We close the course with one last exploration of the world that Turing made. The Turing Machine (TM) was originally imagined as a way of thinking about mathematical algorithms, but not long after Turing’s original publication his way of thinking led to real machines, machines like those that challenge the chess grand master Kasparov or spell-check your term papers. It’s a spectacular case of basic research with an unforseen but beneficial practical payoffs.

In tonight's lab, we will look at the practical extension of Turing’s ideas in the field of AI (Artificial Intelligence). In 1950, Turing looked at the first massive and primitive computers and asked a 21st century question: How can we determine if a computer is intelligent? Turing recognized that looking or sounding like a human being was inessential to intelligence, and likewise that being housed in a gray box and made of silicon did not, by itself, guarantee stupidity. So how could you tell? He proposed a test which he called the Imitation Game. (It has since been called the Turing Test.) Turing argued (in "Computing Machinery and Intelligence") that the ability to carry on a sensible, coherent conversation was the best general test of intelligence. He also realized that you didn’t have to see and hear your conversational partner -- a good conversation could occur between two people at keyboards, each seeing only the typed words of the other. The Turing test, accordingly, involves two conversations-in-print: One terminal is linked in conversation to a human in another room; the other is linked to a computer. Your job, as the judge, is to determine which is which. If the computer fools you by imitating a plausible human, and does it so well that you can’t tell who is human and who is not, then the machine is intelligent.

However interesting the idea of machine intelligence may be, there is another issue here as well. Once we begin debating whether a given computer has passed the Turing test, or once we open the discussion on strategies to be followed by intelligent software, we suddenly find ourselves examining what it is to be human. In many ways the impact of computers goes beyond their practical and theoretical natures. It is not about them; it is about us, just as is all of cognitive science, in the end.

For forty years after the publication of Turing’s article, the Turing Test was merely a thought experiment. But that changed in 1991, when a modified version of the Turing Test took place at the Computer Museum in Boston. A Trinity delegation was there, and later I acquired transcripts from several conversations in the test (some human, some machine). As a warmup, you get to play the imitation game yourself, looking over the shoulder of the judges in the 1991 event.

1. Read through the four transcripts and decide which terminals are human, and which are computer output.

2. For the humans, circle one or more exchanges that helped you decide that it was a human writing. Next to the exchange, write what it is about it that makes it seem human to you.

3. For the computers, circle one or more exchanges that helped you to decide that it was computer output only. Next to the exchange, write what it is about it that makes it seem nonhuman to you.

We will discuss your observations up to this point in class. After this discussion, find one partner and turn your attention to ...

Turing imagined the Imitation Game occurring via teletype machines, huge clunky mechanical typewriter-printers, one of which was wired to a computer, the other to a hidden human teletype operator. Turing never foresaw the World Wide Web, with its endless conversations with invisible partners. In the next century, who knows how many of these conversations will have brains on one end, software on the other. Tonight we will meet a few of Turing’s mind-children, programs that talk with you. At present, these programs are rather crude, but it is quite amazing that they exist at all.

This lab has been developed by two of your hardworking TAs, Chris Marvin and Ian Sample, who have searched the Web for interactive programs that simulate human conversation. The sites they located have all been collected on a Natural Language homepage. This page, by the way, is part of a massive and fascinating Philosophy site built by Chris Marvin and the legendary Frank Sikenertsky -- which you can look at via buttons on the bottom of the page.

1. Open the Information Services folder in your Mac HD. Click on Netscape.

2. Open the following location:

One way to do this is by pulling down the File menu and highlighting Open Location.

3. To engage in conversations, click on the underlined phrases, and follow instructions as they are given. If you seem stuck, click on the Back button (or pull down the Go menu and highlight Back). Be patient, however, since sometimes the web can be a bit slow.

Since there will be many users from our class, you might choose a random starting point and bounce around in the list.

There are two questions; divide the labor as you see fit with your partner, if you have one. Make sure that both of your names appear on both parts of the report.

1. After you look over the conversationalists, select one that did moderately well, and one that did poorly, and discuss what "doing well" means. In what ways did the good performer seem (almost) human? In what ways did the poor performer obviously flop? Use examples and evidence to support your claims. (Don’t be distracted by web limitations, like speed or clunky interface issues.)

2. Since this is a new lab, and the first to exploit the web (in this course), propose some alternative labs that might cleverly examine the same issues. Be creative, go wild. Include enough explanation in your revised lab plan so that it is easy to see how to implement it in the future.