The Turing Test: The History Behind Measuring Artificial Intelligence

Michael Koch originally published this blog on his professional overview website. Check back for weekly updates on the state of AI, SaaS, PaaS, and the tech industry.

People often wonder the origin of AI theory, I want to give a quick background so people can better understand the fundamentals and breakthroughs on this subject. As a thought leader in the future of Cognitive Computing, I want people to first understand the past before we tackle the more complex cases in the evolution of AI and big data.

The Turing Test was developed by Alan Turing in 1950, and is a test of a machine’s ability to exhibit intelligent behavior. It is done through a text interface between a human evaluator, the machine, and another human. The evaluator does not know who he or she is talking to — so the machine and the human are hidden from him or her. The evaluator must then try to determine, through language, which is the machine. If the human is unable to do so, then the test succeeds in exhibiting that AI is equivalent or has surpassed the human in intelligence.

Although the test is not perfect, there are some strengths too it. Humans are creatures of language, and it is through language that we have evolved into the intelligent species that we are today. The fact that the test is done through a language interface would signal a good start to identifying whether something is intelligent or not. Through language we should reasonably be able to measure intelligence, and so a strength would be that the test gives something that is measurable. If a machine could answer questions through language, it would have to be able to reason with the information it was given. The ability to reason means that some level of conscious work would needed to be used by the machine — to make sense of the information whilst applying logic.

The Test could, however, be a flawed way of measuring intelligence. For example: the machine could try and mimic typical human errors in language, and so trick the evaluator. Also, a machine would have to downplay, and act less intelligent for some questions. If the machine answered a very complicated math question in a short space of time, the evaluator would know that is was the machine that was answering. We can see the gaps start to appear in the test. Although the Turing test may have some relevancy, it seems there are too many holes for it to be considered a complete approach in measuring AI. The journey continues.

We live in an exciting time where Artificial intelligence is becoming more involved in everyone’s day to day and ESPECIALLY their business decisions with data. Understanding the foundational aspects of this evolving space will ensure that no person or (company) is left behind in this new age of computing.