Summary with the article: Social interaction with computers. An interpretation of Weizenbaum's ELIZA an her heritage by Pruijt (2006). Illustrating some aspects of early artificial intelligence research.

It is not difficult to see that automation is on the rise. But people ask themselves what the limits are. There have been many debates about this subject. One contribution to this debate is ELIZA. This is a chatterbot. ELIZA played a role in the debate on artificial intelligence in the 1970s. It was especially used to show that artificial intelligence can hardly be more than simulated rather than emulated intelligence. ELIZA still exists and recently took part in a discussion on identity and online interaction, the future of narrative and emotional design. ELIZA is not so technically sophisticated, but became popular because the social interaction with computers. The aim of the writers is to provide more information on ELIZA and to give more information on the limited store of case studies that are possible to use in social science computing courses.

Most literature usually tells about just one small portion ELIZA can do. One research explained ELIZA as an early NLP (natural language processing) program. This program mimicked a psychoanalyst by picking out key words in the typed input and it printed out modifications of a compile of standard phrases. This is correct, but ELIZA does more. ELIZA uses pattern matching to try and detect words of phrases in the input to determine what the user is talking about. Using this, appropriate responses are constructed. The writer sees ELIZA as a good partner in social interaction. ELIZA is much different from natural language help systems, like the system present in MS Office. The writer gives a sample conversation between ELIZA and a social research advisor. ELIZA’s responses are Italicized (this conversation can be found on page 3 and 4 of the article):

Hi. Tell me about your research ideas.

I want to study street violence.

Could you describe the problem of street violence in the shape of a conceptual model?

One independent variable is aggression.

How do you wish to test aggression?

Maybe there is a standard test.

It is always important to choose the right test.

Or a questionnaire

Could you tell me something more about street violence?

It happens mainly after dark.

Let us return to aggression.

Often there is alcohol involved.

Good.

And it is a male affair.

Could you tell me something more about the subject?

Gangs are a big part of the problem.

Let us try to focus our conversation a little bit more.

As you can see, ELIZA accepts an input sentence from the user and transforms it into an output sentence. There are four steps needed for this transformation

• Step 1. The program gets rid of anything that was entered after the first colon.

• Step 2. ELIZA uses a search-and-replace table to make substitutions. One example is that I becomes you. This table is put in a worksheet and it is therefore easy to change or add search-and-replace combinations.

• Step 3. In this step, the input sentence is broken down into its components parts. This is done by certain decomposition rules.

• Step 4. ELIZA assembles an output sentence. ELIZA may use words from the input sentence and also use new material. Reassembly rules can be used.

The rules and the search-and-replace table constitute the script. ELIZA has one worksheet to show the recorded conversation and another worksheet shows a trace of all the operations performed by the program. A realistic conversation should be possible when the context is chosen in a way that it is logic for the user to supply all the substantive information that is needed to keep the conversation going.

The writer wrote an ELIZA script. The context was one in which the student (user) who is planning to do research, asks a lecturer (ELIZA) how to proceed. The researcher doesn’t know anything about the subject but tries to help the student clarify his or her thoughts. The writer asked students to help writing scripts for ELIZA. Apart from writing scripts, students could also analyze and describe bugs in ELIZA as well as suggest improvements. With the help of the students, the writer could make a great version of ELIZA.

Of course, there are limitations to computer programs. Dreyfus said that common-sense knowledge about the world is essential to intelligent behaviour. He also said that computers can only process such knowledge if humans can reduce it to discrete, determinate elements and rules that define relationships between these elements. But humans do not think by manipulating elements and relationships and research in artificial intelligent has not found the faintest notion of what common sense knowledge primitives should look like. Researchers have warned that we can’t make programs that can ever be made to understand anything. Searle said that intentionality was missing in computers. He saw intentionality as a unique feature of thinking as a biological process.

Interaction with computers does not only depend on the characteristics of the computer program, but also on the human capability of adaptation. According to research, users’ reactions to ELIZA showed a tendency to treat responsive computer programs as more intelligent than they actually are. Users believed that ELIZA could understand the information that they entered into the system. Another thing research might look at is emotion. One researcher used ELIZA to induce emotions in people. Research could possible look at whether ELIZA (or ELIZA-like programs) can cause or reinforce positive emotion.