Author Archives: Kevin Hess

Alan Turing’s “Computing Machinery and Intelligence”

06 Tuesday Dec 2011

Posted by in Reading Summary

≈ 3 Comments

http://www.loebner.net/Prizef/TuringArticle.html

Turing begins his paper with a description of a set of rules for a test he calls the “imitation game,” (what would later come to be known as the Turing Test) as a means of answering the question, “can machines think?” In the game, there is an interrogator, a man, and a woman. The object of the game for the interrogator is to determine who the man is and who the woman is only by asking each of them questions. The questions are administered in such a way that the interrogator gleans no additional information from them besides the answers themselves – for example, through the passing of typed notes. The argument suggested by Turing in the paper is that if the man or the woman were replaced by a machine and the interrogator finds it equally difficult to distinguish between human and computer, then it can be said that the machine in question “can think.”

In the next section of his paper, Turing discusses possible criticisms of the new way in which he has framed the question. He argues that his proposed method factors out the physical appearance of a machine in our perception of whether or not it can think, and that “[t]he question and answer method seems to be suitable for introducing almost any one of the fields of human endeavour that we wish to include.” He argues that even though it seems as if the game heavily favors the human (it’s very difficult for a human to trick someone into thinking they are a computer, too), this doesn’t matter as long as one can accept that it is possible for a machine to be built that can take this test.

In the next few sections of the paper, Turing further clarifies the definition of a “machine” in his description of the game to mean “digital computer” and then goes on to describe various qualities of digital computers. He discusses the elements of a digital computer (“store,” “executive unit,” and “control”) and describes their finite state nature. A reference is made to Babbage’s Analytical Engine as an example of a machine that is a digital computer despite not being an electronic one. A section is also spent arguing that because of the finite state nature of a digital computer, it is possible for a digital computer to simulate any discrete-state machine. This implies that if any one machine can be constructed to play the imitation game, it answers the broader question of “can machines think.”

These sections specifying the machine described in the original outline of the imitation game are followed by a list of possible arguments in opposition to the claim that it is possible to construct a machine that can think. These arguments and my brief interpretations of Turing’s responses to them are as follows:

  1. “Thinking is a function of man’s immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or to machines. Hence no animal or machine can think.”

Turing’s response: If God is truly an omnipotent being, then should it not be within his power to assign a soul to an animal, or, similarly, to a machine, and thus also give the power to think?

2. “The consequences of machines thinking would be too dreadful. Let us hope and believe that they cannot do so.”

Turing’s response: This argument is so trivial that it needn’t even be considered.

3. Mathematics has shown that there are problems which cannot be solved mathematically. Doesn’t this mean that there are problems which can’t be solved by digital computers as discrete state machines, which could be solved by humans?

Turing’s response: Although this is a strong argument, do we assign too much importance to our ability to answer questions that a machine theoretically cannot? Those that make this argument would be okay with discussion the question through the criteria of the imitation game anyway.

4. Unless a being can express emotions and be conscious of these emotions, it cannot be said that this being can think.

Turing’s response: It’s possible to test this quality using the imitation game – saying a machine has been programmed such that it writes a poem. It’s possible to for an interrogator to ask questions about the poem to assess whether the machine was conscious of its decisions in the writing of the poem.

5. There will be things that a machine cannot do (“…be kind, resourceful, beautiful, friendly…”).

Turing’s response: This is simply an issue of storage capacity – given infinite storage capacity, machines can have a large diversity of behaviors.

6. Machines can only do what they are programmed to do – they cannot exhibit some behavior that was not already defined in the programming.

Turing’s response: What this argument is really suggesting is that machines cannot surprise. Can it really be said that humans are capable of new thought, if all their ideas are based on things that they have learned?

7. The nervous system is not a discrete state machine – it is continuous. How can a computer simulate human thought as a discrete state machine then?

Turing’s response: A digital machine can simulate a continuous machine so closely that the difference will not be clear to the interrogator in the imitation game.

8. There is no set of rules which can describe what a person should do for every possible scenario.

Turing’s response: Although it may be difficult to comprehend, we cannot say for sure that there is not one set of rules which can be used to predict all of our behavior. Even in a simple case where a computer is given a number and then returns another with no indication of what it has done, it is difficult or impossible to guess the rule that will predict every possible output – that does not mean that one does not exist.

9. A machine cannot exhibit extra-sensory perception.

Turing’s response: In this case, a special imitation game will have to be set up with a “telepathy-proof room” to be sure that the machine is not being influenced by psycho-kinetic powers.

In the final section of the paper, Turing again addresses oppositional argument number 6 – Ada Lovelace’s argument that a computer can only do what it has been programmed to do. Turing discusses a process which he believes could overcome the difficulty in programming a machine that could successfully pass the imitation game as well as disprove the Ada Lovelace argument. In this process, rather than trying to program a fully functional machine from the start, it might be better to create a “learning machine” – a machine that begins with only a base set of rules and then continually updates these rules of interaction as it learns.

Class Summary: 10/31

01 Tuesday Nov 2011

Posted by in Class Summary

≈ Comments Off on Class Summary: 10/31

Class began with a discussion of some of the factual inaccuracies found in Jacquard’s Web. Although the book is very readable, some technical accuracy was sacrificed for the sake of the narrative of the book. One example of a factual inaccuracy found on the book was that it suggested that the ENIAC was programmed using punched cards, when in reality the machine was programmed using patch cables.

Discussion turned to the reading “The Past and Future History of the Internet,” by Barry M. Leiner et al., and to the early formation of the internet. One of the earliest forms of the internet began with DARPA (Defense Advanced Research Projects Agency) and the creation of ARPANET (Advanced Research Projects Agency Network). The first successful communication over ARPANET was sent on October 29, 1969, between UCLA and Stanford.

Log showing first communication over ARPANET

The question of who, exactly, invented the internet has an ambiguous answer. Because the creation of the internet was such a collaborative, community effort, the best answer is probably that not one single person was solely responsible.

Of important note was the use of packet switching, rather than circuit switching, in ARPANET. In circuit switched networks, a direct, physical connection has to be made between the two parties communicating. To make different connections, the actual infrastructure of the network has to be changed (example: telephone operators switching cable connections). In a packet switched network, on the other hand, lines in the network are shared (multiplexing) and traffic is managed by routers. In this system, the physical infrastructure of the network doesn’t have to be changed to accommodate different connections, and ideal routes through the network can be determined dynamically. This kind of network is what makes the internet as we know it possible.

The next topic of discussion was competition between humans and computers, and, more specifically, the supercomputers Deep Blue and Watson. Deep Blue is the name of a chess-playing computer that was created by IBM for the sole purpose playing chess. Deep Blue calculated its moves using brute force analysis – meaning that millions of possible moves would be considered every turn to find the most advantageous one. This kind of processing heavy analysis was possible because of Deep Blue’s advanced processing capabilities and its specialized hardware. At its time, Deep Blue was the biggest and most powerful supercomputer in the world – it could calculate around 200 million moves per second. Renowned chess player Garry Kasparov was defeated by Deep Blue in 1997.

Video: Deep Blue beat G. Kasparov in 1997

The other supercomputer we discussed was Watson, also designed by IBM. Watson was created to be a contestant on the game show Jeopardy. Because being successful on Jeopardy requires speedy interpretation of puns and other lingual tricks, this is a daunting task for a computer; it requires complex language analysis. But, with the ability to evaluate around 200 million pages of content per question and almost 3000 processor cores, Watson was able to defeat Jeopardy star Ken Jennings in a special match (video below). A simplified explanation of Watson’s method: it selects key words from clues, runs them through its 15 terabyte knowledge stores, and then calculates the probability of the answer it has found being correct. If this probability meets a certain threshold, then Watson buzzes in.

Video: Jeopardy! IBM Watson Day 3

Although Watson’s algorithms and processing speed allow it to determine the correct answer a lot of the time, its occasional erratic behavior betrays its non-human nature. For example: choosing a person’s name as an answer when it’s apparent that the clue is suggesting a book, or the oddly specific bet amounts chosen through statistical analysis. This, however, begs the question – is “human-like” behavior the ideal for artificial intelligence, or simply a bar to be exceeded?