocabulary of Silicon Valley with terms of their own to explain the concealed impact of big data. Questioning data's objectivity, Melissa Gregg wrote, ‘The fetishization of scale absolves considerations of infrastructure and thus ownership.” Relentless pursuit of data scalability, she argued, must not make us blind to the environmental, ecological, and social  consequences of scale. 

Without knowing it, Gregg and her co-contributors already built the vocabulary to describe the decay of the Internet. Gregg used the term ‘data exhaust’ to represent leftover material after data has been used in a transaction, intentionally using the word exhaust to bring to mind ecological connotations. Nick Seaver talked about how we seem to be “operationalizing” concepts like culture in the big data age, or trying to turn human emotions and cultures into measurable quantities that can be optimized. We’re only becoming more aware of this in the age of AI, but big data has been driving this phenomenon for years. 

Moreover, as Dead Internet Theory would have it, the Internet is running out of real human material to extract, so it's starting to consume its leftover data exhaust. This is already creating problems. AI trained on AI slop tends to produce lower quality output with each iteration, eventually generating meaningless blobs of output. It also exacerbates preexisting bias based on race, gender, and class, according to research from IBM. More simply, there’s widespread concern about the accuracy of what AI creates. As Genevieve Bell writes in Data, Now Bigger and Better!, “algorithms that clean data also bend the data toward being better looking than it really is.” We are left with false slop that feeds on itself, leading to more false slop. 

As slop inflates the data and poisons the web, it’s also destroying the physical world.  The Washington Post recently uncovered an elaborate operation by Anthropic to scan millions of books—then shred them. Project Panama, as Anthropic called it, was yet another demonstration of the lengths Big Tech would go to to obtain “colossal troves of data” to train their software. This is what Gregg was talking about when she highlighted how the tech industry celebrates the processing power at the expense of the materiality of the processed. She foreshadowed a future in which our technology consumes and discards physical information—and controls access to the data. 

Data! Now Bigger and Better isn’t about generative AI, but it offers a framework for understanding it. We must be mindful of what Gregg calls the “symbolic violence of scale.” LLMs may draw on vast amounts of data, but that data is often flawed and skewed—scale alone does not guarantee accuracy or value. We must also interrogate the intentions behind data, treating it, as Gregg says, the “result of hegemonic struggle.” When model developers intentionally forbid chatbots from answering questions about certain social issues while letting it provide its unsolicited ‘opinion’ on others, this is not accidental but an intentional choice made by corporate actors. 

Data now is certainly bigger and better than ever before, but it remains imperfect and iterative. If we allow it to grow unchecked, it will consume our digital and physical lives. Then we’ll be left, as Bradbury wrote, “alone in a city of rubble and ashes.”