The fourth bound is attention. Simon identified it in 1971, writing about the design of information systems: 'What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention.' The formulation treats the relationship between information and attention as metabolic rather than competitive — information is the fuel, attention is what burns, and more fuel does not produce a larger fire if the furnace is the same size. The fourth bound is the bound that no tool relaxes, because it is not a property of the information environment but a property of consciousness. Every conscious agent, however powerful the tools available to her, must still decide where to direct her attention, and that decision is bounded by the same cognitive architecture — roughly four to seven chunks of working memory, a serial processing system that fatigues, a biological attentional apparatus unchanged since the Pleistocene — that bounded the decisions of Simon's 1947 municipal administrators.
The fourth bound is why the relocation thesis holds. The three bounds AI relaxes are all environmental — information exists outside the mind, computation can be externalized, time is a parameter of the interaction. Attention is internal. It is the resource the mind deploys to evaluate everything the environment provides. No expansion of the environment can expand it, because it is the mechanism by which the environment is processed at all.
Simon's 1971 formulation anticipated by decades the empirical findings that would later constitute attention economics. He saw the logic before the infrastructure existed to demonstrate it: information technology would expand without limit while attention remained fixed, producing a systematic mismatch between what systems could deliver and what recipients could process. The social media era made the mismatch visible. The AI era has made it severe.
The fourth bound is what makes the current moment different from every previous technological transition. Each prior transition shifted generation costs while leaving evaluation costs largely unchanged — producing the pattern where better tools produced better output and more exhausted evaluators. AI extends this pattern, but also intensifies it: the gap between what can be generated and what can be evaluated has widened to the point where evaluation quality becomes the primary determinant of output quality, and the organizational and educational structures that develop evaluation capacity become the binding constraint on civilizational capability.
Simon's 1971 paper was delivered at a symposium on the economics of information processing. The audience was computer scientists and economists considering how to design information systems for organizational use. Simon's argument was that these systems should be designed as attention conservers rather than information maximizers — that the design goal should be to deliver the minimum information necessary for the decision at hand, filtering everything else before it reached the decision-maker.
The prescription was largely ignored. The subsequent half-century of information technology has been, with remarkable consistency, a history of systems designed in violation of Simon's principle. Email inboxes, search engines, social media feeds, and now AI tools have all been built as information expanders rather than attention conservers. The fourth bound has been systematically overwhelmed by the three bounds that technology has relaxed.
Information consumes attention. The relationship is metabolic, not competitive — more information does not coexist with the same attention; it depletes it.
Attention is a property of consciousness. It cannot be expanded by any tool that operates on the information environment, because it is the mechanism by which the environment is processed.
The bound is biological. Working memory, serial processing, attentional fatigue — all are features of the cognitive architecture that no foreseeable technology alters.
Design should conserve rather than maximize. Well-designed information systems filter before delivering, compressing the vast space of available information into the small space that bounded attention can productively engage.
The prescription has been ignored. Most information systems, including most AI tools, are designed as expanders, producing the exhaustion that Simon's framework predicts.