Galilean resonances: the role of experiment in Turing’s construction of machine intelligence

ABSTRACT In 1950, Alan Turing proposed his iconic imitation game, calling it a ‘test’, an ‘experiment’, and the ‘the only really satisfactory support’ for his view that machines can think. Following Turing’s rhetoric, the ‘Turing test’ has been widely received as a kind of crucial experiment to determine machine intelligence. In later sources, however, Turing showed a milder attitude towards what he called his ‘imitation tests’. In 1948, Turing referred to the persuasive power of ‘the actual production of machines’ rather than that of a controlled experiment. Observing this, I propose to distinguish the logical structure from the rhetoric of Turing’s argument. I argue that Turing’s proposal of a crucial experiment may have been a concession to meet the standards of his interlocutors more than his own, while his construction of machine intelligence rather reveals a method of successive idealizations and exploratory experiments. I will draw a parallel with Galileo’s construction of idealized fall in a void and the historiographical controversies over the role of experiment in Galilean science. I suggest that Turing, like Galileo, relied on certain kinds of experiment, but also on rhetoric and propaganda to inspire further research that could lead to convincing scientific and technological progress.

experiment, the 'Turing test' has been widely construed in analytic philosophy and cognitive science (Moor 2003;Shieber 2004;Copeland and Proudfoot 2009;Proudfoot 2013), artificial intelligence (Hayes and Ford 1995;Proudfoot 2011), and within Turing scholarship itself (Copeland 2017;Proudfoot 2017), as a kind of crucial experiment to determine the existence of an intelligent machine.Interestingly, Turing offered something of a testable prediction, which comes in his 1950 text immediately after his discursive presentation of his test: I believe that in about fifty years' time it will be possible to programme computers, with a storage capacity of about 10 9 , to make them play the imitation game so well that an average interrogator will not have more than 70 per cent, chance of making the right identification after five minutes of questioning.(Turing 1950: 442) This prediction has often been read to mean that if a machine can play the imitation game well enough in five-minute sessions to fool an average interrogator in at least one-third of the sessions, then it passes the Turing test.But this has led to controversy.
Kevin Warwick and Huma Shah (2015) organized a practical implementation of the Turing test based on the above reading of Turing's 1950 prediction, which led them to announce a chatbot that would have been 'the first to pass the Turing test'.When questioned about the chatbot, which relied on informationretrieval tricks to deceive observers, Warwick and Shah (2015) insisted that their Turing test experiment was designed 'as set out by Alan Turing'.To refute their Turing test-pass claim, Copeland (2017) argued that Turing's 1950 prediction does not specify the rules of the test, but rather refers to the rate of progress Turing expected towards passing his test (272)(273).Copeland interprets Turing's 1950 prediction with the support of another prediction Turing made two years later in a BBC broadcast (1952).When asked how long it would take for a machine 'to stand any chance' in his test 'with no questions barred', he replied 'at least 100 years, I should say' (495).However, to refute their Turing test-pass claim, Jack Copeland (2017) argued that Turing's 1950 prediction does not implement the rules of the test, but rather refers to the rate of progress that Turing considered for his test to be passed (272)(273).Now, this interpretation selects one particular projection and focuses attention on it, while there are several others that Turing made rhetorically, in passing, in the context of a debate.In May 1951, for example, he had given yet another estimate: 'I think it is probable for instance that at the end of the century it will be possible to programme a machine to answer questions in such a way that it will be extremely difficult to guess whether the answers are being given by a man or by the machine ' (1951b, 484).Further, Copeland's interpretation can be seen as sacrificing consistency within Turing's 1950 examination ' (1951a: 484) and multiple times to '[his] test ' (1952), including 'one of my imitation tests ' (503).In this article, following Turing's own usage, I will refer to all of these terms indistinguishably.
paper itself as a single exegetical unit, for it is in tension with how Turing himself refers to his prediction in his very 1950 paper and the support he expects from the outcome: The reader will have anticipated that I have no very convincing arguments of a positive nature to support my views … If I had I should not have taken such pains to point out the fallacies in contrary views.Such evidence as I have I shall now give.The only really satisfactory support that can be given for the view expressed at the beginning of §6 [his stated prediction], will be that provided by waiting for the end of the century and then doing the experiment described.But what can we say in the meantime?What steps should be taken now if the experiment is to be successful?(Turing 1950: 454-455) The prediction is twice referred to as 'the experiment', and it is the 'only really satisfactory support' for his view that machines can think.Turing went on to describe a research strategy for building 'learning machines' that would play the imitation game well, so that the experiment could be 'successful' when performed 'at the end of the century ' (1950: 455-459).Turing writes in that passage as if he were a card-carrying empiricist who believed that controlled experiment was the only source of reliable knowledge about the natural world, as if he believed that experiment had a demonstrative purpose and that his question -Can machines think?should be decided by an experiment. 2  However, a close reading of Turing's 1950 paper, in conjunction with earlier and later primary sources, can reveal important tensions in interpreting his rhetoric of experiment at face value.In several passages, Turing shows a milder attitude towards the confirmatory power of his 'imitation tests', and he refers to the demonstrative power of scientific and technological progress rather than to that of a controlled experiment.In Turing's major report on 'Intelligent Machinery' (1948) the rhetoric of crucial experiments does not appear.It was rather 'the actual production of the machines' that could 'have some effect' in convincing critics and opponents, because 'the idea of "intelligence" is emotional rather than mathematical' (411).Later in (1950), when Turing formulates his 1950 prediction, he carefully distinguishes his imitation test as but 'the more accurate form of the question'.He writes: 'The original question, "Can machines think?",I believe to be too meaningless to deserve discussion'.'Nevertheless', he adds, 'I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted' (442). 3Furthermore, in (1951a;   2   In this case, a one-way crucial experiment, because according to this view while a positive result would be a decisive, confirmatory event for the existence of an intelligent machine, negative results are trivial, since they cannot exclude the possibility of an intelligent machine in the future.This epistemological asymmetry is justified by the tautological fact that 'machines can think' is an open-ended, 'strictly existential statement' (Popper 1959: 47-48).The proposition cannot be refuted, since no run of the imitation game, or any conceivable experiment, will ever be able to prove empirically that machines cannot think.
3 Shlomo Danziger (2022) explores this point in Turing's thought in detail.I interpret his contribution as strengthening a revealing aspect in Turing's philosophy previously identified by Juliet Floyd (2017), namely the importance of common sense.Here I explore Turing's mild attitude towards the power of a controlled experiment, the 1951b) Turing discusses his belief in the possibility of machine intelligence at length, and does not refer to experiments except for a minor mention of his 'viva voce examination ' in (1951a: 484).In (1952) he reformulates his imitation test into a different scenario, where the machine is interrogated by 'a jury, who should not be expert about machines' (495).Referring to this additional variant of the imitation test, Turing drops the language of associating thinking and intelligence with machines, which he had admitted to be 'heretical' only a few months before on the radio (1951a), and suggests instead that machines passing such a test could be considered 'grade A' machines (1951a).Overall, he never provides detailed settings for any of his various imitation tests as controlled experiments.
Observing these tensions, I propose to distinguish the logical structure of the Turing test argument from its rhetoric.I argue that the idea of a crucial experiment appears in Turing's work as a specific move in 1950, probably as a concession to meet the standards of his interlocutors more than his own.I will start by examining Turing's construction of machine intelligence, in which a method of successive idealizations and exploratory experiments can be observed ( §2).
Next, I will propose to read Turing's 1950 text according to a specific logical structure ( §3).The proposed analysis will also relate the main logical step of Turing's 1950 argument to a possible source, namely Bertrand Russell's account of the Socratic dialectic method in his A History of Western Philosophy, first published in 1945.I will then draw a parallel with Galileo's construction of idealized fall in a void and the historiographical controversies surrounding his rhetoric of experiment ( §4).Following Galilean studies, I will argue for a distinction between the method Turing actually used and his comments on methodology, since it can be noted that his method was not a direct result of his comments on methodology and vice-versa.Overall, I argue that Turing, comparably with Galileo, relied on certain kinds of experiment, but also on rhetoric and propaganda, apparently recognizing the need to inspire further research that could lead to convincing scientific and technological progress ( §5).The main points are summarized at the end ( §6).

Turing's construction of machine intelligence and his imitation tests
At the outset, it may be worthwhile to distinguish between Turing's construction of machine intelligence and of his imitation tests.On the one hand, his imitation tests can be understood as illustrating that the human perception of result of which is, of course, to be interpreted by specialized intellectuals.I emphasize that in considering the possibility of a positive answer to his original question, 'Can machines think?',Turing shifts the focus of decision-making to 'general educated opinion'.Furthermore, in considering the possibility of a positive answer to his 'more accurate form of the question', the imitation game, he proposes that the interrogator and judge should not be an expert, thus making the non-expert profile a structural element of the test's design.This can be seen as one of the elements that makes the Turing test a thought experiment, rather than a controlled experiment (Gonçalves 2023).
intelligence is an emotional phenomenon, and as suggesting a criterion of success and a goal for the study of machine intelligence.On the other hand, it can be observed that he developed a theory of mechanical intelligence inspired by human cognitive development and the workings of the human brain, particularly the cerebral cortex of a human child.
I will now examine the role of experimentation in Turing's development of these two elements.I will draw attention to Turing's use of a methodology of successive idealizations and exploratory experiments, most of which he described in his major report 'Intelligent Machinery' (1948).

Exploratory imitation tests on the human perception of machine intelligence
In 1948, Turing still did not have access to a digital computer, something that he had to wait for until 1951. 4He thus conceived the notion of a 'paper' machine: It is possible to produce the effect of a computing machine by writing down a set of rules of procedure and asking a man to carry them out.Such a combination of a man with written instructions will be called a 'Paper Machine'.(Turing 1948: 416) Turing described how he studied the human perception of machine intelligence by means of thought-guided experiments on chess-playing 'paper' machines.Arguing against the idea that machines could not surpass the intelligence of their creators, he reports to have actually experienced the 'sort of situation', where the machine goes beyond the intelligence of its creator, 'arises in a small degree'.'Playing [chess] against such a machine', he wrote, 'gives a definite feeling that one is pitting one's wits against something alive' (412).
The general point that Turing was trying to make, and which he seems to have learned through his exploratory experiments, is that the perception of intelligence has a subjective component.He had just stated that some objections to the possibility of machine intelligence, which he considered 'purely emotional', could not be 'wholly ignored, because the idea of "intelligence" is itself emotional rather than mathematical' (411).He returned to this important point in a final section of his report (1948), 'Intelligence as an emotional concept', writing: The extent to which we regard something as behaving in an intelligent manner is determined as much by our own state of mind and training as by the properties of the object under consideration.If we are able to explain and predict its behaviour or if there seems to be little underlying plan, we have little temptation to imagine intelligence.With the same object therefore it is possible that one man would consider it as intelligent and another would not; the second man would have found out the rules of its behaviour.(Turing 1948: 431) 4 Simon Lavington (2019) refers to Turing's correspondence from Manchester since April 1949 to emphasize Turing's frustration with the delays in the availability of MADM, the Manchester Automatic Digital Machine, which he helped to design (see, for example, 25).
Turing notes that intelligence is perceived by an observer in the behaviour of an agent.The observer's perception is influenced by the agent's actual capabilities as well as the observer's own biases.The less the observer knows about the agent's internal mechanisms, the more (s)he is tempted to 'imagine' intelligence.It is interesting to note that Turing arrived at this insightful observation with the help of an initial experiment with a paper machine, which he describes in the same passage, at the end of his report: It is possible to do a little experiment on these lines, even at the present stage of knowledge.It is not difficult to devise a paper machine which will play a not very bad game of chess.Now get three men as subjects for the experiment A, B, C. A and C are to be rather poor chess players, B is the operator who works the paper machine.(In order that he should be able to work it fairly fast it is advisable that he be both mathematician and chess player.)Two rooms are used with some arrangement for communicating moves, and a game is played between C and either A or the paper machine.C may find it quite difficult to tell which he is playing.(This is a rather idealized form of an experiment I have actually done.).(Turing 1948: 431) This can be seen as a first variant of an imitation test, in this case using chess as opposed to conversation as the intellectual task of the test.Turing even notes that this is 'a rather idealized form of an experiment', as if aware that his style of reasoning may not meet the highest empiricist standards of what qualifies as a controlled physical experiment.
Turing reiterated his point about the perception of intelligence as an emotional phenomenon in 1950, 5 and 1952. 6

Exploratory experiments on child machines and their learning capabilities
Also in 1948, Turing referred to another kind of initial experiment on machine intelligence to explore the possibility of 'organizing' an initially 'unorganized' machine.The reference for an organized machine was his 1936 'universal' machine, which was an idealization of '[a] man provided with paper, pencil, and rubber, and subject to strict discipline ' (1948: 416).Turing used this reference to explain the 'unorganized' machine instead as an idealization of the brain cortex of a human child, which can be moulded: 5 When discussing the fifth objection to machine intelligence, the '[a]rguments from various disabilities', Turing wrote (1950): 'Usually if one maintains that a machine can do one of these things, and describes the kind of method that the machine could use, one will not make much of an impression.It is thought that the method (whatever it may be, for it must be mechanical) is really rather base' (449-450).
6 When asked how a machine could learn by analogy, Turing argued (1952): 'I've certainly left a great deal to the imagination.If I had given a longer explanation I might have made it seem more certain that what I was describing was feasible, but you would probably feel rather uneasy about it all, and you'd probably exclaim impatiently, "Well, yes, I see that a machine could do all that, but I wouldn't call it thinking."As soon as one can see the cause and effect working themselves out in the brain, one regards it as not being thinking, but a sort of unimaginative donkey-work' (500).
We believe then that there are large parts of the brain, chiefly in the cortex, whose function is largely indeterminate.In the infant these parts do not have much effect: the effect they have is uncoordinated.In the adult they have great and purposive effect: the form of this effect depends on the training in childhood.A large remnant of the random behaviour of infancy remains in the adult.All of this suggests that the cortex of the infant is an unorganised machine, which can be organised by suitable interfering training.(Turing 1948: 424) The 'unorganized' machine differs from the 'universal' machine in its flexibility and structural disposition for learning.It is worth pausing for a clear understanding of Turing's machine conceptions.The unorganized machine, as a working idealization of a human child's cortex, should have the potential to balance 'discipline' with 'initiative': If the untrained infant's mind is to become an intelligent one, it must acquire both discipline and initiative.So far we have been considering only discipline.To convert a brain or machine into a universal machine is the extremest form of discipline.Without something of this kind one cannot set up proper communication.But discipline is certainly not enough in itself to produce intelligence.That which is required in addition we call initiative.This statement will have to serve as a definition.Our task is to discover the nature of this residue as it occurs in man, and to try and copy it in machines.(Turing 1948: 429) This passage gives in general lines what Turing explored in his initial experiments on how to develop machine intelligence.The experiments were paper-based and focused on the unorganized machine, while taking the universal machine as a reference for overly rigid but functional behaviour.He expected to be able to partially organize the former under the interference of a teacher giving it reward and punishment feedback.If successful, this would be a proof of concept that a machine could potentially exhibit functional (and in this case flexible) behaviour through learning, without being directly told what to do.In Turing's view, this would make the machine intelligent.Along these lines, he entitled a section ( §10) of his 1948 report 'Experiments in organising.Pleasure-pain systems' (424) and thus opened it: It is interesting to experiment with unorganised machines admitting definite types of interference and trying to organize them, e.g. to modify them into universal machines … I have investigated a particular type of pleasure-pain system, which I will now describe.(Turing 1948: 424-425) Turing explicitly noted that his experiments were exploratory in nature and their point was to allow him to refine his unorganized machine as an idealization of a human child's cortex: The actual technique by which the 'organising' of the P-type [pleasure-pain] machine was carried through is perhaps a little disappointing.It is not sufficiently analogous to the kind of process by which a child would really be taught … I feel that more should be done on these lines.I would like to investigate other types of unorganised machine, and also to try out organising methods that would be more nearly analogous to our 'methods of education.(Turing 1948: 428) Later, in (1950), Turing reiterated the same point about the nature of his paperbased experiments on machine intelligence.Referring to the 'unorganised' machine as a 'child-machine', he wrote that he had 'succeeded in teaching it a few things, but the teaching method was too unorthodox for the experiment to be considered really successful' (457).
The question now arises as to the importance of paper-based, as opposed to computer-based, experiments in Turing's methodology.It is known that he expected to perform experiments on digital computers as soon as they became available (see note 4 above).So we may ask, why did he eagerly await the availability of a digital computer?Was it because he thought that computer-based experiments would be more meaningful or significant?The answer is hinted at in Turing's own reports on his experiments: I made a start on [machine organising methods that would be more nearly analogous to our 'methods of education'.] but found the work altogether too laborious at present.When some electronic machines are in actual operation I hope that they will make this more feasible.(Turing 1948: 428) That Turing expected to use digital computers specifically to scale and accelerate his exploratory experiments, and not to elevate their status to a demonstrative role, is further emphasized in another passage where he describes '[o]ne particular kind of phenomenon' that he 'had been hoping to find in connection with the P-type machines', which was 'the incorporation of old routines into new'.He studied this phenomenon in the context of teaching the machine arithmetic operations (e.g.multiplication as an extension of summation) and language (e.g.irregular verbs as a variation on the rules of the regular verbs).Of his partial results on both problems, he wrote respectively: 'Although I was able to obtain a fairly detailed picture of how this might happen I was not able to do experiments on a sufficient scale for such phenomena to be seen as part of a larger context; ' and '[c]learly this could only be verified by very painstaking work' (429).
Turing's major 1948 report on 'Intelligent Machinery' shows the systematic use of a method of successive idealizations and exploratory experiments, and does not refer to experiment as having any demonstrative role. 7This motivates a closer look at his 1950 paper, which is the main locus of Turing's rhetoric of crucial experimentation.

Turing's 1950 paper, an exegesis
Turing's 'Computing Machinery and Intelligence' has been said to be accessible to a general readership. 8But it has also been said to be a complex, multi-layered text (Genova 1994), too ambiguous for interpretation (Hayes and Ford 1995;McDermott 2014).An intriguing puzzle about Turing's 1950 text is why he considered the original question, whether machines can think, to be 'too meaningless to deserve discussion' and proposed to replace it with a test, an experiment, and yet spent most of his paper (sections § §6, 7, almost 70% of the paper, as detailed below) discussing the question.
I will propose a way of reading Turing's 1950 text that can resolve these tensions ( §3.1), and suggest a possible influence on a key aspect of its structure ( §3.2).

The logical structure of Turing's 1950 paper
Turing's text, which he divided into seven sections, §1 to §7, can be read according to these three main logical steps: . (The proposal, §1 to §3).A new proposal is made about how best to approach the original question, whether machines can think.One possibility is to discuss the then existing common-sense notions of machines (e.g. a steam engine) and thinking (whathumans do).It is noted, however, that this would make the question paradoxical from the outset, and indeed absurd.
The imitation game is introduced as an idealized scenario designed to be a sensible and appropriate substitute for what is seen as obsolete common sense.The appeal and the settings of the idealized scenario are commented on; in particular, why blind conversation makes sense as an intellectual task to empirically evaluate the cognitive abilities of digital computers (the new machines then in existence) to do something that, if done by a human being, should be called 'thinking'.The imitation game is thus presented as a vivid and picturesque image that contains an (epistemological) 'criterion for thinking'.Based on the imitation game, two variants (man-imitateswoman and machine-imitates-woman versions) of the new question are described in place of the original one.Further variations of it will be suggested as the text progresses to its next logical step. .(The science, §4 to §5).The digital computer is explained in language widely accessible to readers in philosophy, mathematics, and science, if not to the general public.It is emphasized that the proposal (the first 8 Max Newman (1955) wrote: 'Since the paper is easily accessible and highly readable, it would be pointless to summarize it.The conversational style allows the natural clarity of Turing's thought to prevail, and the paper is a masterpiece of clear and vivid exposition' (261).Turing's mother, Mrs. Sara, wrote (1959): 'Almost the whole of this article is within the comprehension of the average reader and is highly entertaining' (94).
logical step) is a philosophical reflection on a sciencenamely, Turing's mathematical science of universal computing (1936).This science was combined with the technology of stored programs developed in the computer building projects of the postwar years.This combination was not accidental but fine-tuned in order to make digital computers behave or perform as universal computing machines.At the end of §5, 'Universality of Digital Computers', it should be clear that the proposal does not suggest some arbitrary idealized scenario, but one that is informed and constrained by the science and technology of digital computers.(The integration of common sense and cutting-edge science is a hallmark of thought experiments in the scientific tradition.) .(The discussion, §6 to §7).On the original question of whether machines can think, a negative ( §6) and a positive ( §7) argumentation is presented using the scientifically informed proposal.First, Turing's beliefs and views are explained.The scientific status of the question is considered open.Turing's own conviction is that the answer to the question is positive, but he would rather avoid saying so directly, for the very reason that he has outlined the proposal in the first placenamely, to provide a basis for the discussion not to be meaningless.The discussion itself unfolds by addressing each of a series of nine objections to the possibility of an intelligent machine, systematically referring back to the imitation game.This is the negative argument.Then a research project is presented for the development of 'learning machines' that could be made to play the imitation game well.These, once provided with the necessary storage capacity and a suitable program, would be able to learn by themselves.By analogy with the education of a human child, the proposed approach is to find a 'child program' that would have little structure at first and would grow in complexity with the machine's experience, so that the machine could eventually exhibit its own intelligence in the imitation game.This is the positive argument.
It is worth noting that the whole discussion is developed on the basis of the imitation game.
It is worth noting that the focus of Turing's argument, as indicated by the development of the text, is not on the first step, the proposal, but on the third step, the discussion of the original question whether machines can think.(Turing's argument reaches its climax in section §6 and its full development in section §7.)Without the proposal (first step), however, the discussion would be based not on science but on culture, namely, the commonsense notions of 'machine' and 'thinking' at the time. 9Given the intended conceptual 9 The Oxford English Dictionary's definition of 'machine' in the early 1950s (cf.Mays 1952: 149) implies that machine behaviour was synonymous with unintelligent behaviour, and intelligence was considered an intrinsic property of humans.
change in the meaning of these words based on a new science, this would indeed be absurd.Furthermore, without an explanation of the new science (second step), the proposal could be understood as nothing more than an imaginary exercise in fiction or fantasy.But having established such basic premises, the discussion could finally unfold.It would then have an empirical basis in the proposed (epistemological) 'criterion for "thinking"'.At the same time, this criterion was embodied in a reasonable idealized scenario (the imitation game) in order to maintain an appeal to common sense.This brief analysis may facilitate the understanding that Turing's test proposal was in fact a means, not an end. 10e can now proceed to examine what kind of philosophical argument Turing developed at the climax of his paper, his discussion of the original question, and what is a possible source to have influenced it.

The Socratic dialectical method of philosophical discussion
Turing explicitly referred to Bertrand Russell's A History of Western Philosophy (1945) in his discussion of the 'theological' objection (1950: 443).Russell's book had appeared only five years earlier, and is one of the few works in the bibliography of Turing's 1950 paper, the only work from philosophy.Russell (1872Russell ( -1970) ) wrote a chapter on Socrates, from which I will quote at length, in which he introduced the Socratic method of dialectic: Dialectic, that is to say, the method of seeking knowledge by question and answer, was not invented by Socrates … But there is every reason to suppose that Socrates practised and developed the method … Certainly, if he practised dialectic in the way described in the [Apology], the hostility to him is easily explained: all the humbugs in Athens would combine against him.(Russell 1945: 92) Parallel to Russell's introduction of 'the method of seeking knowledge by question and answer', Turing would justify his choice of the intellectual task to be addressed in the imitation game: 'The question and answer method seems to be suitable for introducing almost any one of the fields of human endeavour that we wish to include ' (1950: 435).And if there was hostility to Socrates in Athens, there was certainly hostility to Turing in postwar Britain, leading him to title one of his BBC radio broadcasts 'Intelligent Machinery, a Heretical Theory ' (1951a).
Russell also noted that the dialectical method was the method used by Galileo in his dialogues to argue his theories and overcome prejudices.Russell went on to reflect on the limitations of the method as paradigmatically demonstrated by Galileo.Note the parallel, as Turing would celebrate Galileo at the end of his rebuttal of the theological objection,11 which is precisely where he cites Russell's History, and later again in his 1951 BBC radio lecture 'Intelligent Machinery, a Heretical Theory'.12Specifically, Russell wrote: The dialectic method is suitable for some questions, and unsuitable for others … Some matters are obviously unsuitable for treatment in this wayempirical science, for example.It is true that Galileo used dialogues to advocate his theories, but that was only in order to overcome prejudicethe positive grounds for his discoveries could not be inserted in a dialogue without great artificiality.Socrates, in Plato's works, always pretends that he is only eliciting knowledge already possessed by the man he is questioning; on this ground, he compares himself to a midwife.(Russell 1945: 92-93) Russell refers to the image of Socrates as a midwife to make the point that philosophical discussion can produce conceptual clarity, but it cannot produce new knowledge about the natural world.He went on to explicitly state that the Socratic method of dialectic does not apply to empirical problems, such as 'the spread of disease by bacteria'.He advised against using philosophical discussion as if it could provide any positive grounds for discovery about the natural world.
The parallel with Turing's argumentative approach as shown in the third logical step of his 1950 text ('the discussion') can be revealing.In the positive part of his discussion ( §7 of his text), Turing would write: 'The reader will have anticipated that I have no very convincing arguments of a positive nature to support my views'.He added: 'If I had I should not have taken such pains to point out the fallacies in contrary views'.Whether under the influence of Russell's History or not, it can be seen that Turing revived the Socratic dialectical approach, while respecting the limits suggested by Russell and allegedly shown by Galileo regarding any positive claims to his empirical question of whether machines can think.
Russell concluded by emphasizing that the proper use of the Socratic method is for questions about the meaning and use of words: The matters that are suitable for treatment by the Socratic method are those as to which we have already enough knowledge to come to a right conclusion, but have failed, through confusion of thought or lack of analysis, to make the best logical use of what we know.A question such as 'what is justice?' is eminently suited for discussion in a Platonic dialogue.We all freely use the words 'just' and 'unjust', and, by examining the ways in which we use them, we can arrive inductively at the definition that will best suit with usage.All that is needed is knowledge of how the words in question are used.But when our inquiry is concluded, we have made only a linguistic discovery, not a discovery in ethics.(Russell 1945: 93) The meaning and common use of the words 'machine' and 'thinking' or 'intelligence' were just the central topic of Turing's 1950 paper.Nonetheless, again in parallel with Russell's point that no positive discovery could come out of an application of the dialectical method, Turing emphasized that he did not expect to have very convincing arguments of a positive nature to support his views.He declared instead that '[t]he only really satisfactory support' that can be given for the belief on his prediction 'will be that provided by waiting for the end of the century and then doing the experiment described' (455).Turing's use of Socratic dialectic 'to point out the fallacies in contrary views' compelled him to acknowledge that he was also limited by it, and thus only a practical experiment could provide satisfactory support for his views.
For an example of what Turing considered to be a logical fallacy of the contrary views, see the fourth objection, the 'Argument from Consciousness ' (1950: 446).Geoffrey Jefferson demanded machine sentience as a condition for machine intelligence, and Turing argued that this was a claim for solipsism that had no basis in common sense.Here Turing fits the Socratic image of a midwife supposedly giving birth to truth.In fact, Russell's point about the Socratic method as capable of producing truth after correcting 'logical errors' might have been compelling to Turing': We can, however, apply the method profitably to a somewhat larger class of cases.Wherever what is being debated is logical rather than factual, discussion is a good method of eliciting ruth.Suppose some one maintains, for example, that democracy is good, but persons holding certain opinions should not be allowed to vote, we may convict him of inconsistency, and prove to him that at least one of his two assertions must be more or less erroneous.Logical errors are, I think, of greater practical importance than many people believe; they enable their perpetrators to hold the comfortable opinion on every subject in turn.Any logically coherent body of doctrine is sure to be in part painful and contrary to current prejudices.The dialectic method … tends to promote logical consistency, and is in this way useful.But it is quite unavailing when the object is to discover new facts.(Russell 1945: 93) In summary, Turing separated his negative use of dialectic to discuss the original question as a logical question ( §6) from his positive proposal of a research project to address it as an empirical question ( §7).He referred to the imitation game as a criterion by which he could expose such logical fallacies, since it placed machines and humans on the same level, side by side.

Archival source
Following the clue of Turing's explicit citation of Russell's History as one of the few works in his bibliography, we have seen the analytical similarity between Turing's philosophical approach in his 1950 paper and the recommended method for philosophical discussion in Russell's book.Moreover, there is an archival finding which can further support my hypothesis that Russell's History may have been a source for Turing's philosophical approach.
A letter to Turing from the close friend of Russell, Rupert Crawshay-Williams (1908-1977), who would publish his Russell Remembered in 1970, describes Russell's reception of Turing's 1950 paper: Dear Turing, I meant ages agoto thank you for sending me the offprint of your Mind article … I am most pleased to have it, as I enjoyed it very much when it first came out.And, you may be amused to hear, so did Bertrand Russell who was here at the time.We read it and discussed it together.We liked not only (of course) the general approach (the assumptions underlying your argument) but also the particular method and the examples.How did you discover about the Encyclopedia Britannica? … I hear you've been made F.R.S.Many congratulations.But as I don't know whether it's official yet I won't stick it in the envelope.

Yours sincerely, Rupert Crawshay-Williams 13
Crawshay-Williams' reference to the possibility of Turing being amused to hear of Russell's appreciation of his paper and the 'of course' note suggests a strong connection with Turing's 'general approach', 'the assumptions underlying his argument'.Their shared appreciation of 'the particular method and the examples' used by Turing (my emphasis) is also significant.Beyond Turing's acquaintance with Russell's philosophy as part of the Cambridge milieu, the letter may also indicate some prior context regarding Turing's paper. 14 It may also be suggestive that the same letter that informs Turing of Russell's reception of his paper is the one that congratulates Turing on being made a Fellow of the Royal Society (FRS) of London for Improving Natural Knowledge.Russell was an FRS, and the Royal Society was an institution known for its support of experiment as the primary source of knowledge about the natural world. 15Turing probably wrote his paper from late December to early January 1950, and in March 1951 he would be elected an FRS with the support of two sponsors, Russell and Max Newman. 16  13 Archives Centre, King's College, Cambridge, AMT/D/5.Rupert Crawshay-Williams to Turing, April 19, 1951.  1There is one event reported by Max Newman's son, William Newman (2006), referring to his experience with Turing, presumably shortly after Turing moved to Manchester in the summer of 1948: 'He came with us on a brief spring holiday in Criccieth, North Wales, where we rented a house.There were lively discussions in our seafront living room between him, my father, Bertrand Russell, Rupert Crawshay-Williams, and others' (186-187).If nothing else, this event, which probably took place in the spring of 1949, may have been the origin of the connection between Turing, Crawshay-Williams and Russell. 15This topic appears, e.g., in Leviathan and the Air Pump (Shapin and Schaffer 1985).Thomas Hobbes is eventually excluded from the Royal Society largely for his anti-experimentalism, which contrasted with Robert Boyle's 'modest, humble, and friendly' experimental philosophy. 16The mentor of Turing's academic career, Newman was also an FRS (Grattan-Guinness 2013).
Far from suggesting that satisfying Russell was a primary goal for Turing, I suggest that Turing cared about being understood and knew how to adapt his language to address specific readers, and he had Russell as an intended interlocutor.Turing was considered an avowed non-scholar in the sense that he avoided relying on the work of others, preferring to figure things out and solve problems on his own (Newman 1955).However, when it came to being read and understood, he certainly showed interest in the reception of his work.For example, while in Princeton, in the United States, Turing carefully instructed his mother to distribute offprints of his 'Computable Numbers' (1936) in England. 17One of the few people on his list, and the one who deserved special instructions, was Russell.Knowing that Russell was, in Turing's words, 'inclined to be ashamed of his peerage', Turing thought 'the situation calls for tact'.He added: 'I suggest that the correct address for an earl be used on envelope, but that you mark the reprint itself 'Bertrand Russell' on the top right hand corner of the cover'.
Galilean resonances can be found both in Turing's method of successive idealizations and exploratory experiments, and in his rhetoric of experimentation.

Galileo's construction of idealized fall in a void
According to the story of the Leaning Tower of Pisa, one of the most famous anecdotes in the history of science, sometime around the year 1590 Galileo would have climbed to the top of the tower and dropped two unequal weights.He did this, so the story goes, to disprove Aristotle's law of fall, which claimed that the speed at which bodies fell was proportional to their weight.By showing that the objects reached the ground simultaneously, Galileo would have demonstrated to the professors and students gathered around the tower that Aristotle was wrong.This legendary story will provide us with a case study in the history and historiography of science whose parallel to the problem of Turing's rhetoric of experimentation is arguably significant.
The story is never mentioned in Galileo's own writings.It was reported 12 years after his death by one of his closest students and collaborators, Vincenzo Viviani (1622-1703), as part of a biography of Galileo written in 1654 and first published posthumously in 1717. 18The Leaning Tower demonstration has often been considered a turning-point in the history of science, and many authors who believe that Galileo's science was mainly empirical have produced it as a classic example of the superiority of empirical science over a priori science.For centuries, Galileo was largely considered 'the first true empiricist' (cf.Segre 1989b: 207), which may explain Russell's account described above.
In summary, despite the beliefs of Stillman Drake (1978), 19 it is unlikely that Galileo performed the legendary tower experiment as described by Viviani (cf.Cooper 1935;Koyré 1937;Segre 1989a).And if he did, he could hardly have obtained the claimed results under the conditions available to him at the time (cf.Naylor 1974;Adler and Coulter 1978;Segre 1980).But there is a caveat.Galileo seems to have done various exploratory experiments.He seems to have used what he learned from such experiments to infer what should be expected in scenarios and conditions whose ideal experimental conditions were beyond his reach.He combined reason with experiment in his theory-building process, and used conceptual experiment in his dialogical arguments.A corollary of this, as suggested by Segre in the Galilean case (1980), is that 'we must notice that not all experiments are meant to be performed' ( 228).(I will suggest in §5 that Segre's point can also apply to the Turing case, despite Turing's rhetoric about performing a crucial experiment.) In the following, I will briefly discuss the findings of Galilean studies in relation to two questions.The first question is: ( §4.1) How did Galileo establish his law of free fall?Did he rely on crucial experiments, as Russell's account of him as an empiricist might suggest?The second shifts the problem from the history of science to its historiography: ( §4.2) Why was a rhetoric of crucial experiments important to Galileo and to his student and first biographer, Vicenzo Viviani?The answers provided by Galilean studies can shed light on the problem of Turing's rhetoric of crucial experimentation.

Galileo and the law of free fall
According to Paolo Palmieri (2005b), Galileo's journey to formulate 'the most beautiful thought experiment in the history of science' in his Two New Sciences (1638) began some five decades earlier in the drafting of his unpublished De Motu (1590).Palmieri reconstructs Galileo's investigations from his early Archimedean arguments about floating bodies and finds him discovering paradoxical phenomena in Aristotle's arguments against the possibility of the void.Galileo would have started from the analogy that the reason why objects of the same kind, though different in volume, fall at the same speed is the same as the reason why both a chip of wood and a large wooden beam float.The reason why the beam behaves the same as that of the chip is that both must lift a quantity of water equal to their volumes as they fall (1909: vol. I, 263-264).Following this line of reasoning, Galileo would eventually come up against Aristotle's claim that if motion occurred in the void, then both heavy and light bodies would move at the same speed, since the resistance of the void to their motions would be zero, which is unnatural (inconveniens) (Galilei 1909: vol. I, 401).Aristotle's reasoning, according to Galileo, is that faster bodies cut through the medium more strongly, but since the void exerts no resistance, all motions must occur at the same speed.However, while Aristotle would have seen this as a proof of the impossibility of motion in the void, Galileo recasts it as a reason to suspend the common sense that heavy and light bodies move at speeds proportional to their weights (Galilei 1590: 34).
Fast-forwarding five decades to The Two New Sciences, Galileo formulates his famous thought experiment through the character of Salviati: SALVIATI: But without experiences, by a short and conclusive demonstration, we can prove clearly that it is not true that a heavier moveable is moved more swiftly than another, less heavy, these being of the same material, and in a word, those of which Aristotle speaks.Tell me, Simplicio, whether you assume that for every heavy falling body there is a speed determined by nature such that this cannot be increased or diminished except by using force or opposing some impediment to it …

[SIMPLICIO acquiesces]
Then if we had two moveables whose natural speeds were unequal, it is evident that were we to connect the slower to the faster, the latter would be partly retarded by the slower, and this would be partly speeded up by the faster …

[SIMPLICIO agrees again]
But if this is so, and if it is also true that a large stone is moved with eight degrees of speed, for example, and a smaller one with four [degrees], then joining both together, their composite will be moved with a speed less than eight degrees.But the two stones joined together make a larger stone than the first one which was moved with eight degrees of speed; therefore this greater stone is moved less swiftly than the lesser one.But this is contrary to your assumption.So you see how, from the supposition that the heavier body is moved more swiftly than the less heavy, I conclude that the heavier move less swiftly.(Galilei 1638: 66-67) Even if the premises are true, Michael Stuart (2020) argues, following Tamar Gendler (1998), Galileo's conclusion need not follow, 'as it is open to the Aristotelian to distinguish between the weights of united and unified entities or to deny that composite objects have determinable weights' (Stuart 2020;Gendler 1998: 973;405, their emphasis).However, although Galileo's use of imagination can be seen as epistemically unjustified, Stuart continues, it led to a change in the natural interpretation of the phenomenon, which allowed for its retrospective justification.An epistemic use of imagination may be considered invalid at one time, but depending on its future consequences, Stuart argues, its status may be reversed at another time (974).Overall, there are uses of imagination 'that are needed to break today's constraints in order to make progress tomorrow'.
In Postils to Rocco (1909: vol. VII, 731), Galileo offers an ex post facto reconstruction of his discovery process, writing 'I formed an axiom such that nobody could ever object to … ' (The axiom corresponds to a slightly more general version of Salviati's first assumption quoted above, which Simplicio accepts.)Galileo continued to outline how he deduced the above reductio ad absurdum, opening the way for the law of free fall and, more importantly, for the possibility of motion in a void.According to Palmieri (2005b: 231), Galileo suggested that it was reason, not experience, that convinced him of the law of free fall for a certain class of phenomena (when bodies of different weights but the same material).Nearly five decades later, then in possession of the law of accelerated fall, and of a complex theory of fluid resistance, Palmieri argues (2005b: 224-225), Galileo would have 'recast his original thought experiment into the punchy presentation that was eventually published in Two New Sciences' (1638).As noted by Naylor (1976), Galileo made quite distinctive uses of (real) experiment and (didactic) demonstration (my emphasis).Now, note that Galileo's reduction suggests an anomaly in Aristotle's theory of motion, but it cannot, in Russell's terms, provide a positive ground for Galileo's own (existential) hypothesis, namely, that there can be motion in a void.Strictly speaking, the empirical study of this hypothesis within a given location requires the removal of the medium from that location.In the specific case of the nearly 200-foot Leaning Tower of Pisa, this would require a chamber the size of the tower and the removal of all air from it.Thus, the empirical evaluation of Galileo's hypothesis at that time would involve begging the question or assuming the conclusion, since creating the conditions that would make the crucial experiment feasible would require pursuing the hypothesis in question through long-term research and development. 20Historically, by the time a crucial experiment was possible, the related Galilean science was so far advanced that there was little point in doing the experiment except to honour Galileo or for entertainment.It was not until the space programs of the second half of the twentieth century, which were themselves based on Galilean science, that the ideal conditions became available to carry out the legendary tower experiment. 21 I do not intend to suggest that experiment did not play an important role in Galilean science, for it did (cf.Naylor 1974Naylor , 1976)), but only that it did not play a demonstrative role (Segre 1980, my emphasis).According to Palmieri (2005a), 20 Interestingly, the same has been claimed about the Turing test: 'The [Turing] tests are circular: they define the qualities they are claiming to be evidence for' (Hayes and Ford 1995: 974).In this connection, here is Turing saying on the radio (1952): 'You might call it a test to see whether the machine thinks, but it would be better to avoid begging the question, and say that the machines that pass are (let's say) 'Grade A' machines (495). 21In 2015, the experiment was performed for a BBC broadcast from NASA's Space Power Facility in Cleveland, Ohio.Available at: <https://www.youtube.com/watch?v=E43-CfukEgs>.Accessed April 9, 2023.
elaborating on William Shea (1972), the methodology of Galileo's intellectual revolution consisted basically in the mathematical study of classes of phenomena under certain idealized conditions.Galileo's investigations relied heavily on thought-guided experiments, some of which also involved what Palmieri calls 'participation, namely, material or bodily activities ' (2018).The intriguing relationship between different experimental settings and scenarios used by Galileo had been observed earlier by Ernst Mach (1897).Mach found that continuous variation between various conditions and scenarios is the fundamental method of both thought and physical experiment.This helps to explain the close relationship between some of Galileo's experiments.For example, Koyré (1953) notes, 'It is well known with what extreme ingenuity, being unable to perform direct measurements, Galileo substitutes for the free fall the motion on an inclined plane on one hand, and that of the pendulum on the other' ( 224).So much for the first question on how Galileo would have established his law of free fall, we can now move on to the second.If Galileo did not rely on crucial experiments, why was a rhetoric of crucial experimentation important to Galileo, or at least to his student and biographer, Vicenzo Viviani?The legendary tower experiment provides us with a particularly interesting case study of the rhetoric of crucial experiments in the history and historiography of science.

The social and cultural dimension of the Galileo's legendary tower experiment
Early on in his De Motu writings (1590), Galileo reports on tower experiments. 22However, as Segre (1989a) notes, in these specific reports Galileo repeatedly and explicitly stated that bodies of different weights fall at different speeds.For example, he reports dropping two different bodies, one of lead and one of wood, from the top of a high tower: 'the lead moves far out in front.This is something I have often tested ' (107).And yet the biographer Vicenzo Viviani, who studied with the late Galileo, would describe Galileo's feats in Pisa according to Galileo's mature views: And then, to the dismay of all the philosophers, very many conclusions of Aristotle were by him [Galileo] proved false through experiments and solid demonstrations and discourses, conclusions which up to then had been held for absolutely clear and indubitable; as, among others, that the velocity of moving bodies of the same material, of unequal weight, moving through the same medium, did not mutually preserve the proportion of their weight as taught by Aristotle, but all moved at the same speed … demonstrating this with repeated experiments from the height of the 22 His mentions of tower experiments appear in Drabkin's translation (1960: 27, 38, 87, 101, 107, 127).The date of Galileo's De Motu is uncertain.Most of it is usually dated to c. 1589-1592, when Galileo was a professor at the University of Pisa, but it is known that some parts were written earlier (Drabkin 1960).
Campanile of Pisa in the presence of the other teachers and philosophers, and the whole assembly of students.(Galilei 1909: vol.XIX, 606) 23   Faced with Viviani's embellishment of Galileo's exploits, Segre contributed an important insight by asking the historiographical question, 'Why did Viviani think it important to report such an experiment?'He looked at Viviani's literary context and found that in Viviani's milieu a biography had to follow certain standards.The paradigmatic example was the Vite, the famous collection of biographies in art history written by the Mannerist painter and architect, Giorgio Vasari (1511-1574).Segre found concrete examples of Vasari's form and style in Viviani's biography of Galileo, including the embellishment of the portrayed artist's image with quasi-true anecdotes.Moreover, Viviani's intended audience, after Galileo himself in his lifetime, included the general educated public (the nobility, the learned clergy, and the academic community).
Segre highlighted an example of the expectations of such an audience in the correspondence between Galileo's followers, Bonaventura Cavalieri (1598-1647) and Evangelista Torricelli (1608-1647).Cavalieri wrote to advise Torricelli on the occasion of the latter's admission to the Accademia della Crusca: 'I hear that they expect physical rather than mathematical things … It is advisable to meet their expectation, and more than that, the universal expectation that has little esteem for mathematics, unless it sees some applications' (cf.Segre 1989a: 447).Segre further noted that Viviani revised the first version of his Galileo biography to emphasize the tower anecdote over the abstract thought experiment that sought to refute Aristotle, probably in an attempt to be more convincing to such an audience.In sum, the anecdote of the leaning tower experiment was an important device for providing the general educated public with a physical, tangible story and application of the otherwise overly abstract Galilean science.The anecdote may have been invented by Viviani and not by Galileo himself (Segre 1989a: 441, 444).However, in Galileo's works themselves, such as the Dialogo and the Discorsi, we can see, as Naylor (1976) notes, Galileo's 'ambition to extend the results of his original experiments to important practical problems' (400).Centuries later, the 'Apollo 15 Hammer-Feather Drop' was a live anecdotal demonstration of Galileo's legendary tower experiment by mission commander David Scott for the television cameras at the end of the last Apollo 15 moonwalk on August 2, 1971.Far from the Earth's atmosphere, essentially in a vacuum, the astronaut simultaneously released a heavy object (an aluminum geological hammer) and a light object (a falcon feather) from approximately the same height (about 1.6 m), which fell to the ground at the same rate to the naked eye.The performer, who attributed their successful mission in part to 'a rather significant discovery about falling objects in gravity fields' made long ago by 'a gentleman named Galileo', celebrated: 'How about that!Mr Galileo was correct in his findings'. 24 Undoubtedly, there is a remarkable historical connection between the Apollo 15 hammer-feather drop and Galileo's public demonstrations and legends, especially the story of the Leaning Tower of Pisa experiment.Robert Crease (2003) pointed out that Galileo's experiments 'slowly transformed from genuine scientific inquiries into public displays', greatly influencing a next generation of scientists.These included Robert Boyle (1627-1692) and Willem's Gravesande (1688-1742), who built air pumps and special chambers to study vertical fall in evacuated environments.Authorities such as King George III, for example, once witnessed a demonstration of a feather and a one-guinea coin falling together inside an evacuated tube.'The popularity of such demonstrations', Crease notes, 'continues to this day' and is included in many handson science exhibits.For him, even if there was no original (leaning tower) experiment, Galileo inspired a whole genre of experiments and demonstrations, and 'we might as well refer to these as the offspring of Galileo's experiment at the Leaning Tower of Pisa'.Boyle would indeed become a master at using instruments to present to the public indisputable facts produced in crucial experiments (Shapin 1984).And he certainly helped promote Galileo's hypothesis that there could be motion in a void.As a teenager, Boyle visited Florence with his French tutor shortly before Galileo's death and was impressed by 'the new paradoxes of the great star-gazer Galileo' (Fulton 1960: 119).If Boyle was recruited and pushed forward a rhetoric of crucial experimentation, the Galilean stories and propaganda seem to have contributed to it. 25 Segre (1980) emphasized the importance of distinguishing between Galileo's method and his methodology.In other words, in practice Galileo's method 'was not a direct outcome of his methodology and vice-versa (249).While Galileo, the scientist, relied largely on a method of successive idealizations and exploratory experiments, and resorted to thought experiments for didactic demonstration, Galileo, the methodologist, promoted a view of himself as an empiricist at strategic moments, and resorted to a rhetoric of decisive experiments and propaganda.I propose to apply a similar distinction to Turing.

Turing's use of thought experiment and propaganda
In light of the historical case of Galileo, we can now revisit Turing's construction of machine intelligence to emphasize his use of thought experiment ( §5.1) and propaganda ( §5.2). 24 A technical description appears in the 'Apollo 15 Preliminary Science Report'.The video is available at <https:// nssdc.gsfc.nasa.gov/planetary/lunar/apollo_15_feather_drop.html>.Accessed April 11, 2023. 25Paul Feyerabend (1970) is known as one of the first scholars to call attention to Galileo's propaganda.Michael Stuart (2021) studied what Feyerabend later called 'the epistemology of drama', inspired by Galileo's tactics, especially his thought experiments.Segre (1988) presents a portrait of Galileo as a politician.

Turing's use of thought experiment
By 1948, as we have seen, Turing had already developed his concepts of machine intelligence and imitation testing.In the absence of actual digital computing equipment to scale his initial experiments, he explored machine intelligence by simulating on paper the learning process of successive models of his 'unorganized' machine.
Turing also explored human perception of machine intelligence through the abstraction of a chess-playing paper machine.'Playing [chess] against such a machine', Turing wrote, 'gives a definite feeling that one is pitting one's wits against something alive' (412, my emphasis).Writing in 1948, it seems that Turing himself needed no further evidence to believe in the possibility that average interrogators in the future could be fooled by a sophisticated machine in a third of the experimental sessions after five minutes of questioning.In 1950, he would outline his 'beliefs' and conclude the point stating that '[c]onjectures are of great importance since they suggest useful lines of research' (442).If Turing was himself already convinced, then there may be no reason to understand his 1950 crucial experiment as something other than a rhetorical strategy.Arguably, it was more a concession to meet the standards of his interlocutors than his own, which resonates with Galileo's moves from his real, relatively messy exploratory experiments to his sanitized, didactic demonstration. 26oving from chess to conversation, Turing would make the step from his exploratory experiments in 1948 to his thought-led imitation tests in 1950.Gonçalves (2022Gonçalves ( , 2023) ) historicized how this came about in the context of a controversy about minds and machines, and this is consistent with the parallel suggested here with Galileo's cosmological polemic.The idealizations that Turing introduced in his 1950 imitation tests can be regarded, as Karl Popper proposed in his Logic of Scientific Discovery (1959), as 'concessions to the opponent, or at least acceptable to the opponent' (466).For example, the change from chess to conversation was probably largely a response to criticism he had received from Polanyi and Jefferson, both Fellows of the Royal Society, sacrificing his caveat that 'many people' thought that 'a very abstract activity, like the playing of chess, would be best ' (1950: 460). 27He also introduced a third player, representing the imitated type, to help the human judge to identify the imitating machine.He seems to have done this, somewhat satirically, to address another point raised by Jefferson in Jefferson's criticism of Grey Walter's mechanical tortoises.Overall, it can be seen that Turing used his imitation tests both to criticize opposing views and to heuristically present his own views.He varied his imitation tests by varying their experimental conditions and settings, which, as Mach describes (1897), extends 'the scope of ideas (expectations) tied to them'.
When it comes to experimentation, Turing's method was to design and conduct initial experiments to explore the phenomena under study and to refine his hypotheses and analogical models.Thus experiment played an important role in Turing's actual construction of machine intelligence and imitation tests, but apparently not a demonstrative role.Turing's overarching method for establishing rigorous results was the axiomatic method, not the experimental method.In his obituary of Turing, Max Newman emphasizes Turing's reliance on mathematical proof: The central problem with which he started, and to which he constantly returned, is the extent and the limitations of mechanistic explanations of nature.His way of tackling the problem was not by philosophical discussion of general principles, but by mathematical proof of certain limited results … .(Newman 1955: 256) Another testimony was given by R. K. (Richard) Livesley, a graduate of the Mathematics Department at Manchester who met Turing regularly in the Computing Machine Laboratory from 1951 to 1954 (Lavington 2019: 38-39).According to Livesley, 'Turing was fond of saying "An ounce of mathematics is worth a ton of computing". 28Compare Galileo in note 26 above, which refers to his mathematical proof of the properties of projectile launching.
Further examining Turing's writing in 1948, it was 'the actual production of the machines'not a crucial experimentthat could outweigh the reactions to the possibility of machine intelligence, some of which he expected to be 'purely emotional': The objections (a) and (b) [which in 1950 he called the 'Heads in the Sand' and the 'Theological' objections, respectively], being purely emotional, do not really need to be refuted.If one feels it necessary to refute them there is little to be said that could hope to prevail, though the actual production of the machines would probably have some effect.(Turing 1948: 411) This can be read as suggesting that, for Turing, the best demonstration of machine intelligence would come with its scientific and technological development. 29Turing's postwar career shows that he was committed to the project of building an intelligent machine. 30egre (1997) notes that Galileo 'was, no doubt, a master of rhetoric', but his arguments were 'not meant only to persuade' (498).Here, another parallel may be helpful, for Turing's belief in 'the actual production of machines' for persuasive purposes is reminiscent of figures such as the French inventor and engineer Jacques de Vaucanson (1709-1782), famous for his artificial duck.The designers of early modern automata in the seventeenth and the eighteenth centuries were criticized for being driven by futile, purely entertaining motives.Historians of technology have disputed this.For example, David Fryer and John Marshall (1979) criticized the 'claim that the primary objective of Vaucanson's work was 'to astonish and amuse the public' (267-268).'Vaucanson', they argued, 'was an entertainer, but he was also deeply committed to the development of an explanatory psychology'.This analysis of Vaucanson's motives belongs to a class of studies presented earlier by Silvio Bedini (1964) and Derek de Solla Price (1964), which showed that early modern automata were neither 'trivial toys' nor 'immediately useful inventions'.Rather, they were simulacra or models 'whose very existence offered tangible proof, more impressive than any theory, that the natural universe of physics and biology was susceptible to mechanistic explication' (Price 1964: 9). 31Turing's project to demonstrate an intelligent machine can be understood in this tradition.
But how could Turing best contribute to such a project at a time when, as he himself explained, machines lacked sufficient storage capacity and computing power, and the science of machine learning that he proposed was still in its infancy?There is a secondary source known to Turing scholars which suggests an answer to this question.

Turing's use of propaganda
Robin Gandy (1919Gandy ( -1995) ) was one of Turing's closest friends in his postwar years (Hodges 1983).Regarding Turing's purpose in writing his 1950 paper, Gandy wrote: It was intended not so much as a penetrating contribution to philosophy but as propaganda.Turing thought the time had come for philosophers and mathematicians and scientists to take seriously the fact that computers were not merely calculating engines but were capable of behaviour which must be accounted as intelligent; he sought to persuade people that this was so.He wrote this paper unlike his mathematical papers quickly and with enjoyment.I can remember him reading aloud to me some of the passages always with a smile, sometimes with a giggle.(Gandy 1996: 125) Gandy's anecdote supports the approach taken in this article to distinguish the logical structure of Turing's 1950 paper from its rhetoric. 32It also supports an understanding of Turing's test more as a philosophical argument and less as a crucial experiment to decide about the possibility of machine intelligence.Turing dressed his thought experiment in the imitation game, inspired by popular material culture (Gonçalves 2023: 18), and in the few years that followed he went on public radio three times to argue for his hypothesis that machines can think (Turing 1951a(Turing , 1951b;;Turing et al. 1952).So the record suggests that Turing was interested in public engagement and propaganda, as was the case with Galileo's biographer Viviani, who largely followed Galileo himself (Segre 1988).
Turing's test proposal seems to have influenced a next generation of scientists, notably John McCarthy (1927McCarthy ( -2011McCarthy ( ), who wrote (1956)): 'The problem of giving a precise definition to the concept of "thinking" and of deciding whether or not a given machine is capable of thinking has aroused a great deal of heated discussion'.'One interesting definition', he continued, 'has been proposed by A. M. Turing'.McCarthy also attributed to Turing '[t]he first scientific discussion of human level machine intelligence' (2007), referring to Turing's lecture to the London Mathematical Society in (1947).'The notion was amplified as a goal in [Turing's 1950 paper]', McCarthy added.Thus, McCarthy referred to the Turing test both as a definition of the concept of thinking (early in his career) and as a research goal for human-level machine intelligence (later in his career).
Marvin Minsky (1927Minsky ( -2016)), a pioneer with McCarthy in the field of artificial intelligence, commented to the media on the results of the practical implementation of the Turing test announced in 2014 ( §1): 'Nothing is learned from poorly designed "experiments".Ask the program if you can push a car with a string.And, if not, then, why not?'33A year earlier, Minsky had said in another interview: The Turing test is a joke, sort of, about saying "a machine would be intelligent if it does things that an observer would say must be being done by a human"; so it was suggested by Alan Turing as one way to evaluate a machine but he had never intended it as the way to decide whether a machine was really intelligent … . 34t is worth noting that McCarthy and Minsky, leading scientists of the next generation after Turing, had no problem in distinguishing between the substance of Turing's 1950 paper and its rhetoric of a crucial experiment.
Less than two decades after Turing's 1950 paper, his imitation test was on its way to becoming an icon of popular culture.It was featured in Arthur Clarke's influential novel, 2001: A Space Odyssey: The sixth member of the crew cared for none of these things, for it was not human.It was the highly advanced HAL 9000 computer, the brain and nervous system of the ship … Whether HAL [Heuristically programmed ALgorithmic computer] could actually think was a question which had been settled by the British mathematician Alan Turing back in the 1940s.Turing had pointed out that, if one could carry out a prolonged conversation with a machinewhether by typewriter or microphone was immaterialwithout being able to distinguish between its replies and those that a man might give, then the machine was thinking, by any sensible definition of the word.HAL could pass the Turing test with ease.(Clarke 1968: 97) Hodges (1983: 533) asked whether Clarke might have chosen the year 2001 to fulfil Turing's prediction quoted here at the beginning ( §1).
Here I have suggested a parallel between Turing's and Galileo's methods, but in fiction and popular culture the products of their science and propaganda meet.

Concluding remarks
Turing articulated a rhetoric of a crucial experiment to determine the existence of an intelligent machine ( §1).And clearly, experiment played an important role in Turing's construction of machine intelligence ( §2).However, it can be seen that his rhetoric of crucial experimentation appears as a specific move in his 1950 paper, which is followed in later primary sources by a milder attitude towards the confirmatory power of what he called in 1952 his 'imitation tests'.Earlier, in his major 1948 report on 'Intelligent Machinery', Turing refers to the 'the actual production of machines,' somewhat alluding to the persuasive power of scientific and technological progress itself rather than that of a controlled experiment.
These tensions in Turing's primary sources motivate a distinction between the logical structure of the Turing test argument and Turing's rhetoric of experimentation.An exegesis of Turing's seminal 1950 paper has been proposed ( §3), identifying in it Turing's philosophical discussion of the question 'can machines think?' as its main logical step.I have argued, both analytically and historically, that Turing's discussion may have been influenced by Bertrand Russell's account of the Socratic dialectical method in A History of Western Philosophy.Further, I have argued that some influence of Russell, and possibly of other interlocutors as well, including the Royal Society itself as an institution supporting controlled experimentation as the primary source of knowledge about the natural world, can partly explain Turing's rhetoric of crucial experimentation.The latter may have been a concession to meet the standards of others more than his own.
The frame of analysis has been broadened by looking at the maturity of Galilean studies.I have drawn a parallel with Galileo's construction of idealized fall in a void and have briefly reviewed the historiographical controversies about the role of experiment in Galilean science ( §4).We have seen that Galileo could hardly have obtained the results claimed by the story of the Leaning Tower experiment.To perform such an experiment under the ideal conditions necessary to validate Galileo's existential hypothesisnamely, that there can be motion in a voidwas beyond the possibilities at that time.The empirical evaluation of Galileo's hypothesis implied begging the question or assuming its conclusion, since making the crucial experiment feasible required pursuing the hypothesis itself through long-term research and development.Both in the secondary literature and in Turing's own statements, there are indications of the perception that Turing's imitation tests are circular.As has been argued in the Galileo literature, not all experiments are meant to be performed.This seems to be true for the Turing test.
Finally, I have emphasized in Turing's construction of machine intelligence his use of thought experiment and propaganda ( §5).This resonates with important findings in Galileo scholarship, particularly the proposed distinction between the method Galileo used and his alleged methodology.Turing, comparably with Galileo, I suggest, recognized the need to inspire further research that could lead to convincing scientific and technological progress.Galileo's construction of idealized fall in a void, which is an important case in the history and historiography of science, shows that such tactics have been used with great success in the past to change the existing natural interpretation of phenomena.We may be on the verge of seeing history partially repeat itself with Turing's construction of machine intelligence.