Turing Test: 50 Years Later

Visual Turing test for computer vision systems

“…Consequently, to generate tests with more than 10 or so questions, we are obliged to make "invariance" assumptions to allow for data pooling to expand the number of images from which these relative frequencies are computed. Specifically, if we assume that X q , q ∈ Q inst , given the history H ∈ H, depends only on a subsequence, H q ′ of H, then the distribution on X q is invariant to the questions and answers in H that were dropped, and the estimator [2] can be modified by substituting the condition HðIÞ = 1 by H′ q ðIÞ = 1. Let w ∈ W be the localizing region, possibly the entire image, referenced in the instantiation question q. H q ′ is derived from H by assuming that the event X q = x is independent of all attribute and relationship questions in H and all existence and uniqueness questions that involve localizations w′ ∈ W that are disjoint from w, with the important exception of uniqueness questions that answered positive (q′ ∈ Q uniq , X q′ = 1) and therefore instantiated a new object.…”

“…Alan Turing (1) proposed that the ultimate test of whether a machine could "think," or think at least as well as a person, was for a human judge to be unable to tell which was which based on natural language conversations in an appropriately cloaked scenario. In a much-discussed variation (sometimes called the "standard interpretation"), the objective is to measure how well a computer can imitate a human (2) in some circumscribed task normally associated with intelligent behavior, although the practical utility of "imitation" as a criterion for performance has also been questioned (3). In fact, the overwhelming focus of the modern artificial intelligence (AI) community has been to assess machine performance more directly by dedicated tests for specific tasks rather than debating about general "thinking" or Turing-like competitions between people and machines.…”

The machine as psychotherapist: impersonal communication with a machine

“…He, like many psychotherapists, believed that a computer would always fail the Turing test (Box 1), that it could so convincingly imitate a human being in its responses to any text input that it would be taken to be a human being. Reasons for failing the Turing test include an inability to be truly generative in response and an inability to have an adequate theory of mind (Saygin et al, 2000). Whether or not computers or other devices will always fail the Turing test for these or other reasons is a matter of prejudice rather than science.…”

A formal approach to exploring\\ the interrogator's perspective in the Turing test

“…Although it has been proposed more than fifty years ago, the TT is still considered as an attractive and fruitful idea, when it comes to its theoretical aspect (see [20,21,4]) as well as its practical applications (e.g. Loebner Test and captcha systems).…”