The Thought Experiment in the Foundations of Artificial Intelligence - Research Paper Example

? Thought experiments in Artificial Intelligence Turing’s Imitation Game Introduction The Turing’s imitation game, also known as the Turing’s test is a well-known experiment designed to determine a machine’s ability to display intelligence equivalent to that of a typical human being. It was introduced by Alan Turing (1950) to deal with the question whether a machine can have the ability to think. Since the term “thinking” was too abstract, Turing chose to use a different term; whether a machine can have the ability to “imitate” a typical human’s response to a series of simple questions. However, this is limited to a text only channel such as a keyboard and display screen to display the results. The test involves an interrogator, a machine and a person. The interrogator should be in a room separated from both the machine and the other person. All participants are in fact completely separated from each other, whereby both the machine and the person are designated labels X and Y. This means that he does not know which one is the machine and which one is the person at the beginning of the game. The interrogator’s objective is to distinguish between the two using a series of questions on the machine and the other person as well. The questions could be of the form “Will X please tell me whether X plays chess?” (Oppy Graham, Dowe David, 2011, The Turing Test, para. 7). Both must answer the question. The machine’s objective is to try and convince the interrogator that it’s indeed the other person, while the person tries to help him identify correctly which one is the machine. If the interrogator fails to reliably tell the machine from the human, then the machine is said to have passed the test. The test does not check the ability to give the correct answer; however it checks how closely the answer resembles that of a typical human answer. Turing believed that in the next 50 years, computers with the capacity of 109 would have been developed and they would be able to play the imitation game so well that an average interrogator will not have more than 70% chance of making the right identification after five minutes of questioning (Oppy & Dowe, 2011, The Turing Test, para. 8). Relevance of the Turing Test in AI The Turing test, though being an incredible method for testing intelligence in machines, may be way ahead its time. The technology available now is far from achieving a credible human-like conversation for five minutes as Turing believed it would. The dynamics of human conversation are far too complex as factors like arrangement of words come into play. This can be well observed in the Loebner Prize Competition; an annual event in which computer programs are submitted to the Turing Test. Such competitions have catalysed the growth of AI technology substantially over the years, whereby programs like ELIZA came up. In 1997, one exceptional program called CONVERSE, developed by David Levi and his team, including a well-known researcher in computational linguistics, Yorick Wilks, won the Loebner Prize competition (William J. Rapaport, 2005, The Turing Test). Turing’s experiments focuses mainly on an AI machine’s ability to understand natural language. However this test has undergone some objections over time whereby some people suggest that it is chauvinistic i.e. it only recognizes intelligence in things that have the capacity to sustain a conversation with humans. Others thought that the Turing Test is not sufficiently demanding. Turing (1950) however considered possible objections to his claim that machines can “think”. He went ahead and labelled them such as; The Theological objection, the “Heads in the sand” objection, The Argument from Consciousness, Arguments from Various Disabilities, Lady Lovelace's Objection, Argument from Continuity of the Nervous System, The Argument from Informality of Behaviour and finally, The Argument from Extra-Sensory Perception (Oppy & Dowe, 2011, Turing (1950) and Responses to Objections, para.1). A rather simplified variation of the Turing test is having an interrogator and one participant (William J. Rapaport, 2005, The Turing Test and its variations, para. 4). This can either be a human or a computer. Passing the test again requires the computer to convince the interrogator that he or she is interacting with a human. This method is employed in most modern versions of the test. Turing considers this version an oral exam in which the subject passes by convincing the examiner that he understands the content being tested. A number of variations to the Turing’s experiment have come up with time but what if we could consider a machine that can “turn the tables”?, i.e. a machine with the ability to become the “interrogator” rather than the “subject”. This means that the computer is able to ask the questions then the human responds to them, thus achieving a rather credible interactive human conversation. We consider a computer program that asks questions, while at the same time appears to respond in a way that a human would. The main objective here is for the computer to take control of the conversation, thus it computes an appropriate response to the answers provided, this time by the human. In order to achieve this, the program has to be able to formulate questions whose answers may be categorised according to two major characteristics; Positive answer, or a Negative answer. This is further elaborated below as I try to formulate a simple thought experiment. Formulation of a thought experiment. In formulating a simple thought experiment that may help in the intelligence of AI machines, one can consider the machines ability to sway the interrogator into providing a feedback to the machine, rather than the machine giving feedback to the interrogator. This can be achieved by ensuring the machine asks a question after every response. The question here assumes that all the possible answers provided (this time by the interrogator); either falls under positive feedback or under negative feedback. Here is an example of a simple conversation between two people who just met. For the sake of simplicity I took two arbitrary names: Tom and Jane; Tom: Hi, how are you Jane? Jane: I am fine. How have you been? Tom: I am a little disappointed today. Spent the whole afternoon waiting for a lecturer, who, eventually did not show up. Jane: Am sorry to hear that. Did you try to solve it? Tom: No. I got tired and left. Jane: Do not worry, it happens. ……….. Notice how in the conversation, Jane, being the one who was initially asked a question, turned into the one interrogating Tom. In AI this can be achieved by a machine with the capability to turn into the interrogator, thus taking control of the conversation. For instance, the way Jane answered the first question “How are you Jane?”, then she went ahead and asked Tom a question. Notice that for every response Jane makes, she goes ahead and asks another question. Assuming Jane was the program and Tom the interrogator in a Turing experiment, the machine formulates a question related to the answer it gave. Now the question is formulated in a way that the only possible answer that Tom would provide is either a positive one, or a negative one. The word “disappointment” indicates negativity hence the computer is programmed to respond with a statement of either encouragement or sympathy, such as “I am sorry to hear that” or “that must be awful. You will be fine”. Say if Tom had responded positively, for example; “I had a great day”. The machine also provides positive feedback, using statements like “that’s good to hear” followed by a question directed to Tom, that would make Tom respond either positively or negatively. An example of such a question would be “Did you accomplish all your necessary tasks for the day?” After which Tom would have to answer with a “Yes” or “No”; positively or negatively, and the conversation goes on. This hypothesis displays simple human interaction in which one engages in a conversation with the other and asks them some questions, making it seem as if the machine is another human being that is actually quite interested in the conversation. However, due to the complexity of human language, some assumptions are made. Assumptions made One assumption is that the human only gives answers that are either positive or negative. This may not be the case in a normal human conversation as some neutral statements that are neither positive nor negative exist. Say asked the question “how was your day?” then provided the answer “Just had a normal day”. It may be tricky to determine whether the person is pleased about the way he or she spent the day, or if he or she is disappointed. References Oppy, G and Dowe, D. The Turing Test. Retrieved from The Stanford Encyclopaedia of Philosophy (Spring 2011 Edition), Edward N. Zalta (ed.). William J. Rapaport. (2005). The Turing Test. Buffalo Buffalo: State University of New York, NY 14260-2000. Read More