Measuring the Responses of AI Programs: Alan Turing’s Test of Machine Intelligence Applications

Alan Turing wrote Computing Machinery and Intelligence in 1950. This paper, written before any AI applications were created, was purely theoretical. In it, Turing described his ideas about artificial intelligence, as well as describing ways to test for human-like responses. Using Turing’s criteria, thousands of AI programs are put to the test every year, competing for the $100,000.00 Loebner Prize in Artificial Intelligence.

AI Programs Mimic Human Responses

In what Turing called the “Imitation Game”, AI software is often programmed to respond following a complex set of rules in order to appear as human-like as possible. Despite the years that have been devoted to the attempt to teach AI programs to mimic human responses, success has not yet been achieved on this front.

The ability to successfully deceive a judge into believing an AI program’s responses were created by a human being is the main criteria for passing the Turing Test. In his paper, Turing described a situation in which a judge is allowed to ask questions, and evaluate typewritten responses. When a judge is unable to determine which respondent is the machine, and which is the human, the AI program is deemed to be successful in its mimicry.

The Loebner Prize Provides Incentive for AI Research and Development

The Loebner competition was established in 1990, as a cooperative effort between Hugh Loebner and The Cambridge Center for Behavioral Studies. Each year, AI programmers compete in hopes of developing the first AI program to pass the Turing Test. A $2,000.00 consolation prize is awarded each year, along with a bronze medal, to the AI software program that achieves the most human-like responses during the contest.

Controversy Surrounding the Loebner Prize

While the Turing test is commonly accepted as the measure by which all artificial intelligence programming should be measured, it is possible that “teaching to the test” may be stunting the growth of artificial intelligence programs. The ability to formulate human-like answers would demonstrate excellence in the field of natural language processing, but other areas of artificial intelligence are neglected by the Turing Test. For example, abilities of automated vehicles, such as BOSS, the AI program that drove an SUV to win the DARPA Urban Challenge, cannot be measured by the criteria set forth in Turing’s paper on AI.

Artificial Intelligence Programming is a branch of science with widely varying criteria and usage. Artificial Intelligence Software can be used to protect consumers from credit card fraud, or to Enhance the Smart Home Security System, creating a safer environment for the elderly and disabled.

Leave a Reply

Your email address will not be published. Required fields are marked *