2050-11-17 Press Release: Hemingway® AI Achieves New Level Of Processing
From: Carl Douglas, Head of Hemingway Development Team
To: Various Media Outlets and Research Journals
Subject: Hemingway® Passes Turing Test
Hemingway Systems is proud to announce that our breakthrough AI technology has passed one of the most difficult tests for an AI - the Turing Test.
The Turing Test is the first step in evaluating the effectiveness of an AI. At its simplest, a questioner sends queries to two respondents. One is a human, and the other an AI. The respondents answer the questions. The questioner is then asked to determine which of the answers came from the AI and which from the human. If the respondent cannot determine with greater than 50% accuracy which answers were from the AI, the machine passes the test.
The goal of the Turing Test is to see if a computer can pass for humans. While this is a very basic test, we are incredibly happy with the results. In 300 trials, in only 2 were the questioners able to pick out the AI answers more than 50% of the time. In all the other trials, the selection range was 33.2%-46.7%. This indicated that the questioners were fooled by the AI more often than not.
This is, of course, the first step. Hemingway's AI algorithms are learning written and verbal speech patterns, and applying them to their responses. With each question and response, the AI modifies its "personality" and response patterns in an effort to respond more naturally to queries.
To aid in this learning process, the team has been feeding Heminway with great works of literature from all eras. Each piece of literature is "digested" by Hemingway and compared to other literature already digested. Hemingway is then instructed to compare works of literature and assign them to classes, such as fiction or non-fiction, or genres of writing. As Hemingway makes the comparisons, our team reviews the ratings made by Hemingway and makes corrections to the comparisons as needed. By then adding these corrections to its computations, the accuracy of its comparisons increases. The goal is to then have Hemingway begin to use these comparisons to adjust its own communications. The Turing Test was the first step in seeing whether we were moving in the right direction. Our next test will be a bit of a reverse. We will have two human respondents talking to Hemingway in a three-way "conversation". At the end of each trial. One participant (randomly chosen between the Ai and the humans) will be given a topic to discuss. The AI will participate in the conversation as will the humans. If, in the end, neither human can identify the computer responses with more than 50% accuracy, we will consider the test a success. We expect to begin these trials in January of 2051.
The goal of the Turing Test is to see if a computer can pass for humans. While this is a very basic test, we are incredibly happy with the results. In 300 trials, in only 2 were the questioners able to pick out the AI answers more than 50% of the time. In all the other trials, the selection range was 33.2%-46.7%. This indicated that the questioners were fooled by the AI more often than not.
This is, of course, the first step. Hemingway's AI algorithms are learning written and verbal speech patterns, and applying them to their responses. With each question and response, the AI modifies its "personality" and response patterns in an effort to respond more naturally to queries.
To aid in this learning process, the team has been feeding Heminway with great works of literature from all eras. Each piece of literature is "digested" by Hemingway and compared to other literature already digested. Hemingway is then instructed to compare works of literature and assign them to classes, such as fiction or non-fiction, or genres of writing. As Hemingway makes the comparisons, our team reviews the ratings made by Hemingway and makes corrections to the comparisons as needed. By then adding these corrections to its computations, the accuracy of its comparisons increases. The goal is to then have Hemingway begin to use these comparisons to adjust its own communications. The Turing Test was the first step in seeing whether we were moving in the right direction. Our next test will be a bit of a reverse. We will have two human respondents talking to Hemingway in a three-way "conversation". At the end of each trial. One participant (randomly chosen between the Ai and the humans) will be given a topic to discuss. The AI will participate in the conversation as will the humans. If, in the end, neither human can identify the computer responses with more than 50% accuracy, we will consider the test a success. We expect to begin these trials in January of 2051.
Comments