The AI Journal

A telephone operator could connect a call with a pair of cords, a socket and a question: “Number, please?” Yet by the early twentieth century, that apparently simple action had become the central difficulty of the American telephone system. The problem was not that operators were incapable of connecting calls. They were exceptionally capable. The problem was that the telephone network had become too dependent on everything they did.

This is why the history of automatic switching is often told backwards. The familiar story begins with a technical deficiency: people once had to connect calls manually, then engineers invented switches, and eventually machines replaced operators. But automatic switching was available surprisingly early. Almon Strowger patented an automatic telephone exchange in 1891. Engineers continued to improve electromechanical switching for decades. Yet the Bell System did not simply install the best available machine and dismiss the operators. For much of the period, it hesitated. The important fact is therefore not that automatic switching existed, but that a technically plausible alternative could remain organisationally unattractive for years.

Milton Mueller’s study of the manual exchange shows why. The switchboard problem was fundamentally one of scale. As exchanges grew, the number of possible connections increased faster than the convenience of simply adding more people and more cords. Operators had to locate subscribers, interpret requests, signal other exchanges and coordinate traffic. The exchange was not merely a machine operated by people; it was an organisational system whose signalling procedures, physical layout and labour practices had evolved together (Mueller 1989). A larger exchange therefore did not behave like a small exchange with more operators. Growth altered the problem itself.

The operators were consequently both solution and constraint. Kenneth Lipartito’s history of the industry shows how technological development and the organisation of operators’ work were inseparable. The introduction of new switchboards changed the skills required, while management simultaneously standardised the movements and procedures by which calls were handled (Lipartito 1994a). The company was trying to make a human system behave predictably enough to deliver an increasingly industrialised service. The more successful it became, the more tightly it had to specify what an operator should do.

That creates an unusual technological relationship. Usually, automation is imagined as the replacement of an inefficient worker by an efficient machine. In the telephone exchange, the machine first had to replace a worker whose performance had already been engineered to resemble a machine. Operators were timed, trained, supervised and disciplined. Their voices were standardised. Their movements were prescribed. Their behaviour towards customers was regulated. Bell Telephone’s celebrated ideal of the polite, controlled “voice with a smile” was not ornamental. It was an attempt to turn a highly variable human interaction into a reliable component of a communications system.

Venus Green’s account of Bell System automation makes the contradiction especially clear. The company had spent decades developing “personal service” as a selling point, particularly as it sought to make the telephone useful beyond a narrow business clientele. The operator helped transform an awkward instrument into a service: she knew how to interpret a request, locate a subscriber and deal with the irregularities that machinery could not easily handle. But that same personalisation became increasingly expensive as the network expanded. Automation therefore required abandoning part of the service model that had helped establish the network in the first place (Green 1995).

The crucial change was not simply the invention of a better switch. It was the changing relationship between the switch and the organisation around it. Lipartito’s analysis of automatic switching shows that the technology was a form of component innovation whose usefulness depended on the investment strategies and existing competencies of telephone companies (Lipartito 1994b). An automatic exchange could not be evaluated as an isolated object. It affected staffing, capital expenditure, maintenance, signalling, subscriber equipment and the geography of exchanges. The machine became valuable when enough of the surrounding system could be reorganised to accommodate it.

This helps explain an otherwise puzzling feature of telephone history: automation did not advance uniformly. Large markets offered conditions in which the fixed costs of automatic switching could be justified. Smaller exchanges could continue to rely on operators because the organisational arithmetic was different. The machine was not simply “better” in the abstract. Its advantage depended upon traffic, scale and the cost of reorganising the network around it.

The human consequences were correspondingly uneven. Operators are sometimes treated as an obsolete occupation whose disappearance illustrates the inevitable march of technology. But their disappearance was not an automatic consequence of invention. It was the result of an institutional decision about how a particular network should be organised. Elinor Carmi has shown how Bell’s attempts to regulate operators increasingly treated their conduct as a source of “noise” within the communications system. The company sought to control not only what operators did but how they behaved, spoke and presented themselves (Carmi 2015). The eventual appeal of automatic switching was therefore partly that it removed a troublesome category of variability from the system.

The strongest objection to this argument is straightforward: perhaps this makes organisational history do too much work. Automatic switching was expensive because it was technically difficult; early systems were unreliable; patents and competing firms complicated investment; and the Bell System had perfectly rational economic reasons to wait until the machines were sufficiently cheap and dependable. On this account, the organisational story is secondary. Technology matured, costs fell, and adoption followed. There is no need to claim that operators themselves constituted the decisive problem.

That objection is powerful, and the claim should be narrowed. Operators were not the sole cause of automation, and organisational considerations cannot substitute for engineering or economics. Lipartito’s work demonstrates that the technical development of automatic switching had its own trajectory, while the later economic evidence shows that adoption depended strongly on local conditions. Feigenbaum and Gross, studying the mechanisation of telephone operation across American cities, find that automation between 1920 and 1940 displaced operators on a large scale but that its labour-market effects depended upon the timing and circumstances of adoption (Feigenbaum and Gross 2024). The machine mattered.

But this does not restore the simple story of technical substitution. It strengthens the more limited claim that technical capability became transformative only when it fitted an organisational problem. A machine capable of replacing an operator was not necessarily a machine worth installing. The relevant comparison was between two complete systems: a growing exchange with people, training, supervision and physical switchboards, and a redesigned exchange with automatic switching, different capital requirements and different maintenance arrangements. Automation won when the second system became more economical and manageable than continuing to enlarge the first.

The distinction matters because it changes what “replacement” means. The operator was not merely occupying a job that happened to be vulnerable to invention. She embodied a collection of functions that the company had to redistribute when the job disappeared. Some were mechanised; some were standardised; some moved elsewhere in the organisation; some vanished with the service model itself. The machine replaced the operator only because the institution had first learned which parts of the operator’s work it could afford to lose.

The telephone exchange therefore offers a sharper historical lesson than the usual morality tale of humans versus machines. Automation did not begin when the machine became clever enough to do the human’s task. It began when the organisation could no longer comfortably remain the kind of organisation that required the human task to be performed in that way. The operator disappeared not at the moment the machine became her equal, but when the network had become ready to become something else.