Technology

How Technologies Spread

Technology adoption follows a characteristic pattern that sociologist Everett Rogers mapped in his 1962 study Diffusion of Innovations: a slow start among early adopters, a steep acceleration as the mainstream follows, and a plateau as the remaining holdouts either adopt or are bypassed. This S-curve has held remarkably stable across centuries of technological change - from the printing press to the automobile to the smartphone - though the time scale has compressed dramatically. Electricity took about 35 years to go from 10% to 80% of US households (1908-1942). Smartphones went from 35% to 81% of American adults in under eight years (2011-2019).

Network effects accelerate adoption once a critical mass is reached. A telephone becomes more valuable as more people have telephones. A social media platform becomes harder to leave once your social graph is embedded in it. Robert Metcalfe's observation that the value of a network grows roughly with the square of its users (Metcalfe's law) captures this dynamic, though the precise mathematical relationship is debated. What is less debated is the practical consequence: network effects create winner-take-most dynamics that concentrate entire categories around one or two dominant platforms, from operating systems (Windows/macOS, iOS/Android) to messaging (WhatsApp, WeChat) to search (Google, Baidu).

Adoption curves differ across societies, and the reasons are instructive. South Korea had as many broadband subscriptions per person in 2002 as the EU had in 2008. The reason was not superior technology but deliberate state policy, including public investment and a 1999 certification scheme for broadband-ready buildings. India largely bypassed the landline telephone phase, moving directly to mobile connectivity, with the Unified Payments Interface (UPI) enabling digital payments at scale that parts of the developed world still have not matched. Africa's M-Pesa mobile money system, launched in Kenya in 2007, created a financial infrastructure layer in a context where traditional banking was inaccessible to most of the population. These examples illustrate that technology adoption is shaped at least as much by institutional context, state policy, and existing infrastructure as by the technology itself.

The adoption gap between old and new technologies has compressed from decades to years

Path Dependency and Lock-In

The QWERTY keyboard layout was designed in the 1870s, by the usual account to stop mechanical typewriter keys from jamming (some historians point instead to telegraph operators' needs). You are almost certainly typing on that layout now, long after the mechanical constraint vanished. Whether QWERTY is actually suboptimal is debated. The economist Paul David made it the textbook case of lock-in in 1985; Stan Liebowitz and Stephen Margolis replied in 1990 that the case for the rival Dvorak layout's superiority was overstated. Either way, once a technology is widely adopted and infrastructure is built around it, switching costs can lock in designs for decades or centuries, whether or not better alternatives exist.

Economist W. Brian Arthur developed the theoretical framework for this dynamic: increasing returns to adoption. The more people use a particular standard, the more complementary products, skills, and infrastructure are built around it, making it increasingly costly to switch. Railroad gauges are a canonical example. The 4 feet 8.5 inches (1,435 mm) standard gauge used across most of Europe and North America was not chosen because it was optimal - it was the gauge George Stephenson happened to use in his early British railways, and it became the standard through network effects and interoperability requirements. Russia settled on a broader five-foot gauge (now 1,520 mm), a choice that still shapes logistics and trade infrastructure across the former Soviet space today. The popular explanation - that it was deliberately chosen to obstruct invading armies - is repeated far more confidently than the evidence supports; the contemporaneous documentation points mainly to the recommendation of the American engineer hired to build the Moscow-St. Petersburg line, and the military advantage was noticed afterward rather than designed in.

Path dependency operates at much larger scales. Japan's Shinkansen (bullet train) system, launched in 1964, created an infrastructure-led development model that shaped urban geography, property values, and commuting patterns for generations. Once cities were built around high-speed rail nodes, the system became self-reinforcing. The US made a different infrastructure choice in the same era - the Interstate Highway System (1956) - which similarly locked in patterns of suburbanization, automobile dependency, and land use that persist today, even as the costs (congestion, emissions, maintenance debt) have become increasingly visible. Neither choice was "wrong" in any simple sense; both created path dependencies that constrain options decades later.

Gauge choices made nearly two centuries ago still constrain logistics infrastructure across continents

How Tools Reshape Cognition

Plato recorded one of the oldest known critiques of a new communication technology: in Phaedrus, Socrates warns that writing will weaken memory, because people will rely on external marks rather than internal recall. He had a point: literate societies do develop different memory strategies than oral ones. Few would argue, though, that writing was net negative. The pattern has repeated with many major cognitive tools since: the technology offloads a cognitive function, the offloaded capacity atrophies, and new capacities develop in its place.

Calculators shifted numeracy from mental computation to problem-setup and estimation. Navigation offers the cleanest evidence in both directions. Eleanor Maguire's studies of London taxi drivers found that acquiring "the Knowledge" - the years of route memorization the license demands - was associated with a measurably larger posterior hippocampus, and that the effect grew with time in the job. Running the comparison the other way, work by Louisa Dahmani and Véronique Bohbot found that heavier lifetime GPS use predicted worse spatial memory, and in a small three-year follow-up, those who used GPS more declined more steeply. The two lines of evidence are separate studies on separate populations rather than a single before-and-after, and neither settles causation, but together they point the same way: the navigating gets done somewhere, and where it gets done shows up in the person.

Betsy Sparrow's 2011 research at Columbia, published as "Google Effects on Memory," found something similar for facts. When people expect to look information up later, they remember where to find it rather than the thing itself. That is a shift from content memory to source memory. It is rational rather than lazy, and it changes which cognitive skills matter most. One caveat: another of the paper's four experiments could not be replicated in 2018 or 2020, so hold the specific effect loosely. The broader phenomenon of cognitive offloading is not in doubt.

The current transition with large language models raises this pattern to a new scale. When a tool can draft text, summarize research, translate languages, and generate code, the load-bearing skills shift from production to evaluation: can you tell whether the output is accurate, well-reasoned, and complete? This is a real cognitive shift, not a trivial one. Optimists hear in "AI is making us stupid" the anxiety Socrates expressed about writing, and argue that societies which adapt their education systems to new demands do better. Skeptics reply that this tool can do the reasoning itself, not only the remembering, and early studies point both ways. High school students who practiced math with an unrestricted chatbot did worse once it was removed than peers who never had it; physics students taught by a carefully designed AI tutor learned more than they did in an active-learning class. What the new demands will ultimately be is still uncertain.

Unintended Consequences

Chlorofluorocarbons (CFCs) were developed around 1930 as a safe, non-toxic refrigerant, replacing ammonia and sulfur dioxide, which were dangerous. For decades, CFCs seemed like a pure improvement. Nobody anticipated that they would accumulate in the stratosphere and catalytically destroy stratospheric ozone. The warning by Mario Molina and Sherwood Rowland in 1974, made urgent when the Antarctic ozone hole was reported in 1985, led to the Montreal Protocol (1987) - widely considered the most successful international environmental agreement in history. The lesson is not "scientists should have known." The atmospheric chemistry was new. The lesson is that complex systems produce consequences that are fundamentally unpredictable from the properties of the technology alone.

Tetraethyl lead, added to gasoline from the 1920s to prevent engine knocking, was known to be toxic at the time of its introduction but was defended as safe at environmental concentrations by General Motors, which had developed it. Geochemist Clair Patterson, while studying the age of the Earth through lead isotope ratios, discovered that atmospheric lead levels had risen dramatically since the introduction of leaded gasoline. His decades-long campaign against the lead industry eventually contributed to the phase-out of leaded gasoline, completed globally only in 2021. A 2022 study estimated that over 170 million Americans alive in 2015, more than half the population, had high blood lead levels as young children, mostly during the peak decades of leaded gasoline. It put the cost at an average of 2.6 IQ points per American.

Social media algorithms, designed to maximize engagement, have been accused of amplifying polarization, radicalization, and mental health harm. The evidence is more contested than popular accounts suggest. In experiments run with Meta during the 2020 US election, teams including Brendan Nyhan (Dartmouth) changed what users saw for about three months and found no measurable change in polarization. Jonathan Haidt (NYU) argues that smartphones and social media caused a global mental health crisis among teenagers, and one quasi-experimental study found that Facebook's rollout across US colleges worsened students' mental health. Researchers like Andrew Przybylski (Oxford) and Candice Odgers (UC Irvine) reply that the measured effects are small and mostly correlational. A fair summary is that social media algorithms produce real effects on information environments, but the magnitude and causal mechanisms are still being worked out, and confident claims either way outrun the evidence.

Technologies can produce consequences that nobody could predict from their design alone

Who Controls the Tools

Political scientist Langdon Winner asked in 1980: "Do Artifacts Have Politics?" His answer was yes. The design of a technology embeds power relationships. Robert Moses's low-clearance overpasses on Long Island's parkways kept buses off the main routes to his public beaches and, in Robert Caro's famous account, kept away lower-income and minority residents who relied on public transit. Some historians dispute the intent and the effect: commercial vehicles were barred from parkways anyway, and buses could reach the beaches by local roads. Whether or not this specific example holds up to scrutiny, the general point is well-established: technologies create differential access, and the choices embedded in design are often political choices, whether or not they are recognized as such.

The economist Carlota Perez offers a framework for understanding how technology and power interact at civilizational scale. In her model, technological revolutions (the mechanized factory, steam and railways, steel and electricity, oil and the automobile, information technology) follow a recurring pattern: an installation phase of speculative investment and disruption, followed by a turning point (usually a financial crash), followed by a deployment phase where the technology becomes embedded in institutional structures. Each installation phase concentrates wealth and power in the hands of whoever controls the new infrastructure. The railroad barons of the 19th century were replaced by the oil and automobile industrialists of the 20th, who are being replaced by the platform and AI companies of the 21st. In her account, though, the full benefits of each revolution arrive only after institutions are rebuilt around it.

Bi Sheng invented movable type in China in the 1040s, roughly four centuries before Gutenberg. But the technology had different consequences in different institutional contexts. In Europe, the printing press loosened the Catholic Church's hold on written knowledge, helped the Protestant Reformation spread, widened vernacular literacy, and ultimately contributed to the scientific revolution. In China, where the imperial examination system already provided a meritocratic pathway and the logographic writing system made movable type less practical, the same technology did not produce the same institutional disruption. Technology does not operate in a vacuum. Its effects are shaped by the institutional environment it enters. Today's parallel case is the divergence between the Western and Chinese tech ecosystems: different institutional contexts are producing different technology architectures, different data governance regimes, and different distributions of power, even when the underlying engineering is similar.

The Innovation Dilemma

Joseph Schumpeter's concept of creative destruction - the process by which new technologies render old ones obsolete, destroying existing firms and industries while creating new ones - remains the standard framework for understanding technological change in market economies. The automobile destroyed the horse-drawn carriage industry. Digital photography drove Kodak into bankruptcy in 2012. Streaming destroyed video rental. The process is economically productive in aggregate but painful for the specific workers, communities, and firms that are displaced.

Clayton Christensen's The Innovator's Dilemma (1997) refined the story: incumbents fail not because they are incompetent but because they rationally serve their best existing customers, leaving space for disruptive technologies that initially seem inferior but improve rapidly. This framework has been enormously influential in business strategy. It has also been criticized. Historian Jill Lepore argued in The New Yorker (2014) that Christensen's case studies were selectively chosen, that many of his "disrupted" companies actually survived, and that the theory functions more as an ideology of innovation than a reliable predictive framework. Christensen replied that her essay broke the scholarly rules it accused him of breaking. The debate is instructive: critics argue that innovation theories often serve as justifications for particular policy choices (deregulation, venture capital culture, weak labor protections), while defenders answer that a theory can be misused and still describe how incumbents fail.

The Soviet Union provides a useful counterpoint. Its space program - from Sputnik (1957) to the first human in space (1961) - demonstrated that centralized state direction could produce cutting-edge technology. But the same system that concentrated resources on strategic priorities consistently failed to produce consumer goods that matched Western quality or variety. The lesson is not that markets are universally better at innovation; it is that different institutional arrangements produce different kinds of technological output, and no single system optimizes for all dimensions simultaneously.

Where Analysts Disagree

Has innovation slowed down? Robert Gordon (Northwestern) argues in The Rise and Fall of American Growth that the innovations of 1870-1970 (electricity, indoor plumbing, the automobile, antibiotics) were qualitatively more transformative than anything since, and that the digital revolution has produced disappointing productivity gains compared to those earlier revolutions. Erik Brynjolfsson (Stanford) counters that digital gains are real but poorly measured by GDP statistics, and that AI may produce a productivity acceleration that vindicates optimism. The data so far is mixed. Productivity growth in most developed economies slowed after the mid-2000s, though US growth has picked up since 2019 (about 2% a year, against 1.5% in 2005-19). Whether the slowdown was a measurement problem, a temporary lag, or a lasting plateau is still unresolved.

Techno-optimism versus techno-skepticism. Joel Mokyr, a 2025 Nobel laureate in economics, traces two centuries of sustained growth, and the escape of vast numbers from poverty, to technological progress in societies open to new ideas. Marc Andreessen's 2023 "Techno-Optimist Manifesto" goes further, calling technology the main driver of growth and the precautionary principle "deeply immoral." Daron Acemoglu and Simon Johnson counter in Power and Progress (2023) that technology's gains have spread widely only when workers and institutions forced them to. Evgeny Morozov argues that Silicon Valley optimism ignores distributional consequences - who benefits and who is harmed. Shoshana Zuboff's The Age of Surveillance Capitalism argues that digital technology has enabled a fundamentally new form of economic extraction based on behavioral prediction. These are not just academic positions; they map onto real policy choices about AI regulation, antitrust, data governance, and public investment.

AI safety: how worried should we be? The spectrum runs from existential risk (Yoshua Bengio, Geoffrey Hinton) through near-term harm focus (Timnit Gebru, Emily Bender - bias, environmental cost, the treatment of workers) to skepticism about existential risk (Yann LeCun, Andrew Ng - current systems are not close to the capabilities that would justify existential concern). All positions are held by credible researchers; the disagreement is real and unresolved. The field does not yet have the conceptual tools to confidently predict the trajectory; the camps differ on whether that uncertainty is itself a reason for caution.

Every tool reshapes the world it enters, in ways its creators did not fully foresee. The printing press was designed to reproduce books; it helped produce the Reformation, mass literacy, and eventually the scientific revolution. The Interstate highways were built mainly for civilian traffic, with defense a secondary aim; they accelerated suburbanization, automobile dependency, and the reshaping of American geography. Understanding technology is not about predicting the next gadget. It is about recognizing that the tools a society builds become the infrastructure that builds the society back.