How everyone is wrong about AI
The newspaper was an unexpected development
It was not a product that anyone asked for. Like many innovations, it wasn’t an invention. It was not patented or discovered in a laboratory. It emerged as a result of the convergence of three disparate technologies: continuous roll printing presses, the telegraph, and the railroad. This convergence occurred by the 1830s, facilitated by a business model aptly named “the penny press”.
Stepping back, the printing press was initially developed in the 1450s, 40 years before Columbus’ voyage. It went through numerous evolutions. The big change was moving from batch (page-at-a-time) to continuous (rotary) processes. In the 1600s, Gutenberg type presses could operate at the rate of 240 impressions per hour. By 1818, The Koenig press could reach 2400 impressions per hour.
The steam-powered rotary printing press, invented in 1843 in the United States by Richard M. Hoe, ultimately allowed millions of copies of a page in a single day. Mass production of printed works flourished after the transition to rolled paper, as continuous feed allowed the presses to run at a much faster pace. Hoe’s original design operated at up to 2,000 revolutions per hour where each revolution deposited 4 page images, giving the press a throughput of 8,000 pages per hour. By 1891, The New York World and the Philadelphia Item were operating presses producing 90,000 4-page sheets per hour. –Wikipedia
By harnessing steam engine power, the marginal cost of printing approached zero and typesetting productivity was such that tens of these pages could be laid out in hours. The production problem was solved.
The telegraph gave rise to news services which could relay national or international news to local papers nearly instantly. Small towns did not need to employ correspondents to reside on assignment to report what was happening in any and all national capitals. New syndication business models represented by Reuters and the Associated Press would distribute to thousands of smaller papers and communicate the stories via the Victorian Internet. Local reporters could flesh-in the paper with the latest from the community. The content sourcing problem was solved.
Finally the railroad offered daily distribution from urban to rural populations. This meant the entirety of a country’s population would have access to news and commentary and, “all that was fit to be printed” on a daily basis. The distribution problem was solved.
A Bundle and Institution
From the capital point of view, the greatest investment for a newspaper publisher was the printing plant. The bigger the plant, the bigger the circulation. And the bigger the circulation, the bigger the profit. Printing presses kept getting bigger and faster. An old newspaper man told me that a newspaper is nothing more than a mechanism for depreciating a printing plant.
In order to pay off this investment, publishers not only sold papers but also advertising. To appeal to more readers they added a variety of content types beside news. Sections emerged to cover sports, lifestyle, photographs, opinion, comics, serialized novels, financial information, entertainment/gossip (thus forming the Hollywood star system). This included style pages, advertising and serialized novels. Some of the greatest literature of the time by authors like Mark Twain and Charles Dickens was first published in newspapers. Weekend editions were an obvious product extension.
Get the best analysis on the business of Apple and surrounding industries.
For more articles like this, subscribe to Asymco One →
Thus an institution was born. The newspaper became the primary source of information for the world. It became highly influential and a political (and propaganda) instrument. It came to be the historical record. We still seek out old papers in archives to find out what happened day-to-day in history.
The bigger the bundle the better the business
What made it resilient was its continuous bundling of additional information types, catering to diverse audiences. There was something in a newspaper for everyone. For men and women, from youth to elderly and for rich and poor. Since the cost of adding pages to the paper was marginal, bundling ensured a growing audience. This audience attention could be monetized through newsstand sales, subscriptions and advertising.
Obviously nobody would read a newspaper cover-to-cover. There was far more information than could be consumed and also far more than was interesting to any one person. The style reader probably did not care much for sports. The Hollywood gossip reader did not probably check the finance pages. But most readers of newspapers bought the bundle because it offered optionality.
They did not need to read it all but they could if they wanted. Maybe a scandal in Wall Street would attract the gossip reader occasionally. Maybe a pennant race would attract the style reader. If something exciting happened, you could read all about it.
Consumers were happy with the bundle and the newspaper became a tentpole holding up the information a society needed to function. It swayed elections, influenced public opinion and investigated the bad and the ugly in society. By the turn of the century, more than 9000 newspapers were publishing daily or weekly in the US. A proliferation such that almost every one of the 3140 US counties had more than one.
Unbundling the Newspaper
Then the internet arrived. Gradually, it became possible for every single piece of the bundle in a newspaper to be made available as a web page, free with some display ads. If you were interested in style, there were 1,000 style sites. If you were a sports fan, there would be 1,000 sports sites. News junkies could get the latest news faster online. Financial news was now real-time. No longer did you have to wait for the morning to find out what happened in the markets the day before.
As national and global news could be obtained on web sites, newspapers hoped to continue as sources of local information. That also gradually got peeled away by social media. The result was nothing short of catastrophic. Since 2004, the United States has lost one-fourth (2,100) of its newspapers. Weekday circulation declined 45%, from more than 122 million to 68 million. Consequently, half of the counties only have one newspaper, and two-thirds lack a daily newspaper. The loss of newspapers has resulted in the displacement of 36,000 journalists, including photographers, editors, comic strip artists, and copywriters.
Get the best analysis on the business of Apple and surrounding industries.
For more articles like this, subscribe to Asymco One →
The internet unbundled the newspaper but also the music album, the magazine, the TV channel and almost all forms of entertainment. Unbundling built new empires: iTunes/Apple Music, Netflix, Spotify, YouTube, Facebook.
This unbundling led to a new way of finding opportunities. You could actively look for bundles and once you found one you could ask when and how it would get unbundled. And vice versa. Jim Barksdale summarized it as: “i know of only two ways to make money: bundling and unbundling.”
So much so that if you found a huge bundle that somehow resisted unbundling you would have to investigate it for what’s wrong with it. The classic example is the automobile. As a bundle of many types of trips and configurations into a simple but over serving business model clearly it was ripe for an unbundling. That is what ride sharing (with or without a human driver), and new modes represented by micromobility seek to do.
The AI Bundle
The iPhone is an enabler in this game, capable of either bundling or unbundling depending on the apps which are built for it. Wearables do some of the side quests but the phone is still the core of the personal network.
I would contend that AI is exhibiting a lot of bundling characteristics. We assume one model is good enough for all types of tasks even though there are pitfalls in privacy, security or complexity. Being a prompt mechanic feels like more of a DIY experience. But indications are out there that AI will be unbundled. According to The Information,
“Apple has complete access to the Gemini model in its own data center facilities. Apple can use that access to produce smaller models that power specific tasks or are small enough to run directly on Apple devices so they can run the tasks faster.”
This approach relies on a process called model distillation, where a larger AI model trains smaller versions that can run efficiently on-device. These smaller models reduce the need for cloud processing and improve speed, which fits Apple’s focus on privacy and performance.
“Since Apple has full access to Gemini, its student model can also learn to imitate the internal computations that Gemini uses to arrive at its answers.”
But it does not stop there. The implication of AI as a job destroyer is also challenged by bundling theory.
Jobs as Bundles
We have long been warned that AI would eliminate millions of jobs. One of many forecasts predicts that by 2030, approximately 10.4 million US jobs, 6% of the workforce, will be lost to AI. The underlying assumption behind these figures is simple: if AI can perform tasks that are currently performed by humans, the human workforce will be rendered obsolete.
A new paper by Luis Garicano, London School of Economics, Jin Li and Yanhui Wu, University of Hong Kong suggests it’s not that simple.
The paper argues that jobs aren’t tasks. They’re bundles. For instance, radiologists don’t merely read scans; they interpret edge cases, collaborate with clinicians, and approve decisions that individuals make. Replacing the image-reading aspect doesn’t necessarily equate to replacing the Radiologist job.
Authors draw a line between “weak bundles” and “strong bundles” based on their ability to be split apart. Weak bundles can be easily separated, while strong bundles cannot be without losing their value. You can think of bundling strength as also the degree of modularity/integration. A weak bundle is more modular, a strong bundle is more integrated.
“In weak-bundle occupations, AI automates some tasks and narrows the boundary of the job… In strong-bundle occupations… AI improves performance inside the job, but does not remove the human from the bundle.”
In jobs with weak bundles, AI doesn’t merely replace a task; it fundamentally reshapes the job. The human is left performing tasks that the AI cannot, often in a more specialized and limited capacity compared to the original role.
Once AI takes over a portion of the work, humans cease to divide their time. They dedicate themselves entirely to the remaining tasks, resulting in increased output per worker, i.e. higher productivity. Thus even weak bundles don’t necessarily imply job reduction but increased output.
Artificial intelligence is transforming job roles, rather than eliminating them. While certain tasks may be reallocated and productivity could potentially increase, employment and working hours have remained relatively stable. In many instances, the bundle holds.
In a strongly bundled occupation, AI still improves task-performance, but the human retains both tasks and a larger revenue share. Whether the aggregate labor share rises or falls depends partly on the distribution of coordination costs between tasks within an occupations.
If most employment sits in strongly bundled jobs, the aggregate labor share is more resilient than the task-exposure models predict. The distribution of bundle strength across the occupational structure is therefore a first-order empirical question for understanding AI’s macroeconomic impact.
In other words, the doom/utopia AI outcome dichotomy is false. It depends on the potential for unbundling. If you’re in an integrated (strong-bundle) job, blending judgment, context, and responsibility, AI is sustaining: it will make you faster and better paid. If you’re in a modular (weak-bundle) job, it will shunt you into an integrated role, whereby productivity rises.
The degree of job displacement will depend on the degree of bundling in most jobs across the economy. It may turn out that integrated jobs will be added as the opportunities created by productivity will be market creating.
History is filled with examples of restructuring of ideas and the re-configuration of businesses due to technological change. AI is an example of both enabling and disabling of existing businesses and occupations. Very much the way the Internet turned out to be.
Displacement happens but innovation creates more opportunity and reallocation of effort to better (or at least more) outcomes. The newspaper and internet created markets and jobs. Will AI do the same? Modularity may hold the answer.
Get the best analysis on the business of Apple and surrounding industries.
For more articles like this, subscribe to Asymco One →