Artificial intelligence has a long way to go and one of the challenges that stands on the road to success is the Turing Test. Developed by Alan Turing in the fifties, the test measures AI ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human. Zackary Scholl developed a program to produce a poem that passed the Turing Test, questioning the source reliability of creative materials.
In the classic illustration of the Turing Test, an interviewer communicates with a computer and a human without seeing them. Then the interviewer is asked to state which one of them is more human-like. If the computer is picked over the human, then it means that the computer passes the Turing Test.
This is what Scholl achieved with his undergrad project at Duke University. Now a PhD candidate in computational biology, Scholl stated in a recent blog post that the system he developed uses context-free grammar system to produce full poems. The program operates by incrementally dividing the poem to its components such as stanzas, lines, phrases, verbs, adjectives and nouns. “When a call to create a poem is made, then it randomly selects components of the poem and recursively generates each of those” Scholl explains.
Scholl reveals in his blog that he began by submitting his automated poems to online poetry websites. Then he went on submit them in a literary journal of his University (pictured above). The Turing Test of his AI was to get his poems accepted by a literary journal, which it did ever so successfully.
“Perhaps in the future we will need to question the source of creative materials to determine whether they are indeed human or machine made” Scholl explains in conclusion to his post.
Story via Motherboard, image via Raspberry PI AI