Jevons Paradox and the AI Age

Why Efficiency May Increase Workload, Information Volume, and Decision Pressure

Artificial intelligence is usually introduced with a promise: less effort, faster work, fewer repetitive tasks, and more time for higher-value thinking. That promise is not false. AI can draft a report in minutes, summarise a document almost instantly, translate text, classify data, generate options, and automate administrative work that previously consumed hours.

But there is a problem hidden inside that promise.

AI may not give organisations less work. It may give them the ability, and eventually the pressure, to create more work.

That is the uncomfortable possibility. A tool designed to reduce burden may, at the level of the whole organisation, increase the total amount of writing, reporting, analysis, monitoring, verification, and decision-making expected from human beings. The individual task becomes easier, but the system expands. The worker writes faster, so more documents are requested. The analyst processes more information, so more information is collected. The executive receives more options, more dashboards, more warnings, and more scenarios, but still carries the responsibility for judgement.

This is where Jevons Paradox becomes useful. It gives us a disciplined way to think about why efficiency does not always lead to reduction. Sometimes efficiency expands the very activity it was expected to contain.

What Jevons Saw

William Stanley Jevons made his argument in the nineteenth century in relation to coal. Britain was becoming more efficient at using coal, especially through improved steam-engine technology. A simple assumption would have been that greater efficiency would reduce coal consumption. If engines used coal more efficiently, surely less coal would be needed.

Jevons saw the opposite danger.

More efficient use of coal made coal-powered activity cheaper and more productive. That encouraged wider industrial use. Coal did not simply become a resource used more sparingly. It became the energy base of a larger industrial system. Greater efficiency helped expand the scale of coal-dependent activity.

That is the paradox. Efficiency at the unit level can produce expansion at the system level.

This does not mean efficiency is useless. Nor does it mean that every efficiency gain automatically increases total consumption. The point is more careful than that. When a technology makes an activity cheaper, faster, or easier, people often do more of it. Sometimes the increase is small. Sometimes it offsets part of the saving. Sometimes it is large enough that total use rises.

In modern terms, this is often discussed as a rebound effect. If a car becomes more fuel-efficient, the driver may save fuel. But if driving becomes cheaper, the driver may travel more often, choose longer journeys, or live farther from work. Some of the saving disappears. In extreme cases, total use may increase.

Jevons Paradox is therefore not a slogan against efficiency. It is a warning against naïve efficiency thinking.

The question is not only: “How much does this technology save per task?”

The deeper question is: “How does the whole system behave once the task becomes cheaper?”

The AI Translation

AI changes the cost of cognitive production.

It makes it cheaper to write, summarise, classify, translate, analyse, generate images, produce presentations, prepare meeting notes, create marketing content, and draft policy documents. These are not small changes. They affect the basic work of modern organisations, especially knowledge-based institutions.

But once cognitive production becomes cheaper, demand for it may rise.

A department that once asked for one monthly report may now expect weekly reports. A board that once received a short risk summary may now receive multiple dashboards, scenario packs, AI-generated forecasts, and daily updates. A manager who once requested a single briefing may now ask for ten variants: short version, long version, legal version, financial version, strategic version, public version, internal version, and “just one more” refined version.

The work has become easier at the level of production. But the organisation has quietly raised the volume of expected production.

This is the AI version of Jevons Paradox: AI may reduce the cost of producing information while increasing the burden of deciding what information matters.

That distinction is crucial. Producing information is not the same as understanding it. Generating options is not the same as choosing wisely. Automating analysis is not the same as carrying accountability.

More Reports, Not Fewer

The most immediate example is writing.

AI can help draft reports, emails, policies, summaries, meeting notes, proposals, and board papers. That can save time. But it also changes expectations. If writing becomes faster, the organisation may demand more written justification for everything.

A simple decision that previously required a conversation may now require a written rationale. A project update that was once delivered monthly may become weekly. A manager who once accepted a short note may now expect a full briefing because “AI can produce it quickly.”

The burden does not disappear. It changes shape.

The human being now spends less time starting from a blank page, but more time checking, correcting, refining, approving, and explaining machine-assisted text. The document may be produced faster, but the responsibility for its accuracy remains human.

In that sense, AI does not abolish bureaucracy. Poorly governed, it may accelerate bureaucracy.

More Data, Not More Understanding

The same problem appears in information analysis.

AI allows organisations to examine larger datasets, detect patterns, classify material, identify anomalies, and generate summaries from complex information. Used carefully, this is valuable. In finance, logistics, medicine, intelligence, cyber security, and public administration, AI can help reveal information that would otherwise remain buried.

But there is a danger. If analysis becomes cheaper, organisations may analyse everything simply because they can.

More data is collected. More dashboards are created. More indicators are tracked. More alerts are produced. More “insights” are generated. The organisation becomes information-rich, but not necessarily judgement-rich.

The central problem shifts. It is no longer only, “Can we obtain information?” It becomes, “Which information is decision-relevant?”

That is a harder question. It requires judgement, context, experience, and discipline. AI can support that process, but it cannot remove the need for it.

More Content, Less Attention

Marketing and communication provide another clear example.

AI makes it cheap to produce social-media posts, newsletters, images, video scripts, adverts, and campaign material. The result is predictable. More content will be produced. More variations will be tested. More messages will compete for attention.

At first, this looks like productivity. A small team can now produce what previously required a larger team. But at the level of the public information environment, the outcome may be saturation. Everyone can publish more, so everyone must compete harder to be heard.

The scarce resource is no longer production capacity. The scarce resource is attention.

This is another Jevons pattern. The efficiency gain expands the activity. The cost of content falls, so the volume of content rises. But human attention does not expand at the same rate. The system becomes louder, not necessarily wiser.

More Monitoring, More Control Questions

AI also reduces the cost of monitoring.

Organisations can monitor transactions, communications, behaviour, operational systems, cyber activity, performance indicators, movement patterns, and risk signals. In many cases, this has legitimate value. Fraud detection, infrastructure protection, safety monitoring, cyber defence, and emergency management can all benefit from better detection.

But cheaper monitoring creates pressure to monitor more.

This raises serious governance questions. What is the lawful purpose of monitoring? Who has access to the outputs? How long is data retained? What happens when a system wrongly flags a person, transaction, or behaviour? Who checks the machine? Who has authority to override it?

The issue is not whether monitoring is always bad. It is not. The issue is whether institutions understand that AI can turn limited monitoring into continuous monitoring almost by default.

Efficiency expands capability. Capability expands expectation. Expectation becomes normal practice.

That is why governance must be built before the system becomes too large to control.

More Academic Output, More Verification

Education and academia face their own version of the paradox.

AI can help students draft, summarise, translate, explain, and structure written work. It can help researchers scan literature, organise notes, and prepare early drafts. These uses can be legitimate when handled transparently and ethically.

But the verification burden increases.

Teachers and universities must now ask new questions. Did the student understand the work? Are the sources real? Are the citations accurate? Has the text been generated, copied, edited, or genuinely argued? Does the submission reflect learning, or only output?

AI reduces the effort required to produce academic-looking text. It therefore increases the need to verify understanding.

That is a classic Jevons-type effect. The task of producing text becomes easier. The system responds by generating more text. But the harder task — judging quality, authorship, understanding, and integrity — remains human.

More Options, More Decision Pressure

The most serious effect appears at executive level.

AI can generate options, scenarios, forecasts, risks, recommendations, and decision trees. This can improve decision support. A leader can see more possibilities, test assumptions, and compare alternative courses of action more quickly.

But more options do not automatically produce better decisions.

Sometimes they create hesitation. Sometimes they create the illusion that one more model run will remove uncertainty. Sometimes they diffuse responsibility because the recommendation appears to come from a system rather than from a person. Sometimes they give leaders too much material and too little clarity.

Executives do not need infinite options. They need decision-ready understanding.

This is why intelligence must not be confused with information. Intelligence, in a serious executive sense, is not the accumulation of reports, dashboards, warnings, and data streams. Intelligence is decision-relevant understanding disciplined for accountable judgement before consequence.

That definition matters in the AI age. AI may support intelligence, but it cannot assume judgement. It may accelerate analysis, but it cannot absorb accountability. It may increase awareness, but it cannot carry consequence.

Defence and Security: The Hardest Case

In defence and security, the paradox becomes sharper because time, consequence, and accountability are compressed.

AI-enabled systems can assist with sensing, surveillance, intelligence processing, cyber defence, logistics, targeting support, and command dashboards. These capabilities may improve awareness and speed. But they may also increase the number of signals, alerts, classifications, and recommended actions reaching commanders and decision-makers.

The commander may know more, but also be required to decide faster.

This is not a minor administrative concern. In high-consequence environments, more information can become a burden if it is not filtered, interpreted, prioritised, and governed. A warning that cannot be acted upon is not yet useful intelligence. A dashboard that displays everything may clarify nothing. A recommendation without explainable assumptions may create false confidence.

The danger is not that AI will make defence decision-making too simple. The danger is that it may make decision environments faster and denser while leaving human beings responsible for consequences they had less time to understand.

The Governance Problem

If AI creates expansion, then adoption is not enough. Organisations need discipline.

They need to decide what AI should produce and what it should not produce. They need to decide which outputs require verification, which decisions require human authority, and which uses are inappropriate. They need audit trails, accountability rules, escalation thresholds, and clear decision rights.

A serious organisation should ask basic questions before deploying AI into important work:

• What decision is being supported?

• What information is being used?

• What assumptions are embedded?

• What uncertainty remains?

• Who checks the output?

• Who has authority to decide?

Who is accountable if the decision is wrong?

These are not technical questions only. They are governance questions. They determine whether AI becomes useful decision support or simply a faster machine for producing unchecked organisational noise.

The point is not to reject AI. That would be unrealistic and unhelpful. The point is to govern the expansion that AI makes possible.

Conclusion: The Scarce Resource Is Judgement

Jevons Paradox teaches a simple but powerful lesson: efficiency does not end the problem. It changes the scale of the problem.

AI will make many tasks faster. That much is clear. But the strategic question is what happens next. If every report becomes easier to write, will organisations ask for fewer reports or more? If every dataset becomes easier to analyse, will leaders receive clearer intelligence or more dashboards? If every decision can be surrounded by forecasts, options, and simulations, will judgement improve or become buried under output?

The central question is not whether AI can make work faster. It can.

The central question is whether human institutions can govern the expansion that follows.

AI may reduce the cost of producing information, but it may increase the value of disciplined judgement. In the AI age, judgement — not production — may become the truly scarce resource.

Final line:

Efficiency may multiply activity; only judgement can discipline consequence.

Selected Sources and Evidence

William Stanley Jevons, The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines, 1865. Jevons’ original argument on coal efficiency is the historical foundation for what later became known as Jevons Paradox: the warning that improved efficiency can expand total use when it lowers the effective cost of an activity. A useful public edition and contextual note is available through Yale Energy History.

Centre for Research into Energy Demand Solutions, “The rebound effect and the Jevons’ Paradox: beyond the conventional wisdom.” CREDS explains the rebound effect as the set of mechanisms through which expected savings from efficiency improvements may be partly or substantially offset by increased use. This supports the article’s distinction between unit-level efficiency and system-level expansion.

Steffen Lange, Johanna Pohl and Tilman Santarius, “A multi-level typology of rebound effects and mechanisms,” Energy Research & Social Science, 2021. This peer-reviewed article provides a contemporary framework for understanding rebound effects and mechanisms beyond the original coal example, supporting the article’s wider application of efficiency-rebound logic.

Microsoft Work Trend Index, “Breaking Down the Infinite Workday,” 17 June 2025. Microsoft’s workplace data and analysis show the intensification of digital work through meetings, emails, chats, interruptions, and after-hours communication. This supports the article’s claim that digital productivity tools can increase rather than reduce the total communication and coordination burden.

McKinsey & Company, The State of AI: Global Survey 2025, 5 November 2025. McKinsey’s survey reports widespread AI adoption but continuing difficulty in scaling value, with high-performing organisations more likely to use disciplined management practices, including human validation of model outputs. This supports the article’s argument that AI productivity depends on governance, workflow integration, and decision discipline rather than adoption alone.

National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), 2023. NIST’s framework identifies governance, mapping, measurement, and management as core functions for responsible AI risk management. This supports the article’s claim that organisations need audit trails, verification rules, escalation thresholds, decision rights, and accountability structures before AI expands beyond effective control.

OECD, OECD AI Principles. The OECD principles promote trustworthy, human-centred AI that respects human rights and democratic values. They support the article’s emphasis on human accountability, governance, oversight, and institutional responsibility when AI systems influence consequential decisions.

Author workflow disclosure

This article was produced through an AI-assisted but human-directed workflow. AI support was used for accessibility assistance, article structuring, language refinement, source-discovery prompts, revision planning, and conversion of editorial comments into specific amendments. The author retained responsibility for the argument, accepted or rejected suggested changes, checked the logic of the claims, and remained accountable for the final text. AI-generated material was not treated as empirical evidence, and synthetic or illustrative examples were not presented as observed data.

Image note

The image accompanying this article is AI-generated and is intended for illustration purposes only.

CONTACT

© 2026 Dr Danie Adendorff. All rights reserved. Rumbls.com is an independent analytical blog.

TERMS AND CONDITIONS

DISCLAIMER

COOKIE POLICY

PRIVACY POLICY