Consider the Word Processor

Consider the shipping container. Before it existed, moving goods across the world was a slow, expensive, and labor-intensive process. Cargo arrived at ports in an almost infinite variety of shapes and sizes; crates, barrels, pallets, loose goods, all of which had to be loaded and unloaded by hand, piece by piece. Before the shipping container, a medium-sized cargo ship would often spend more time in port being loaded and unloaded than it did at sea. The cost and friction of moving cargo this way functioned as a practical ceiling on global trade. Goods simply couldn’t move fast enough or cheaply enough to make global commerce economically rational.

Malcom McLean’s standardized steel box, the kind of you see stacked one on top of the other on modern cargo ships, removed that ceiling. Because the containers were uniform in size, the entire process of loading, transit, and unloading could be mechanized and coordinated across ships, ports, trucks, and railways simultaneously. A journey that once required days of manual labor at each end could be completed in hours. The cost of moving goods fell dramatically as a result. It also allowed manufacturing to locate wherever labor and materials were cheapest, because transport costs no longer swamped those efficiency gains. Global supply chains as we know them are substantially a consequence of a simple metal box. It allowed for more trade, cheaper goods, and economies of global scale. This is what a technology genuinely in service of economic productivity looks like. It lowers the cost of doing something useful, expands what it possible, and creates value that did not previously exist.

Consider the word processor. Microsoft Word was released in 1983, and by the end of that decade it had become the standard tool for producing professional documents. It made writing and editing faster, easier, and cheaper. But its more consequential effect was on the economics of revision. Imagine you are a legislative drafter in 1975, working on a seven-page bill. You discover on page three that a definition needs to be added. You cannot simply insert it. The page is fixed. You retype page three entirely, adding in the definition, and then every page after it, all four pages because the text has shifted. Every revision required such a process. Every addition, every deletion, every reorganization meant returning to the point of change and retyping everything that followed.

The cost of revision scaled directly with the length of the document. A five-page bill was manageable. A hundred-page bill was a significant undertaking to revise. A thousand-page bill was effectively impossible to substantially alter once drafted. Brevity was thus a practical necessity imposed by the physical reality of ink on paper. The word processor removed this necessity. And what followed, predictably, was an explosion in the length and complexity of legal text.

In 1955, the United States federal tax code and its accompanying regulations totaled approximately 1.4 million words. Today that figure exceeds 10 million. The average length of a bill passed by Congress has grown from 2.5 pages in 1948 to more than 15 today. The full body of permanent federal law now runs to over 24 million words.

This is not the consequence of an expanding government. During the Second World War, the federal government mobilized the entire American economy, conducted a global military campaign, instituted rationing, and directed the lives of millions of citizens in ways that have no modern parallel. It did all of this with a fraction of the legal language in use today. What has expanded therefore is not the reach of government but its verbosity.

The clearest expression of that verbosity is the modern omnibus bill; single pieces of legislation running to thousands of pages, so large that no legislator reads them in full before voting on them. The Consolidated Appropriations Act of 2023, which funded the entire federal government for a fiscal year, ran to 4,155 pages. Documents of this scale did not exist before they could be drafted, cross-referenced, and coordinated across dozens of offices simultaneously. That capacity arrived with the word processor.

In conventional economic terms, this looks like progress. The word processor freed lawyers and legislative drafters from the labor of editing documents in the physical world. It also freed them from any practical ceiling on document length. More output was possible with the same inputs which is the definition of efficiency.

But output is not the same thing as productivity, or usefulness, and the distinction matters. A longer law is not necessarily a better one. More words do not necessarily produce more justice, more clarity, or more predictable outcomes for the people subject to those laws. In many cases the opposite is true. Length generates ambiguity. Ambiguity requires interpretation. Interpretation requires specialists. The law becomes less a clear set of rules that govern everyone and more a terrain that only experts can navigate. This is not a more productive legal system.

The results are visible in the structure of modern legislative and regulatory text. The word processor made one drafting technique in particular essentially costless: the cascading cross-reference. Rather than stating a rule directly and cleanly, a drafter can now define it by modifying another rule, which itself modifies an earlier one, which references definitions established elsewhere in the text.

Stating something cleanly requires making a decision and making a decision means taking responsibility for it. Cross-reference language defers that responsibility. The meaning is never quite stated. It must be assembled, by someone who already knows where all the pieces are. Before the word processor, maintaining that kind of web of references through every revision was laborious, a practical deterrent against indulging the impulse to obscure. After the word processor, we see the indulgence everywhere.

Take this passage from Section 1 of the Internal Revenue Code, not an obscure corner of tax law, but the very first operative provision, the one that sets individual income tax rates. Open it and this is the kind of language you will find:

“each of the dollar amounts under clauses (i), (ii), and (iii) of subparagraph (B) shall be adjusted in the same manner as under paragraph (1)(C)(i), expect that subsection (f)(3)(A)(ii) shall be applied by substitution ‘2012’ for ‘2016’”

This is not addressed to citizens. It is addressed to people who already live inside the architecture. There is a reasonable defense for this kind of writing. Modern legislation governs genuinely complex systems. Some technical density is unavoidable. But there is a difference between complexity that is necessary and complexity that is merely produced. Legislators have a basic obligation to draft language that at least aspires to be read. The taxpayers funding this machinery are also subject to it. They deserve better than text that requires a specialist to decode what the government has decided they must do.

The downstream consequence of this kind of writing is an economy built around navigating complexity that specific technologies enabled. In 1960, there was one lawyer for every 700 Americans. Today there is roughly one for every 250. The profession did not grow steadily across the century, but rather, accelerated precisely as the tools for producing legal complexity became cheaper and more powerful. Major corporations now maintain armies of in-house counsel and retain outside firms to translate legislative complexity at rates that would have been inconceivable to earlier generations of business leaders. Washington’s lobbying infrastructure, built substantially around knowing where the definitional carve-outs are and how to create new ones, employs tens of thousands of people and moves hundreds of billions of dollars annually.

This is all economic activity, but it is not productivity. It is closer to rent seeking; the capture of value through the manipulation of a system rather than the creation of new value within it. A tax attorney who helps a corporation identify a deduction embedded in the cross-reference structure of the tax code has not made anything. Their fees, however, will be tallied as economic activity. They will appear in GDP figures as output, indistinguishable in the accounting from more productive enterprises.

The aggregate social return on that expertise, across hundreds of firms and millions of hours of similar work, is dubious at best. The word processor did not set out to create this economy. It set out to free people from the drudgery of producing text. What is actually produced, at least in law, was a new and more elaborate drudgery. Some of the most rigorously trained minds in the country now spend their working lives navigating a complexity that would not exist without a specific kind of technological advancement. Surely there are more useful things to do with hands that skilled.

Consider social media. These platforms generate enormous shareholder value, employ highly skilled engineers, and contribute meaningfully to GDP. By every conventional economic measure, they are success stories. What they have not obviously done is make the world more productive, more informed, or more coherent. Their business model is the conversion of human attention into advertising inventory, and the attention consumed does not appear in the accounting as a cost. Entire ecosystems of content creators and influencers have emerged around them, generating real dollars that show up in the economic output figures. Whether any of this constitutes genuine productivity, in the sense of creating something that did not previously exist that improves the lives of people who encounter it, is a much harder question than the GDP figures suggest.

It is not difficult to argue the opposite. Social media platforms have restructured the information environment though which people understand the world, form beliefs, and coordinate with one another. That environment is a form of infrastructure. A functioning society depends on some baseline of shared factual ground, institutional trust, and collective coherence in the same way it depends on roads, bridges, and ports. Social media, engineered to maximize engagement rather than accuracy or understanding, has eroded that infrastructure systematically.

That erosion has real economic consequences, even if they resist precise measurement. A society that cannot agree on basic facts, or trust its institutions, or sustain coherent, collective deliberation is a less productive society in ways that compound over time. None of the damage appears in GDP figures. The advertising revenue does, and it makes the economy look like it’s growing. But a measurement system that records the monetization of incoherence without recording the incoherence itself is not really measuring what it claims to measure and will reliably overstate any benefit a technology is contributing.

Consider artificial intelligence. It is, like the word processor and social media, a digital technology whose outputs are weightless and scalable in ways physical technologies are not. The optimists promise productivity gains of a kind not seen since the industrial revolution. They may be right. But the word processor promised more efficient document production and delivered a tax code no one can read. Social media promised to connect the world and delivered an information environment so degraded that public coherence is collapsing. The pattern is not a guarantee. Digital technologies do not inevitably distort the environments they enter. But they have demonstrated a tendency to generate enormous output, to have that output counted as growth, and to leave the costs of what they displace or degrade unrecorded. There is no particular reason to assume AI will be different.

The shipping container’s productivity was legible because its purpose was unambiguous. It made the movement of goods so efficient at such massive scale that it restructured the entire global economy around the premise that distance would no longer determine cost. You could count the ships, measure the turnaround times, and watch the trade volumes grow. The feedback loop between the technology and its effect on the world was direct and hard to fake. The word processor’s output was, among other things, legal complexity. Social media’s output is engagement. AI’s output will be what? Content? Analysis? Code? Conversation? Synthetic media? All of those things and more. But AI, unlike the shipping container, does not yet answer cleanly the question every technology should be asked; what is the output actually for, and who does it serve? The word processor and social media have provided their answers. What AI leaves behind is still being written. The pattern, however, is not encouraging.