Lovelace and Turing: Can a Machine Originate?
Two arguments, 107 years apart, about what computers can be — a COMP 1150 case study
- Who: Augusta Ada King, Countess of Lovelace (1815–1852), and Alan Mathison Turing (1912–1954) — a mathematician working in 1840s London and a mathematician working in 1930s–50s Britain
- What: Lovelace claimed the new mechanical computer “has no pretensions whatever to originate anything.” A century later, Turing named that claim “Lady Lovelace’s Objection” and argued it was wrong.
- Where / When: London, 1843 (Lovelace’s Notes on the Analytical Engine); Cambridge, Bletchley Park, and Manchester, 1936–1954 (Turing’s universal machine and his essay “Computing Machinery and Intelligence”)
- Why it matters: Their argument is the one we are having right now about large language models. The terms have not changed in two hundred years; the machines have. The case is the clearest single window into how computer science braids together mathematics, engineering, and philosophy.
- Concepts at play: algorithm, the universal machine, the Turing test, what it means for a computation to “originate” something
The Case
In the summer of 1843, Augusta Ada King, Countess of Lovelace, finished a job no one had asked her to do. She had agreed to translate, from French into English, a short article by an Italian military engineer named Luigi Menabrea, describing a machine her friend Charles Babbage had designed but never built — the Analytical Engine. Menabrea’s piece was about thirty pages. By the time Lovelace finished, the translation was thirty pages and her own “Notes by the Translator,” labeled A through G, were sixty-five (Lovelace 1843).
Lovelace was twenty-seven. She had been schooled in mathematics since girlhood. Her mother, Anne Isabella Milbanke, had pushed her into algebra and Euclid for a reason. The reason was to keep her from the temperament of her father, the poet Lord Byron, who had abandoned the family shortly after her birth. The plan worked uncomfortably well. Lovelace became a real mathematician, tutored by Mary Somerville and Augustus De Morgan. She described her own way of working as poetical science — the synthesis of imagination and rigor she felt the Analytical Engine made possible.
The Engine itself, on paper, was a marvel. Babbage had specified a steam-powered, brass-and-iron machine with:
- a store (memory)
- a mill (a processor)
- conditional branching
- loops
- a way of programming it with punched cards borrowed from the Jacquard loom
In Note G of her translation, Lovelace published what is usually called the first algorithm intended to run on a machine. It was a step-by-step procedure for computing the Bernoulli numbers, laid out as a table of operations. Then, in the same Note, she stopped and made a claim that would outlast everything else she wrote:
The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform. … Its province is to assist us in making available what we are already acquainted with.
She wrote this in the same room where she had just designed a program. She did not see the two acts as in tension. To Lovelace, designing the procedure was the original work; the Engine’s execution was the carrying-out. The Engine, in her phrase, could not “anticipate analytical relations.” Originating belonged to us.
Lovelace did her mathematics inside a world that did not formally count her as a scientist. The Royal Society would not admit a woman for the better part of a century. She corresponded with De Morgan as a private student. She published her Notes under her initials, “A. A. L.” Her contemporaries who could have cited her work generally did not. None of this kept her from doing it. It is worth saying. The rest of her century — and a fair part of the next — treated her contribution as Babbage’s footnote. Lovelace died of cancer in 1852, at thirty-six. The Analytical Engine was never built in her lifetime. Her Notes lay mostly unread for almost a century.
When her argument was next picked up — and named — its respondent was Alan Turing. In 1936, still a graduate student, Turing wrote a paper titled “On Computable Numbers, with an Application to the Entscheidungsproblem” (Turing 1936). He asked what it could mean for a number to be computable at all. His answer was a thought experiment. Imagine a machine with:
- an unlimited paper tape
- a tiny set of rules
- a head that can read and write symbols
Anything a human could compute by following rules, this machine could compute. And — the move that mattered — another such machine could be built that read a description of any first machine and simulated it. There was no “machine for Bernoulli numbers” and a separate “machine for solving differential equations.” There was just the universal machine, and the program you fed it.
The Second World War interrupted everything. Turing spent it at Bletchley Park. He designed the electromechanical machines that broke German Enigma traffic. By most historians’ estimates, this shortened the war by two to four years. After the war he returned to the question of thinking machines. In 1950 he published, in the philosophy journal Mind, an essay titled “Computing Machinery and Intelligence” (Turing 1950). Most of it is a defense of one provocation. The question can machines think? is too vague to be useful. We should replace it with a game.
The game has been called the Turing test ever since. It runs by teletype:
- A human judge in one room exchanges typed messages with two players in other rooms.
- One player is a human; the other is a machine.
- The judge tries to tell them apart.
- If the judge cannot do better than chance, then whatever “thinking” amounts to, the machine has done it.
In the middle of that essay, under the heading “Lady Lovelace’s Objection,” Turing names her by title and quotes the famous lines back. He treats her objection as the strongest of nine he means to answer. It is strongest because it sounds reasonable. Surely a machine can only execute what we have told it. Machines take me by surprise with great frequency, he writes in reply. The reason they do, he says, is not that they have escaped the rules we gave them. It is that the rules we gave them have consequences we did not have the time or patience to predict.
Turing did not get to keep arguing. In 1952 he was prosecuted under a Victorian statute against “gross indecency” — a relationship with a man. He was given a choice between prison and chemical castration. He chose the latter. On 7 June 1954, he was found dead at forty-one of cyanide poisoning. The inquest ruled suicide. The British government issued an apology in 2009 and a posthumous royal pardon in 2013. In 2017 it extended that pardon to thousands of other men convicted under the same statute. It is not the moral of the case. But it is the world Turing was making his argument inside, and the case is poorer if we pretend otherwise (Hodges 1983).
One historiographical note is worth flagging before we leave the lives behind. The popular story has Lovelace as “the first programmer,” full stop. Careful archival work — most thoroughly the 2018 study by Hollings, Martin, and Rice — finds that the truth is more collaborative and more interesting (Hollings et al. 2018):
- Babbage drafted earlier procedures.
- Lovelace’s Bernoulli table was her own.
- Her broader claims about what such a machine could and could not do were genuinely hers.
History is contested. That, too, is worth knowing on the way in.
The argument did not die with either of them. It became routine. It is taught in introductions to computing. It is debated in artificial-intelligence research. It was refought in the 2020s, when large language models began producing prose that interrogators sometimes could not distinguish from human writing. We have built things that pass restricted versions of the Turing test. Meanwhile critics produce a version of the Lovelace Objection updated to apply to them: they can only recombine what they’ve been shown. Whose machine is it, after all.
How It Worked
Behind the philosophy are two technical ideas, both small enough to write in a few lines of Python.
The first is what Lovelace meant by an algorithm. An algorithm is a finite list of unambiguous steps that, followed exactly, produces a specific result. Lovelace’s Note G gave one for the Bernoulli numbers — a sequence of fractions \(B_0, B_1, B_2, \ldots\) that appear all over mathematics. Her full procedure was several pages long, written as a numbered table of operations for the Engine. The shape of an algorithm, though, is simple enough to fit in a few lines. Here is one any reader can follow by hand:
Procedure: factorial(n)
Computes n! = 1 × 2 × 3 × … × n.
1. Let result ← 1.
2. For i from 1 up to n:
result ← result × i.
3. Return result.Try it for \(n = 5\). Start with result = 1. Multiply by 1 (still 1), then by 2 (2), then by 3 (6), then by 4 (24), then by 5 (120). The procedure has no idea what factorials are. It just follows the rules. Lovelace’s Bernoulli procedure had the same shape — a numbered list of operations on numbered variables — but was longer, and produced a more interesting sequence: \(B_0 = 1\), \(B_1 = 1/2\), \(B_2 = 1/6\), \(B_3 = 0\), \(B_4 = -1/30\), \(B_5 = 0\), and so on. Following the rules is, in Lovelace’s terms, the Engine’s province. Designing the rules — which operations, in which order, to make the right pattern fall out — was hers.
The second idea is Turing’s. He noticed that the rules themselves are a kind of data. If a machine can read symbols and follow rules, then the rules of any other machine can be written down as symbols and read by a single, fixed machine. That machine can then pretend to be any of the others. We call this a universal machine, and every laptop, phone, and server you have ever used is one. The full mathematics is intricate; the idea is not:
Procedure: interpret(program)
Input: program — a list of instructions, written as data.
1. Start with an empty stack.
2. For each instruction in program, in order:
- If the instruction is a number, push it onto the stack.
- If the instruction is "+", pop two numbers and push their sum.
- If the instruction is "*", pop two numbers and push their product.
3. Return the top of the stack.
Examples:
interpret( [3, 4, "+", 5, "*"] ) → 35 # (3 + 4) × 5
interpret( [6, 7, "*"] ) → 42 # 6 × 7This tiny procedure is a working — if minimal — universal machine. Hand it one list of instructions and it does arithmetic. Hand it a different list and it does different arithmetic. In principle, hand it the right (much larger) list and it could simulate any other computer, including itself. That is what Turing’s 1936 paper proved formally: one machine, parameterised by its program, can stand in for all of them.
These two ideas — that a procedure is something we can write down precisely (Lovelace), and that the procedure itself is data the machine can read (Turing) — are the whole basis of modern computing. They also frame the argument the case turns on. If a program is “just” data being interpreted by a universal machine, the question can the machine originate something the programmer didn’t? stops being silly. It becomes the most interesting question in the field.
The Argument Lovelace and Turing Started
Lovelace and Turing never met; she died sixty years before he was born. But their argument has been treated, since 1950, as a direct exchange. Lovelace stating the case for the prosecution, and Turing replying.
The Lovelace Objection
The Lovelace Objection
- The Analytical Engine — and any machine of the same kind — can do only what its programmer has explicitly told it to do.
- To originate something, in the sense relevant to thinking (a genuinely new truth, an unforeseen idea), is to do something other than execute instructions we already understand.
- Therefore, the machine cannot originate anything; it cannot, in any serious sense, think.
The argument’s force is in premise 1. It does not deny that machines are useful. It denies that running rules could ever be the same kind of activity as making the rules. Lovelace had just spent months designing a program, and she knew where the work was. Most of the weight Turing put on the dispute fell on premise 2 — on whether “executing instructions we already understand” really excludes everything we would call thinking.
Turing’s Reply
Turing’s Reply
- The question “can machines think?” cannot be settled directly, because no one can agree on what thinking means; it should be replaced with a test — if a human interrogator, exchanging typed messages, cannot reliably tell a machine from a human, the machine has done whatever thinking amounts to.
- Machines routinely produce outputs their authors did not predict and could not have computed in advance; “doing what we told it” and “doing what we expected” are not the same.
- A machine that learns from experience changes its own behavior in ways its original programmer did not write, which undermines premise 1 of the Objection.
- No one has identified an ability of human minds that we have positive reason to think cannot be present in a sufficiently complex machine.
- Therefore, the Lovelace Objection rests on assumptions we have no good grounds to accept; its conclusion is unproven.
The move that did the most work was premise 2 — surprise. Turing’s example was small and honest. He programmed machines, then guessed wrong about what they would output, all the time. The reason was not metaphysical. It was that the rules he wrote had implications he had not bothered, or been able, to trace. If “originate” means “produce something the programmer did not foresee,” then any non-trivial program does it constantly. If “originate” means something stronger — say, a kind of understanding the program has — then we need to say what that is. Lovelace’s argument has not.
For seventy years this was where the dispute sat. Turing had not refuted Lovelace; he had relocated the argument from can machines do this thing? to what would it take to settle that question? The second is harder.
The stochastic-parrot revival. In 2021, a group of researchers led by Emily Bender published a paper titled “On the Dangers of Stochastic Parrots” (Bender et al. 2021). Its central claim about large language models is recognisably the Lovelace Objection in modern dress:
- An LLM is a statistical pattern-matcher trained on human text.
- It produces plausible continuations of what it has seen.
- It does not understand.
The premise has been updated — “doing what we told it” has become “doing what its training data showed it” — but the structure is the same. The systems are very good at sounding original. That, the argument runs, is not the same as being original.
The surprise reply, updated. The counter has also been updated, and it is recognisably Turing’s. The behaviour of large models routinely surprises the engineers who train them. Capabilities show up that were not designed in, and that were not present in smaller versions of the same architecture. Restricted versions of the Turing test now fall to these systems without much difficulty. Surprise is back on the table, and so is the harder question of what we are willing to accept as evidence of originating.
Where the argument rests now. Two things are true at once. The Turing test was designed to replace the question “can it think.” It is being passed — and we still ask the question. The Lovelace Objection, formulated about a brass-and-steam machine that was never built, sits on top of every press release announcing a new model.
Neither side has gone away. The disagreement was never really about machines. It was about what we mean by thinking. And it was about whether we are willing to apply that word to a thing whose insides we understand. Lovelace, looking at her tables, was not. Turing, taking the strongest reading of his own argument, asked a harder question. Was her unwillingness a discovery about machines, or a habit about people? That is the question two centuries did not close.
Discussion Questions
- Lovelace insisted that designing a procedure was the original work and that the machine’s execution was only the carrying-out. Use an example from outside computing — a recipe being cooked, a piece of music being performed, an athlete following a coach’s plan — to argue for or against her view. Where in your example does the “originating” actually happen?
- Lovelace was schooled in mathematics partly to keep her from her father’s temperament, published under her initials, and was barred from the Royal Society her whole life. Turing was prosecuted for his sexuality and may have died by suicide as a result. Pick one of them. Does the institutional and personal context they worked inside change how you should read their technical argument about machines? What would you lose by treating the work as separable from the life?
- Write The Lovelace Objection and Turing’s Reply in your own words. What is the one thing they really disagree about?
- You are a 2026 conference reviewer. A paper claims its new model “thinks.” What three short questions would you ask the authors before deciding?
- Pick one: ChatGPT, a self-driving car, a chess engine. Does it pass Turing’s test? Does passing the test matter for whether you would call it thinking?
Further Reading
- Ada Lovelace, “Notes by the Translator,” in L. F. Menabrea, Sketch of the Analytical Engine (1843) — the source text, including the Bernoulli procedure and the Objection (Lovelace 1843).
- Alan Turing, “Computing Machinery and Intelligence,” Mind (1950) — the imitation game and the reply to Lovelace by name (Turing 1950).
- Christopher Hollings, Ursula Martin, and Adrian Rice, Ada Lovelace: The Making of a Computer Scientist (Bodleian Library, 2018) — the careful archival account (Hollings et al. 2018).
- Andrew Hodges, Alan Turing: The Enigma (1983) — the standard biography (Hodges 1983).
- Emily M. Bender et al., “On the Dangers of Stochastic Parrots” (FAccT 2021) — the modern revival of the Objection, applied to LLMs (Bender et al. 2021).