YIBADA

Computer With Built-In Algorithm Beats Man In A Turing Test

| Dec 13, 2015 01:25 AM EST

China has made considerable progress in artificial intelligence research.

Humans and artificial intelligence are undeniably different; however, one computer was found to beat man in a Turing test.

New York University scientists have been able to identify a way for computers to recreate simple symbols and drawings that look exactly the same and indistinguishable from those made man humans, according to the journal Science.

The scientists built an algorithm called Bayesian Program Learning (BPL), which aims to transform certain concepts into simple computer programs, enabling computers to learn a catalogue of visual concepts from a single case.

A case in point is letter A in the alphabet. In the digital alphabet, the letter would be represented by code. The twist is instead of a programmer writing the code, the computer generates the code to represent the letter. The researchers also integrated previous and current concept for the model to learn.

In the study, the research team tested their algorithm with humans and machines using the "visual Turing tests." Humans and computers with the algortihm were asked to reproduce a series of handwritten characters after being shown a single sample of each character.

The team gave human judges paired examples of the outputs of the humans and the machines together with the original prompt, and asked them to determine the output that was made by the computer.

According to the researchers, although human judges' correct responses were different for each character in the said test, 75 percent of human judges performed significantly lower than chance in detecting whether a human or a computer made the series of characters.

Allen Institute Intelligence CEI Oren Etzioni told Mashable that they had not beaten the Turing test any more than a calculator did by out-multiplying a human. For Etzioni, it was best termed as a "scientific contribution" and that the research was an invaluable reminder that there was a need of methods that could generalize small numbers to model human abilities.

With the Turing test result, it is assumed that the articifial intelligence could be at par with man's cognitive abilities in certain aspects in the future.

Most Popular

EDITOR'S PICK