Artificial Intelligence's Jevons Paradox: A Look at Contradictory Efficiencies

In the mid-19th century, economist William Stanley Jevons observed a paradox in the coal industry: as technology became more efficient, coal consumption actually increased due to cost reductions and demand elasticity. Fast forward to the present day, and a similar paradox is unfolding in the realm of artificial intelligence (AI).

The rapid advancements in AI technology have led to a compound effect of hyperbolic resource consumption growth. Efficiency gains have not resulted in conservation but rather a surge in AI use, with demand elasticity estimated to be approximately 2-3 times. Even with the adoption of 100% renewable energy, efficient AI remains unsustainable at scale.

The Financial Model of AI includes unit economics, total costs, infrastructure investment, and resource competition. However, the efficiency of AI has followed a paradoxical path: for every 10x efficiency gain, usage increases 100-1000x, new use cases emerge, previously impossible applications become viable, and total compute demand increases.

The Technology Stack of AI includes the model layer, infrastructure layer, application layer, and resource layer, all of which must scale exponentially to meet the demands of AI. Every efficiency gain in AI requires more network capacity, and the Induced Demand: More efficient AI creates more AI use, leading to habitual use and dependency.

The Coding Paradox: More code is written, maintained, and made complex due to AI. GitHub Copilot, for instance, made coding AI efficient, resulting in millions of developers using AI, and total compute demand increased 10,000x. The Content Paradox: Infinite content is created, leading to information overload and quality degradation. The Feature Creep: Every application adds AI, leading to total usage multiplication.

The question isn't how to make AI more efficient, but whether we can survive our success at doing so. As AI becomes more efficient, prices drop toward zero, and demand becomes infinite. OpenAI or other AI firms have not publicly specified precise energy consumption forecasts from 2024 to 2030, but analyses predict that AI data centers' electricity demand could increase dramatically. Deloitte estimates a tenfold rise from 2022 to 2026, reaching 90 TWh, comparable to the energy use of a small country or a large city.

The Jevons Paradox occurs through efficiency gain, cost reduction, demand elasticity, new applications, and total increase in resource consumption. The Efficiency Trap reveals that efficiency is not sustainability, and making something cheaper to use guarantees it will be used more, often overwhelmingly more. The Sustainability Impossibility: Efficiency improvements cannot solve exponential demand growth.

The paradox compounds recursively: AI makes AI development more efficient, leading to better models, more use cases, and more development. The Runaway Train scenario results in a resource crisis by 2030, forced rationing, and societal disruption. We are efficiency-gaining ourselves into a resource crisis, as the demands of AI exceed the capacity of current infrastructures and resources.

The Conscious Constraint scenario involves voluntary limitations, a sustainable AI movement, and managed deployment. The Behavioral Solution requires a fundamental value shift toward digital minimalism, human-first policies, and conscious consumption. The Decision Paradox: Every micro-decision is automated, leading to an exponentially higher number of decisions and increased complexity.

The Distribution Strategy of AI includes democratization, ubiquity, invisibility, and saturation. The launch of ChatGPT in November 2022 resulted in 100M users in 2 months due to its more efficient interface and easier access compared to previous models.

Usage Caps, Progressive Pricing, Resource Taxes, Application Restrictions, and Efficiency Penalties are politically and economically impossible solutions. The Hard Wall scenario results in physical limits being reached, efficiency gains stopping, and system breakdown.

In conclusion, the Jevons Paradox presents a significant challenge in ensuring the sustainable development and deployment of AI. It's crucial to consider the long-term consequences of efficiency gains and to adopt strategies that prioritise sustainability and resource conservation.