AI is often compared with electricity, steam power, computing, or the internet. Each analogy reveals something important, but none captures AI completely.
Which part of AI does each analogy help us understand?
Steam power amplified physical power; AI amplifies certain cognitive and informational capabilities. Both can reduce the cost of performing tasks and stimulate new machines, industries, and business models. Steam produces measurable physical force. AI produces probabilistic outputs that may be plausible but wrong. Cognitive work is also harder to specify and evaluate than mechanical motion.
Electricity is a flexible input across the economy. AI may similarly become embedded in many products, functions, and industries. Both require complementary investments and organizational redesign before major productivity gains appear. Electricity is standardized and highly reliable: the same current powers many devices predictably. AI behavior varies with models, data, prompts, context, and task conditions.
The factory system transformed productivity by reorganizing tasks, workers, machines, supervision, and workflows. AI may matter most when organizations redesign work around new divisions between humans and machines. Factories concentrated production in physical locations. AI can decentralize capability to individual workers and small firms. It also affects judgment and knowledge work, not only physical production.
Computing provided programmable information processing that could be applied across industries. AI extends this by making software more adaptive and allowing people to direct systems through natural language. Conventional software follows explicit instructions and is usually deterministic. AI learns statistical patterns, produces variable outputs, and can be difficult to interpret or verify.
The internet created a shared distribution layer for information, communication, and commerce. AI can become another platform layer, connecting users, knowledge, tools, services, and transactions. The internet primarily connects people to information created elsewhere. Generative AI synthesizes or creates responses, potentially obscuring sources, introducing errors, and mediating what users see.
Mobile made computing ubiquitous, personal, contextual, and continuously available. AI follows the same path through assistants, embedded features, wearables, and on-device systems. Most AI still depends on centralized data centers, extensive computing resources, and cloud providers. AI systems may also act with greater autonomy than conventional mobile applications.
Cloud made computing available on demand through scalable services and APIs. AI is similarly consumed as a metered service, allowing organizations to access advanced capabilities without building the underlying infrastructure. AI is not just infrastructure. Model behavior, training data, evaluation, and safety policies affect outcomes. Switching AI providers may change the system’s behavior, not simply its price or performance.
Each analogy describes a different layer of the transformation:
- Steam highlights capability amplification.
- Electricity highlights pervasiveness and complementary investment.
- The factory highlights workflow and organizational redesign.
- Computing highlights programmability and information processing.
- The internet highlights distribution and network effects.
- Mobile highlights ubiquity and personalization.
- Cloud highlights scalability and platform dependence.
Together, they suggest that AI’s effects will not come from the model alone. They will emerge from the surrounding system of people, tasks, data, workflows, infrastructure, incentives, and institutions.
But AI differs from all these precedents because it operates on language, judgment, perception, and decision support. Its outputs are probabilistic, its apparent competence can exceed its reliability, and its errors can be difficult to detect.
AI resembles earlier general-purpose technologies in how widely it may spread but differs in how uncertain, interpretive, and human-like its outputs can appear.
We should use historical analogies as lenses, not forecasts. For every comparison, ask:
- What does this analogy illuminate?
- What complementary changes does it suggest?
- What does it cause us to overlook?
History cannot remove AI’s distinctive risks or decide what we should build.
Leave a Reply