In this step, we look closer at the groundbreaking work of Alan Turing and the place of learning by imitation within it.
Automatic Computing Engine (1947)
Turing in 1947 held a lecture entitled ‘Automatic Computing Engine’ presented to the London Mathematical Society.
In the lecture, he stressed the importance of developing a machine that could learn from experience and is able to interact with human beings so that ‘it may adapt itself to their standards’. He also maintained that the machine should be treated like a ‘slave’ so that the human user could be aware of what was happening at all times. Furthermore, as the machine became more efficient in conducting activities, the instructions given to it would become ‘altered beyond recognition’. This would be in the same way that ‘a pupil improves their own work after being tutored’.
Such a machine could be credited as having ‘intelligence’, but for the sake of fairness, it was important not to have too high expectations of its performance against humans (Shah 2011). Chess-playing was the platform on which Turing first envisaged a machine imitating the actions of a human chess player.
Intelligent Machinery (1948)
In 1948, Turing published an essay entitled Intelligent Machinery, which is considered to be ‘the first manifesto of Artificial Intelligence’ (Copeland 2000) and focused on the extent to which was possible for a machine to exhibit ‘intelligent behaviour’. In this essay, Turing introduced a comparison or ‘indistinguishability’ test to see if a machine could play chess as well as a mediocre human player. Importantly, he included imitation as part of his test. Shah (2011b) describes the test as follows:
In the 1948 version of the imitation game, a mathematician and chess player acted as ‘B’ operating a ‘paper machine’, while two ‘poor’ chess players, A and C, unseen to each other, played across two rooms. Turing felt it might not be easy for C to say whether they were playing A or the paper machine: ‘C may find it difficult to tell which he is playing’
(Shah 2011: 431).
Importantly in his 1948 essay, Turing posited his controversial notion, that:
….the idea of ‘intelligence’ is itself emotional rather than mathematical
In other words, that intelligence is not adequately measured in intelligence tests, yet Turing advocated a comparison test between a human and a machine.
The evolution of Turing’s ideas on machine intelligence from playing chess to a linguistic challenge is founded in his belief that: the most impressive human activity was the learning of languages.
Computing Machinery and Intelligence (1950)
In 1950, Turing published the highly influential paper Computing Machinery and Intelligence, which has had almost 12,000 citations in Google Scholar (October 2019). In the paper, he developed a conversational test to see if a machine could answer any question in a satisfactory manner in two formats, the simultaneous comparison test and the viva voce test. The two formats are outlined in the table below:
|Simultaneous comparison test||Viva voce test|
|A three-participant game, or simultaneous comparison test involving one human interrogator questioning, by text only, two hidden entities, one human and one machine. In this game, the role of the ‘hidden human’ is to be truthful. The interrogator’s task is to determine which is the human and which is the machine from the answers both give to the interrogator’s question.||A two-participant game or viva voce involves a human interrogator questioning one hidden entity by text interaction only and has to determine whether the answers received are from a human or a machine.|
(Adapted from Turing 1950)
We will now look at each test in more detail.
The simultaneous comparison test
The simultaneous comparison test is visualised in the diagram below. The image shows a computer screen that is split left and right. Both sides are displaying different text-based data. The ‘judge’ computer needs to correctly identify which data represents human conversation.
The winning machine in 2008, Elbot, is the research and development chatbot from conversational AI company, Artificial Solutions, with whom you can chat in the next step. The results from the experiment can be seen in the link below:
The viva voce test
In the viva voce two-participant format of Turing’s imitation game, one human being communicates with one machine and decides whether or not they are interacting with a human or a machine.
The 2012 Turing test experiment was won by a chatbot imitating the talking ability of a 13-year-old male child from Odessa, Ukraine who can speak English. The Eugene Goostman chatbot was developed by a team led by AI scientist Dr Vladimir Veselov and a report on the 2012 experiment can be found on the Turing100 blog here:
Both the simultaneous comparison and the viva voce tests were carried out in one experiment at Bletchley Park, UK, the centre for Britain’s codebreaking in the second world war on the 100th anniversary of Alan Turing’s birth: 23 June 2012.
Discuss Alan Turing’s thought experiment known as Turing’s Imitation Game or the Turing test. What do you consider to be the strengths and weaknesses of this as an interpretation of artificial intelligence?
Further resources Arai, N. (2017). Could an AI Pass the Entrance Exam for the University of Tokyo? TED https://www.ted.com/talks/noriko_arai_can_a_robot_pass_a_university_entrance_exam?utm_campaign=tedspread&utm_medium=referral&utm_source=tedcomshare
ReferencesArtificial Solutions. (2019). Conversational AI Platform. Artificial Solutions https://www.artificial-solutions.com/Copeland, B.J. (2000). What is Artificial Intelligence. AlanTuring.net Reference Articles on Turing http://www.alanturing.net/turing_archive/pages/Reference%20Articles/what_is_AI/What%20is%20AI03.htmlEncyclopaedia Britannica. (2005). Imitation Behaviour https://www.britannica.com/topic/imitation-behaviourHeyes, C.M. and Ray, E.D. (2000). What is the Significance of Imitation in Animals? Advances in the Study of Behavior, 29, 215-245McElroy, M. (2013). A First Step in Learning By Imitation, Baby Brains Respond to Another’s Actions. UW News https://www.washington.edu/news/2013/10/30/a-first-step-in-learning-by-imitation-baby-brains-respond-to-anothers-actions/Shah, H. (2011a.). Deception-detection and Machine Intelligence in Practical Turing Tests [Thesis]. Reading UniversityShah, H. (2011b). Turing’s Misunderstood Imitation Game and IBM’s Watson Success. The 2nd Towards a Comprehensive Intelligence Test (TCIT). Symposium on Reconsidering the Turing test for the 21st Century AISB 2011 Convention, York University, AprilShah, H. (2012). Vladimir Veselov Wins Colonnade Trophy. Turing100 https://turing100.blogspot.com/2012/06/vladimir-veselov-wins-colonnade-trophy.htmlShah, H. (2018). Turing’s Substantive vs. Simulated Intelligence: Can a Machine Answer Any Question? Interdisciplinary Conference on Rethinking, Reworking and Revolutionising The Turing test, Edinburgh University, November 15-16Turing2014. (2014). Eugene Goostman Machine Convinced 33.33% of a New Set of Judges at The Royal Society 6-7 June 2014, following its 29.17% performance at Bletchley Park in 2012. Turing2014. 60th Anniversary of Alan Turing’s Untimely Death http://turingtestsin2014.blogspot.com/2014/06/eugene-goostman-machine-convinced-3333.html University of Reading. (2008). Can a Machine Think? Results from the 18th Loebner Prize Contest https://www.reading.ac.uk/15/research/ResearchReviewonline/featuresnews/res-featureloebner.aspx
© Coventry University. CC BY-NC 4.0