9.4: The Future Arrived Early

Artificial Intelligence, Work, and the Next Global Divide

For years, conversations about automation relied on a familiar image.

The robot comes for your job.

Sometimes it was an assembly-line machine replacing factory workers. Sometimes it was a self-driving truck. Sometimes it was a vaguely menacing humanoid doing something that, judging from most demonstrations, would have taken an actual human about twelve seconds.

Then generative artificial intelligence arrived, and the conversation changed.

Suddenly the occupations assumed to be safest from automation were not necessarily those requiring advanced education or creativity. AI systems could generate text, summarize documents, produce images, write code, analyze data, translate languages, draft contracts, tutor students, imitate voices, design presentations, and perform pieces of knowledge work that had long been treated as distinctly human.

The future did not arrive exactly as predicted.

It rarely does.

Exposure Is Not the Same as Replacement

Public discussion about AI often collapses several different questions into one.

Can AI perform part of a job?

Can it perform most of a job?

Will employers redesign the job around it?

Will the technology increase worker productivity, reduce staffing, create entirely new occupations, or simply add one more tool workers are expected to master without receiving more time or pay?

Those outcomes are not interchangeable.

Recent labor research increasingly emphasizes task transformation rather than immediate wholesale job elimination. The International Labour Organization's global analysis of generative AI estimates that roughly one quarter of workers worldwide are employed in occupations with some degree of exposure to generative AI, but the report stresses that exposure does not mean that an occupation will disappear. In many cases, AI is more likely to alter the composition of work than replace the worker entirely (ILO, 2025).

That distinction matters enormously.

Jobs are bundles of tasks.

A teacher does not simply “deliver information.” A nurse does not simply record symptoms. A lawyer does not simply draft language. A journalist does not merely rearrange facts into sentences. Occupations involve judgment, relationships, accountability, tacit knowledge, physical presence, institutional rules, emotional labor, and ethical decisions that may be difficult to automate even when some component tasks are technically replicable.

AI may therefore remove pieces of work while increasing the importance of other pieces.

The real transformation occurs when institutions decide what humans should still be paid to do.

Productivity for Whom?

Technological change is often justified through the language of productivity.

If workers can produce more in less time, economic theory suggests that organizations can become more efficient and societies potentially wealthier.

But there is an awkward question hiding inside that sentence.

Who gets the productivity gain?

If an AI system allows one employee to complete work that previously required three, several outcomes are possible. Workers might receive higher wages because their productivity increased. Firms might shorten working hours while maintaining output. Organizations might expand services without expanding staffing. Or employers might eliminate positions and capture most of the savings as profit.

Technology does not decide among these possibilities.

Institutions do.

Labor markets do.

Political systems do.

Ownership structures do.

That should sound familiar.

Throughout this book, we have repeatedly encountered technologies or economic systems presented as neutral tools whose consequences turned out to depend heavily on who controlled them. AI is no exception.

The Next Digital Divide

For decades, scholars and policymakers have discussed the digital divide, initially focusing on unequal access to computers and internet connectivity.

Artificial intelligence complicates that divide.

Access still matters, of course. Reliable electricity, broadband infrastructure, computing capacity, affordable devices, and digital literacy remain unevenly distributed across countries and within them. But the emerging AI divide may involve several additional layers: who possesses the computing infrastructure required to build large models, which languages receive high-quality representation, whose data becomes training material, which countries host data centers, who owns intellectual property, and which education systems can prepare people to use AI productively rather than merely consume it.

The distinction between access and capacity is crucial.

A country may have millions of citizens using AI tools while possessing little influence over the technologies themselves.

That is participation without control.

Language Is Infrastructure Too

One of the least visible dimensions of AI inequality is language.

Large AI systems perform best in languages with abundant digitized text, strong commercial markets, large user bases, and substantial technical investment. English enjoys enormous advantages in this environment, while many Indigenous, regional, and lower-resource languages are represented by far smaller datasets.

That matters for more than convenience.

Language carries cultural knowledge, historical memory, social categories, humor, metaphor, political meaning, and ways of understanding the world. If AI systems increasingly mediate education, government services, translation, search, communication, and knowledge production, then uneven language representation can become another mechanism through which some communities participate in the digital future on someone else's terms.

Technological inequality can therefore become cultural inequality.

And cultural inequality, as we saw much earlier in this book, rarely stays cultural for long.

The Cloud Lives Somewhere

AI is often described using remarkably weightless vocabulary.

The cloud.

Virtual assistants.

Digital intelligence.

Machine learning.

The language can make the technology sound almost immaterial, as though AI floats somewhere above the physical world.

It does not.

AI requires semiconductor manufacturing, server farms, electrical grids, cooling systems, water, fiber-optic infrastructure, mineral extraction, transportation networks, and enormous capital investment. Data centers occupy land. Chips depend on highly specialized manufacturing ecosystems. Batteries and electronics require minerals whose extraction carries environmental and labor consequences.

The digital economy has a physical geography.

That geography links AI directly back to environmental politics, labor, development, and inequality.

A student asking whether AI will transform education in California is therefore connected, however indirectly, to semiconductor manufacturing in East Asia, mineral extraction in Africa and Latin America, energy systems in multiple regions, globally distributed data-labeling labor, and regulatory decisions made in capitals around the world.

There is our old friend again.

The supply chain.

It never really leaves.

Invisible Workers Behind Intelligent Machines

Artificial intelligence is frequently described as though machines simply learned everything by themselves.

They did not.

Human beings label images, review harmful content, rank responses, transcribe audio, annotate datasets, test outputs, moderate platforms, translate language, and perform countless forms of digital labor that make automated systems appear more autonomous than they actually are.

Much of this work has been outsourced internationally.

Workers in lower-income labor markets may perform repetitive or psychologically difficult tasks for companies and platforms headquartered elsewhere, often through subcontracting structures that make responsibility difficult to trace.

The pattern is remarkably familiar.

New technology.

Old question.

Who does the hidden work?

Figure \(\PageIndex{1}\): The hidden global infrastructure of AI. AI systems may appear as simple digital tools such as chatbots, apps, or automated platforms, but they rely on a much larger material and social infrastructure, including data, human labeling and moderation, engineering, chips and data centers, electricity and water, critical minerals, and global labor and supply chains. The figure also highlights key political questions surrounding AI: who owns it, who works on it, who bears the environmental costs, and who captures the gains. Source: Author-created via promt generation using ChatGPT, CC BY 4.0.

Figure \(\PageIndex{1}\) This infographic is titled “The Hidden Global Infrastructure of AI.” It shows AI as a layered system rather than a standalone digital tool. At the top of the image is the part labeled “What you see,” which includes a chatbot displayed on a laptop screen, an AI app, and an AI system. This top layer represents the visible, user-facing side of artificial intelligence. Below that, the visual moves downward through a series of stacked layers showing what makes AI possible. The first layer is Data, represented by database imagery and digital information systems. The second layer is Human Labeling and Moderation, represented by people working at computers. This layer emphasizes that AI depends on human workers who sort, label, review, and moderate content. The third layer is Models and Engineering, represented by coders and technical workers building and maintaining AI systems. The fourth layer is Chips and Data Centers, represented by servers, computer chips, and large-scale computing infrastructure. The fifth layer is Electricity and Water, represented by power lines, utility infrastructure, and water use, showing that AI requires major physical resources to operate. The sixth layer is Critical Minerals, represented by mining equipment and extracted earth materials, emphasizing the role of mineral extraction in producing the hardware behind AI. The bottom layer is Global Labor and Supply Chains, represented by ships, shipping containers, trucks, and industrial facilities, showing that AI depends on worldwide manufacturing, transport, and labor systems. On the right side of the graphic, four questions appear vertically: Who owns it? Who works on it? Who pays the environmental cost? Who captures the gains? At the bottom, the figure concludes with the statement: “AI has a global material footprint.”

AI Governance Is Global Governance

Because AI systems cross borders so easily, regulating them presents a difficult problem.

A company may be headquartered in one country, train models using data gathered globally, run infrastructure in several jurisdictions, sell services in dozens more, and produce consequences that national regulators struggle to contain.

Different governments are already developing sharply different regulatory approaches. The European Union has pursued a relatively comprehensive risk-based framework through the AI Act. The United States has historically relied more heavily on sectoral governance and executive action, although policy remains politically contested. China has developed its own regulatory structure around algorithms, generative AI, data, and platform governance. UNESCO, the OECD, and other international bodies have attempted to establish broader principles for responsible AI.

These frameworks differ, but the underlying issue is the same.

Technological power has become transnational faster than political accountability.

That should make us nervous.

Not because the robots are about to take over.

Because humans are very good at building enormously influential systems before deciding who should be responsible when something goes wrong.

The More Important AI Question

So will AI destroy jobs?

Some.

Create jobs?

Almost certainly.

Transform many more?

Very likely.

But Global Studies pushes us toward a harder question.

Will the gains and disruptions be distributed evenly?

There is little reason to assume they will be.

Countries with strong digital infrastructure, research institutions, highly skilled workforces, large technology firms, abundant capital, and reliable energy systems may be positioned to capture substantial productivity gains. Countries lacking those advantages may adopt AI as consumers while remaining dependent on technologies developed elsewhere. Within countries, highly educated workers may gain new tools while other workers experience intensified monitoring, reduced bargaining power, or displacement.

Gender may shape these effects because clerical and administrative occupations with relatively high AI exposure employ large numbers of women in many economies. Language matters. Geography matters. Education matters. Disability can shape both opportunities and exclusions. Labor institutions matter.

In other words, AI does not replace the global inequalities we have spent this book studying.

It plugs into them.

And occasionally amplifies them.

The Future of Work Is Really the Future of Bargaining

This is why the future of work cannot be understood solely as a technological question.

It is a political and institutional question.

Who owns the systems?

Who controls the data?

Who decides how workers are evaluated?

Who receives the productivity gains?

Who bears the cost of retraining?

Who is protected during transitions?

Who can challenge an automated decision?

Who gets to say no?

Those questions are not fundamentally about artificial intelligence.

They are about power.

Technology changes the terrain on which power operates, but it does not eliminate the struggle over how benefits and risks are distributed.

That struggle will shape the global economy students enter after leaving this classroom.

Which raises a fairly practical question.

After an entire book about systems, inequalities, institutions, cultures, conflicts, borders, food, health, gender, climate, and now artificial intelligence, what exactly are you supposed to do with all of this?