Comment permalink

The Elusive Turing Test: What Happens When Two AI's Chat?

Artificial Intelligence may pass the Turing Test some day, but not yet.

The Turing Test, named for mathematician Alan Turing, is a kind of conceptual benchmark for artificial intelligence; one in which an artificial intelligence must be indistinguishable from a human in a conversation. That doesn’t mean that the voice simply sounds human, but that it is able to answer questions, make observations, and deliver instructions as a human would. It’s a deceptively tall order for engineers and robotics gurus to fill, but it looks that we may be on the eve of doing just that.

A recent feature in Wired.com looks at the prevalence of AI’s today, and at some of the recent developments within information technology, to determine whether the Turing Test can be tackled anytime soon. Artificial intelligence, defined as a program that can make a determination based on information and then execute it, is used ubiquitously today in ways that one probably wouldn’t even consider. GPS navigational systems, web browser search algorithms, stock trading algorithms and automated customer service programs are all examples of AI. In addition, two interrelated advances in information systems technology have pushed us closer to crossing the Turing threshold. One is the availability of enormous stores of raw data; from streamed video and audio to text conversations and technical documents. The second is the improved processing power necessary to analyze and utilize these staggering amounts of data.

However, simply warehousing vast amounts of analyzed data doesn’t begin to  recreate the human mind, because connections between data follows a structure of logic. If people are anything, it’s impulsive, erratic, emotional, and anything but logical. Instead, those concerned with creating a human-like conversation AI focus on probability, rather than logical connections; on, “calculating probabilities and producing complex behavior from the interaction of many small, simple processes.”

The result, at least at the moment, is something of a shady mockery of human communication. Attempting to find the kinds of tangential ideas, social and cultural variables, and relational cues that are an inherent part of human conversation is one of the biggest difficulties in recreating human speech. An open ended question or creative request can “break” a conversation AI (or chatbot). These shortcomings are readily apparent when you attempt to place two chatboxes together in a conversation (see the video below).

AI vs. AI. Two chatbots talking to each other

To gauge how far we truly are from a true AI, consider what simply passing the Turing Test actually accomplishes…a vague approximation of human conversation. It’s still mimickry, albeit an intricate and subtle one, of what people are able to accomplish. Even assuming that some AI is able to recreate human conversation, the distance between that and an approximation of something like human ingenuity, or creativity, is still a very long way off.