I’m a firm believer that unless you know what the origin of something is you can never really understand it. For example if a friend is acting out of character what triggered their behaviour? You know when they are in or out of character, if you are paying attention, that is. Likewise, people often use language that gives away what they mean without actually saying what they mean.
Humans are pattern matchers in every sense of the word. The unknown constant is intuition, but for the sake of covering everyone lets keep it really simple. A dog is a dog. We know that because we learnt it as a child, likewise a flower, a cup, a TV, a car and so on and so forth, you get the picture. AI is also a pattern matcher. Interestingly this is where etymology, technology and history collide. There are so many interesting similarities around the vocabulary used to describe artificial intelligence. It contains words whose histories stretch far beyond modern computing.
Me being me, I had a bit of a random thought process last night and ended up writing this about AI or rather IA. It’s not quite a full A-Z but it does have a lot of the good bits.
This is not a tools first make you a gazillionaire by Thursday post if that’s what you are expecting wrong feed love. For those still here, let’s get into it.
I love etymology, always have, so it made sense to dig into the terminology surrounding AI which, interestingly, stems from Latin manuscripts, Greek language and philosophy, and Arabic mathematics. Several of the concepts central to computing were formalised during the middle of the twentieth century; others are thousands of years older. As I’m a bit OCD about naming conventions, the below is organised alphabetically and examines some of the historical origins and connections.
A: Algorithm
The English word algorithm ultimately derives from the name of the ninth-century mathematician Muḥammad ibn Mūsā al-Khwārizmī. Al-Khwārizmī worked in Baghdad during the Abbasid period. A Latin translation of a work associated with his methods of calculation rendered his name in a form such as Algoritmi.
Terms derived from that Latinisation developed over centuries into the modern word algorithm. The word algebra has a separate connection to al-Khwārizmī. It derives from the Arabic al-jabr, appearing in the title of his mathematical work Al-Kitāb al-Mukhtaṣar fī Ḥisāb al-Jabr wal-Muqābala. The two modern computing terms therefore have historical roots in medieval mathematics transmitted from Arabic scholarship into Latin Europe.
So when you’re a slave to, or banging on about the “Algorithm” you now know who and what you are referring to.
B: Bletchley Park
Bletchley Park, the famous code breaking unit in the Second World War, immortalised in numerous films, provides a useful illustration of the distinction between codes and ciphers.
A code can substitute one meaningful unit for another according to an agreed system.
A codeword, for example, can represent a phrase, place, operation or instruction.
A cipher however transforms elements of a message according to a defined system, normally involving a key.
The German Enigma machine used a polyalphabetic substitution cipher generated through electromechanical rotors. Bletchley Park’s work on Enigma therefore involved cryptanalysis of a cipher system, rather than simply discovering the meanings of individual codewords. That differentiation is key because code, cipher and encryption are often used interchangeably in ordinary language despite describing different mechanisms, but they are not the same.
C: Cipher and Cypher
Cipher derives from the Arabic ṣifr, meaning zero or empty. The word migrated via Medieval Latin and European languages before acquiring its later association with secret or encoded writing, anyone paying attention during The Matrix may recognise the connection. The character Cypher uses an established alternative spelling of cipher, and his name fits his role particularly well.
The Matrix itself is code representing a simulated reality. Cypher understands how to interpret that code: the symbols displayed on the screens represent people, objects and events within the simulation. His character therefore sits directly inside the distinction between symbol and meaning, code and interpretation, representation and reality.
There is a second linguistic connection. Cipher has also historically meant zero, reflecting right back to its origin in Arabic ṣifr. The name Cypher therefore carries associations with both encoded information and zero, although the Wachowski brothers Larry and Andy now the Wachowski sisters Lana and Lilly would have to be asked specifically about their reason for choosing the names of the characters. Does anyone know them to ask?
Just a little side note - I played pool with Laurence Fishburne one night in 1998, in a bar in Sydney when they were filming the first Matrix. I didn’t win, perhaps because he’d been loaded up by Tank beforehand and had an unfair advantage, but I digress.
C: Code and Codex
Code derives from Latin codex or caudex. Codex came to mean a book or manuscript composed of bound leaves and subsequently an organised collection of laws, rules or authoritative writings. (That really disturbs the cynic in me, as I’m not a fan of rules.) Over time, code acquired meanings involving systematic collections of rules and systems in which information was represented according to agreed conventions. For example:
- legal code
- criminal code
- building code
- code of conduct
Those are fundamentally rules governing human behaviour, but code also developed the sense of a system of signs or symbols governed by conventions, where one thing represents another. So the same word eventually encompasses:
- rules governing humans → legal and behavioural codes
- rules governing representation → codes in which symbols represent other information
- instructions governing machines → computer code. Computer code is a much later development of the word. The conceptual progression is: codes were written to tell humans how to behave long before we wrote code to tell machines how to behave. OpenAI uses Codex as the name of its software-engineering system. The linguistic relationship between codex and code predates OpenAI, and I haven’t found an OpenAI source confirming that this etymology influenced its choice of name, however that cynic in me is surfacing again but who knows? Claude is also widely reported to have been named after Claude Shannon, although I have not found a primary Anthropic source confirming it, which brings me nicely to………
D: Dartmouth
In 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon prepared a proposal for a summer research project at Dartmouth College [3]. Their 1955 proposal coined the term artificial intelligence and stated the research premise that aspects of learning and intelligence might be capable of sufficiently precise description to permit machine simulation. The Dartmouth project is important not because intelligent machines were first imagined there, but because Artificial Intelligence became the name attached to an organised field of research.
E: Enigma
Enigma derives from Latin aenigma, from Greek ainigma, meaning a riddle or an obscure saying: something whose meaning is hidden and has to be deciphered. Centuries later, Enigma became the name of the German electromechanical cipher machine most famously used during the Second World War. The name was remarkably appropriate. Enigma transformed readable plaintext into ciphertext that appeared unintelligible unless the recipient possessed the correct settings needed to reverse the transformation. Its encrypted messages became one of the major targets of Allied cryptanalysis. Work by Polish cryptanalysts before the war was fundamental to understanding and attacking Enigma, and that work was subsequently developed at Bletchley Park, where Alan Turing, Gordon Welchman and many others contributed to methods and machines used to recover Enigma settings. So the word completes an unusually literal journey:
- Greek ainigma → something whose meaning is hidden
- enigma → a riddle or mystery requiring interpretation
- Enigma → a machine designed to hide meaning
- cryptanalysis → the attempt to recover that meaning
The machine did not give the ancient word a new metaphor. Someone gave a machine built to create riddles, the ancient word for riddle, fascinating.
E: Engelbart
This is a totally different strand of computing history concentrated on increasing human capability. In 1962, Douglas Engelbart published Augmenting Human Intellect: A Conceptual Framework [4]. Engelbart described a programme concerned with increasing a person’s capacity to approach complex problems and improve their ability to derive understanding and solutions. The related expressions intelligence amplification and later intelligence augmentation are commonly abbreviated to IA. AI and IA should not be treated as simple historical opposites. They developed through overlapping fields and intellectual traditions. They do, however, identify different questions:
AI: how might machines perform functions associated with intelligence?
IA: how might machines increase human intellectual capability?
I: Information
Claude Shannon a mathematician and electrical engineer published A Mathematical Theory of Communication in the 1948, Bell System Technical Journal [5]. Shannon developed a mathematical framework for communication and information. Among the concepts associated with information theory are the quantitative treatment of information, entropy, channel capacity and the effects of noise on communication. Shannon’s work did not create human information or communication. It provided a mathematical theory through which aspects of information transmission could be analysed quantitatively.
I: Information Technology
Ten years later, in 1958, Harold J. Leavitt and Thomas L. Whisler published Management in the 1980’s in Harvard Business Review. They used the expression information technology to describe a group of emerging technologies that they expected to affect organisational management. Their conception included techniques for processing information, statistical and mathematical methods used in decision-making, and computer simulation. The expression IT therefore emerged in a management context broader than the modern everyday meaning of an organisational IT department.
P: Pattern
Humans recognise patterns, but we also construct patterns. We identify recurring relationships in the world, group things into categories, give those categories names and develop systems for communicating what we have recognised. Language itself is an example: we learn that very different animals can all belong to the category dog, and that the sound and written word dog can be used to refer to that category. Mathematics, musical notation, maps, calendars and written language all allow us to organise recurring relationships into systems that other humans can recognise and use.
AI is also concerned with patterns, but not in the same way. Large language models learn statistical relationships across enormous quantities of human-produced data. The patterns they identify therefore sit on top of patterns, categories and representations that humans have already created.
R: Representation
Humans recognise patterns. Representation is how we make them transferable.
The Roman numeral X represents the quantity ten. The same quantity can also be represented as 10 using the Western Arabic digits of the Hindu-Arabic numeral system, ١٠ using Eastern Arabic digits, or simply ten in written English.
Different symbols, different systems, same quantity. The quantity has not changed, its representation has, and it works in the opposite direction too. The same graphical symbol can represent completely different things depending upon the system in which it appears. X can represent ten in Roman numerals, a letter in the Latin alphabet, an unknown or variable in algebra, multiplication in some mathematical notation, a chromosome, or a position on a map. Meaning therefore does not reside in the shape X alone. Its meaning depends upon the representational system in which the symbol is being interpreted. Computing depends extensively upon these layers of representation. Numbers can represent characters. Binary states can represent numbers. Sequences of bits can represent text, images, sound or instructions. Programming languages can represent operations that are ultimately translated into instructions a processor can execute. The computer therefore operates across representations created according to defined systems. That distinction becomes particularly important when we reach artificial intelligence. A machine’s ability to process a representation should not automatically be confused with a human understanding of what the representation means. The distinction is simple but fundamental:
the thing → its representation → the rules for interpreting that representation.
Humans have been building those layers for thousands of years, computing builds upon them.
S: Standardisation and ASCII
ASCII stands for American Standard Code for Information Interchange. ASCII was developed in the United States through standards work during the early 1960s, with the first ASCII standard published in 1963. It established numerical representations for a defined set of characters and control functions.
For example, in ASCII decimal notation: 65 represents A. That relationship exists because the encoding standard defines it. ASCII did not arise in isolation. Earlier telegraph, teletype and character-encoding systems preceded it and influenced the development of standardised digital character representation. ASCII was subsequently incorporated into a much larger history of character encoding. Modern computing predominantly relies upon Unicode, commonly encoded using UTF-8, to represent characters across many writing systems.
S: Shannon
Claude Shannon connects several strands of this history. In his 1937 master’s thesis, Shannon demonstrated how Boolean algebra could be applied to electrical switching circuits. That work became foundational to digital circuit design. His 1948 communication theory subsequently provided a mathematical framework for information transmission. We’ve already discussed his contribution to the Dartmouth proposal alongside McCarthy, Minsky and Rochester. Shannon therefore appears at important points in the histories of:
- digital switching
- information theory
- early artificial-intelligence research
Anthropic’s AI system is called Claude. It is widely reported that the name honours Claude Shannon as I mentioned earlier, but that remains unconfirmed.
T: Turing
In 1950, Alan Turing published Computing Machinery and Intelligence in the journal Mind [6]. The paper begins: “I propose to consider the question, ‘Can machines think?’” Rather than attempting to define both machine and thinking directly, Turing developed the idea of an imitation game as an operational way of approaching the problem. What later became known as the Turing Test grew from this argument. Turing’s paper predates the Dartmouth project and the formal adoption of Artificial Intelligence as the field’s name.
X: X
X perfectly demonstrates the distinction between a symbol and what it represents, because few symbols have accumulated quite so many different meanings.
Within Roman numerals, X = ten: a precise, known quantity. Within algebra, x can represent an unknown or variable quantity: something whose value is yet to be determined. Within the Latin alphabet, X is simply a letter.
X can mark a precise location, as in X marks the spot, yet elsewhere signify something unknown, as in Planet X or the X factor. It can represent a kiss, a chromosome, a crossing point, multiplication, cancellation or prohibition. Plus since 2023, X has been the name applied to the social-media platform formerly known as Twitter.
The same symbol can therefore represent the known and the unknown, a precise location or an unidentified one, a number and a letter, connection and cancellation. The physical mark remains recognisable. What changes is the meaning assigned to it within the system in which it appears.
So representation means a symbol does not have to change for its meaning to change, the system interpreting it does. This principle is fundamental to computing, where interpretation depends upon agreed structures determining how representations are processed.
Z: Zero
Zero brings us almost full circle.
Both zero and cipher ultimately trace back to the Arabic ṣifr, meaning empty or nothing. But the mathematical concept and notation have an older history: the positional use of zero developed in India, before being transmitted and developed through Arabic scholarship and later reaching Europe.
So one of the most fundamental ideas in modern mathematics travelled across languages, cultures and centuries:
Indian mathematics → Arabic scholarship → European mathematics → modern computation.
Zero is particularly interesting because it represents nothing, while simultaneously making an enormous amount possible. In a positional number system, its presence determines the value of the digits around it: 1, 10, 100 and 1000 are very different quantities because of where zero appears.
And in binary, the foundation of digital computing, the entire representational system is constructed from just two symbols:
0 and 1.
So we end the alphabet with perhaps the strangest representation of all: humanity invented a symbol for nothing, and that symbol eventually became one half of the basic language through which digital computers represent almost everything.
AI → IA
While writing this I found the chronology provided an interesting pattern. 1948: Shannon formalises a mathematical theory of communication and information. 1950: Turing asks how machine intelligence might be operationally considered. 1955: McCarthy, Minsky, Rochester and Shannon propose the Dartmouth project using the term Artificial Intelligence. 1956: the Dartmouth summer project takes place. 1958: Leavitt and Whisler describe Information Technology. 1962: Engelbart publishes Augmenting Human Intellect. None of these events independently created modern artificial intelligence, together with developments in mathematics, logic, linguistics, electronics, statistics, computing and many other fields, they form parts of its intellectual history. Modern large language models add another layer. They process tokens and learn statistical relationships from very large quantities of data, and their trained parameters encode learned statistical structure. That should not be confused with a literal memory containing every piece of information encountered during training. AI products can separately provide conversation histories, persistent memory, retrieval systems, databases and external tools. Those capabilities should therefore be distinguished from the underlying language model. Which brings me back to two letters. AI for Artificial Intelligence: and IA for Intelligence Augmentation. Artificial Intelligence asks how machines can perform capabilities associated with intelligence. Intelligence Augmentation asks how computational systems can increase human capability. My own published research arrived at this argument from a completely different angle. The Feed Loop observed what can happen when AI-generated instructions begin governing subsequent AI behaviour: patterns can propagate until a rule that appeared explicit quietly stops working [1]. Gates, not prohibitions, indicated the corresponding design principle: telling an AI what not to do is less robust than placing a checkpoint in the system that its work must pass [2]. Neither study set out to investigate AI versus IA, yet both brought me back to the same place. The history of computing is a history of humans creating representations, rules, codes and systems that allow machines to do increasingly extraordinary things with information. Perhaps the mistake is assuming that the logical conclusion of that history is removing the human from it. We have spent thousands of years teaching machines how to work with our representations of the world. The question now is not simply: How intelligent can we make the machine? It is: How intelligently can we design the relationship between the human and the machine? Maybe the future was hiding in the acronym all along. AI → IA.
References
[1] Samantha Maeer, “The Feed Loop,” Zenodo, https://doi.org/10.5281/zenodo.20474271
[2] Samantha Maeer, “Gates and Prohibitions,” Zenodo, https://doi.org/10.5281/zenodo.21206719
[3] John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude E. Shannon, “A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence,” 31 August 1955, https://ebiquity.umbc.edu/paper/html/id/1199/A-Proposal-for-the-Dartmouth-Summer-Research-Project-on-Artificial-Intelligence
[4] Douglas C. Engelbart, “Augmenting Human Intellect: A Conceptual Framework,” October 1962, Doug Engelbart Institute, https://www.dougengelbart.org/pubs/augment-3906.html
[5] Claude E. Shannon, “A Mathematical Theory of Communication,” Bell System Technical Journal, vol. 27, July and October 1948, https://archive.org/details/bstj27-3-379
[6] Alan M. Turing, “Computing Machinery and Intelligence,” Mind, vol. 59, October 1950, pp. 433-460, https://archive.org/details/MIND--COMPUTING-MACHINERY-AND-INTELLIGENCE