In their watershed 1943 paper “A Logical Calculus of the Ideas Immanent in Nervous Activity,” Warren McCulloch and Walter Pitts proposed an artificial neural network based on an abstract model of the neuron. They represented their networks in a symbolism drawn from mathematical logic. They also developed a novel diagrammatic system, which became known as “McCulloch–Pitts neuron notation,” depicting neurons as arrowheads. These inscriptive systems allowed McCulloch and Pitts to imagine artificial neural networks and treat them as mathematical objects. In this manner, they argued, “for any logical expression satisfying certain conditions, one can find a net behaving in the fashion it describes.” Abstract neural networks were born as paper tools, constituting a system for writing logical propositions.

Attending to the written materiality of early neural network techniques affords new historical perspective on the notoriously opaque technology driving contemporary AI. I situate McCulloch and Pitts in a material history of logic understood as a set of practices for representing idealized reason with marks on paper. This tradition was shot through with anxiety around the imperfection of human-crafted symbolic systems, often from constraints as mundane as “typographical necessity.” Like the authors they admired, McCulloch and Pitts had to compromise on their notation, forgoing preferred conventions in favor of more easily typeset alternatives. Neural networks’ origin as inscriptive tools offers a window on a moment before the closure of a potent black box, one that is now shaping our uncertain future through ever more powerful, ever more capitalized deep learning systems.

The history of Brazilian IT in the 1970s is influenced by the work of a group of specialists who occupied spaces in university and technocratic circles to propagate ideas of technological autonomy from the Global North. In this sense, there is a consensus that an elite of this group, acting in the Commission for the Coordination of Electronic Processing Activities (CAPRE), managed to establish a national Informatics policy, giving rise to an indigenous computer industry at the end of the decade. However, there is still much to be explored about the dynamics surrounding CAPRE's different activities and the profile of its “ordinary” technicians, considering the breadth of attributions that the small body assumed in structuring the Brazilian IT field. Our proposal is to map them by combining prosopography and identifying the concepts, cultures and practices that guided its actions, such as the ideas of “rationalization” and “technological nationalism” and the establishment of a technopolitical network with the technical-scientific community of the period, including the first political class associations in the field of Computer Science. The paper will discuss the composition of the group and its expertise and trajectories, as well as the main actions of the technicians aimed at subsidizing CAPRE's decision-makers. In this sense, the considerable degree of cohesion between technicians and its leaders ensured that an autonomous path was established for Informatics in the country, even though they were exposed to the authoritarian context of the period, which led to CAPRE itself being extinguished in 1979.

The collaboration with an academic mathematical laboratory or research institute is a recurring pattern in the genesis of early computer manufacturers: it typically involved financial support and exchanges of patents, ideas and employees.

In my presentation I show how knowledge transfer between academic laboratories and private corporations followed different strategies and was shaped by the contingent policies and contexts in which they unfolded. The presentation focuses on three different case studies: the partnership between the Cambridge Mathematical Laboratory and Lyons, begun in 1947; the example of the Mathematisch Centrum in Amsterdam and its 1956 spin-off NV Electrologica; and the case of the Matematikmaskinnämnden and Facit, the Swedish manufacturer of mechanical calculators, which entered the computer business in 1956.
The three case studies are representative of three distinct patterns. First, knowledge transfer by a sponsorship agreement. Funding and supporting the construction of the EDSAC computer enabled the Lyons catering company (a leader in business methods) to appropriate its design to manufacture its LEO Computers. Second, knowledge transfer through a spin-off. Electrologica (the Netherland’s first computer manufacturer) was established by computer scientists of the Mathematisch Centrum as a spin-off to commercialize the computers designed by the institute. Third, the recruitment of technical staff from a center of excellence. Facit entered the computer business by hiring most of its technicians and researchers from Matematikmaskinnämnden (the research organization of the Swedish government). Taken together the three case studies cast light on how R&D diffused in the embryonic computer industry in post-war Europe.

The computational social scientific projects of the Cold War United States are known for their technocratic and militarized aspirations to political command and control. Between the 1960s and the 1980s, Defense officials built systems that sought to replace cognitively limited humans with intelligent machines that claimed to predict political futures. Less familiar are projects that sought to challenge militarized logics of command and control. This paper shares the story of CASCON (Computer-Aided System for Handling Information on Local Conflicts), a State Department-funded information management system that mobilized the qualitative, experiential knowledge and political acumen of diplomats to challenge U.S. Cold War logics, like arms trafficking and unilateral interventionism. The system’s target users—analysts in the Arms Control and Disarmament Agency and the State Department tasked with monitoring conflicts in the global South—were notoriously skeptical of the Pentagon’s militarism and computational solutionism. Yet users ultimately rejected the system because it did tell them what to do! Despite their protestations, they had internalized the command and control logics of policy computing.

CASCON was an early effort to design around the contradictions produced by coexisting fears of human cognitive and information processing limits, on the one hand, and of ceding human agency and expertise to machines on the other. I conclude by arguing that CASCON reflects the simultaneous seduction and fear of the quest to depoliticize politics through technology—an ambivalence that marks contemporary computing systems and discourse as well.