Reading route · 12 selected stops
History of Computing
A computer brings together ways of representing a task, carrying out instructions, storing information and exchanging messages. This route follows twelve selected publications, experiments and systems to ask how those problems became connected, and what each development actually demonstrated.
From expressing a rule to running a program
Lovelace’s notes invite us to imagine a machine carrying out a sequence of operations. Boole’s work asks how reasoning can be expressed mathematically, while Turing’s model makes rule-following computation precise and exposes limits to what a general procedure can solve. Reading these together helps separate three questions: how to describe a task, what counts as a computational procedure, and whether such a procedure exists for the task. These publications address related questions without constituting a single machine-building project.
The Manchester Baby and FORTRAN bring the discussion to the relation between instructions and working systems. The Baby tested a machine that could execute a program held in electronic memory. FORTRAN’s compiler translated mathematical expressions into machine code, with practical efficiency central to the project. Compare the questions each story answers: where instructions are stored, how people express a calculation, and how a machine carries it out. A useful reading habit is to identify whether a claim concerns a mathematical model, a language or an implemented system.
Devices, communication and learning address different problems
The transistor and integrated-circuit stories concern electronic components and their connections. Shannon’s information theory instead examines how messages can be measured and reliably transmitted under channel constraints. ARPANET then gives a concrete case of computers exchanging packets, while the Web concerns navigating linked information over the internet. These are different layers of a system: an improvement in a component does not by itself create a network protocol, and reliable transmission does not establish that a message is true or meaningful.
Backpropagation and Transformer add questions about how a model is trained and how it represents relationships in sequences. Read the 1986 paper alongside its stated earlier history, and the 2017 paper alongside its machine-translation experiments. Training a model toward an objective leaves questions about its results and evaluation. The route ends with a selected architecture paper; it leaves open many uses of computing and many research directions. Logic, programming, hardware and communication remain useful subjects in their own right, rather than stages destined to culminate in a single form of AI.
The reading timeline
1843 (translation and Notes)
Lovelace explores programmable machines
Begin with the distinction between a published account of a programmable machine and a completed working computer. The notes provide a way to ask what operations a program might express while retaining the wider collaboration around the Analytical Engine.
1854 (Laws of Thought published)
Boole brings algebra to logic and probability
Look at reasoning treated as an object of algebraic analysis. Keep Boole’s mathematical work separate from later adaptations to digital logic and from Shannon’s distinct work on communication.
1936 (computable-numbers paper)
Turing models the limits of computation
Ask what it means to follow a general computational procedure, and why some tasks lack one. The story also makes room for Church’s related result and separates the mathematical model from the construction of electronic computers.
1947 (point-contact transistor experiments)
A point-contact transistor amplifies a signal
Move from formal rules to a physical device that can amplify a signal. Distinguish the point-contact experiments, their later public announcement and the subsequent development of other transistor designs.
21 June 1948 (first program run)
Manchester’s Baby runs a program held in electronic memory
Follow a program into electronic memory and then into execution. This experimental test of stored-program computing has a specific scope, and reading that scope is more useful than assigning an unrestricted title of first computer.
1948 (paper published in two parts)
Shannon gives information and communication a mathematical framework
Compare the question of transmitting a message reliably with the question of understanding its content. Information measured through probability does not directly measure truth or meaning.
1957 (compiler delivered to users)
FORTRAN translates mathematical expressions into efficient machine code
See how a language and compiler changed the work of expressing numerical calculations. The delivery of a usable compiling system is a distinct milestone within the longer development of programming languages.
1958–1959 (experimental circuit and planar integration concept)
Integrated circuits bring components and connections onto semiconductor material
Compare the experimental circuit with the subsequent planar integration proposal. Keeping those stages and their contributors separate helps explain why putting components on one material also required solving connection and manufacturing problems.
1969 (early nodes connected)
ARPANET links research computers through packet switching
Follow computers becoming connected through packet switching, including the need to debug early communication. Keep this research network distinct from the later development of the internet as a whole and from the Web.
9 October 1986 (Nature paper published)
A backpropagation paper shows multilayer networks learning internal representations
Connect the chain rule from calculus with the adjustment of weights in a multilayer network. Read the paper as an influential account with earlier precedents, rather than treating its publication as the invention of every part of the method.
1989–1991 (proposal, implementation and dissemination)
The Web moves from a hypertext proposal to a working service
Shift from connecting computers to linking and browsing documents. The sequence from proposal to implementation and wider circulation helps explain how the Web developed as a service over the internet.
12 June 2017 (first preprint posted)
The Transformer organizes sequence modelling around attention
Examine how attention was organized within a sequence-model architecture and tested on machine translation. Keep this paper’s result distinct from later chat applications, the earlier history of attention and claims about general intelligence.
Questions to take further
Why are ARPANET and the Web separate stops?
They make different problems visible. ARPANET’s story concerns connecting research computers and exchanging packets; the Web’s concerns linked information, browsers and servers operating over the internet. Reading them separately preserves the contributions of networking work, shared protocols and information-service development without assigning the whole internet a single birthday.
How should I interpret the connections between these stories?
Use them to compare questions and dependencies, rather than assume every story directly caused the next. A mathematical model, an electronic experiment, a compiler and a learning architecture establish different things. Turing’s limits do not disappear when a device becomes faster, and the Transformer paper does not make AI an inevitable endpoint of computing.