The Turing Test
The main goal of the Lovelace test is to show whether a computer is ‘genuinely creative’. The test is presented through a computer which creates stories, and if a human can read the story created by the computer, and cannot explain how it was generated, the computer can be called ‘genuinely creative’. This test differs from the Turing Test in the specific goal, aiming to create new material, rather than possibly using existing material to trick player C. However, the Lovelace test also has its own flaws, such as the timeframes for each step of the test not being mentioned and raising questions about the nature of human creativity. In the Poetics , Aristotle discusses the nature of creativity, and originality in terms of Lovelace’s ideas, through poetry, describing how ‘imitation is implanted in man from childhood . . . through imitation he acquires his earliest learning’. This section of the text continues to describe how art and thoughts all came from nature originally, and then from previous thoughts or events. These ideas suggest that no thought is truly original, but generally thoughts can be called original if they are a new combination of previous thoughts and experiences. Applying this logic to the Lovelace Test, a computer would be able to create new material in the same way a human can, using previous ideas in a creative way. Academics suggest that for the Lovelace test, the person will have a reasonably long amount of time to find an explanation for the creation of the computer’s output. This vague limit means that they could not spend infinitely long tracing back all possible contributors to the creation, as you would also be able to do that for a human’s output. The Lovelace Test 2.0 was created by Mark Riedl in 2014. The idea is that the computer must create an artifact, any form of original content, based on a set of constraints set by a human. To pass the test, it must satisfy the set of constraints given, be realistic for a human to write, and for the requirement from the original test, the human cannot be able to explain how this outcome developed. In the paper, Riedl suggests the use of the Lovelace 2.0 Test to ‘quantify the creativity of an artificial agent, allowing for comparison of different systems’. He described giving each agent an increasing number of constraints, recording the amount before it fails the test. This can then be repeated for a mean number of passes, which can be compared between agents. Gary Marcus is a professor of cognitive science at New York University, who has been very critical of existing tests of computer intelligence, suggesting alternatives and variations throughout his career. In one article, Marcus uses an example of a chatbot which claimed to have beaten the Turing Test, when asked simple questions such as ‘How many legs does a camel have?’, the chatbot responds with ‘Something between 2 and 4. Maybe, three…’, showing that if a gap is found in its knowledge, it cannot use common sense and falls back on humour to cover it up. His proposed solutions require the AI model to show reasoning and explanation for its outputs. In 1990, the philanthropist Hugh Loebner set up the Loebner Prize, which uses a modified version of the Turing Test, rewarding the best performance. The competition used the test described by Turing in his original paper. The judges had to speak to a selection of bots and humans through computer terminals, eventually ranking which ones they think are the ‘most human’, where the highest-ranking bot wins the prize of $2,000-$3,000. The promoters of the competition also added an award for the ‘most human human’, which they require to have significant knowledge about chatbots and how they work. The Loebner Prize was not very highly regarded by the AI community and is mostly seen as a publicity stunt. A separate
82
Made with FlippingBook - PDF hosting