Deep Learning System Tops Humans in IQ Test

 Deep Learning  

Using new techniques in deep learning, researchers have created a system that has outperformed human subjects in a test of verbal reasoning. The researchers indicate that the results bring us closer to the true human intelligence.

Researchers from Microsoft and the University of Science and Technology of China built a deep learning system that has outperformed average human scores on the types of problems that have always been toughest for computers, according to their recently released recently released study.

Related articles What Does Watson Tell Us About the Philosophy of AI?  Stuart Armstrong and Nick Hawes Discuss the Future of Artificial Intelligence  The Rise of Artificial Intelligence

The test was made up of three categories of questions: logic questions (patterns in sequences of images); mathematical questions (patterns in sequences of numbers); and verbal reasoning questions (questions dealing with analogies, classifications, synonyms and antonyms), reported The Observer.

Computer scientists have previously used data mining techniques to analyze sets of text to find links between words and how those words related to each other. This technique was useful for translation and other tasks, but it typically restricted words to single meanings, unable to cope with the richness and vagueness that makes up much of language.

The scientists look beyond existing technologies to automatically solve verbal comprehension questions, creating an entirely new framework made up of three components. They called their system the RK Model.

"With appropriately leveraging the deep learning technologies, we could be a further step closer to the true human intelligence."

The first was an element classifier that sorted the specific type of a verbal question, such as analogy, classification, synonym or antonym. The second component involved leveraging a word embedding method that considers the multi-sense nature of words and the relational knowledge among words contained in dictionaries. Third, for each specific type of question, they created a simple yet effective solver based on the obtained distributed word representations and relation representations.

The overall result was a systematized approach for recognizing the different meanings a word can have.

The researchers then compared the deep learning system with 200 human subjects using Amazon’s Mechanical Turk crowdsourcing service.  The artificial intelligence and the people answered the same set of verbal questions.

The system performed better than the average human, concluded the researchers.

“Our RK modelcan reach the competitive performance of the human under the age from 40 to 60, which indicates the great potential of the word embedding to comprehend human knowledge and
form up certain intelligence,” the researchers wrote.

"While this work is a very early attempt to solve IQ Test using AI techniques, the evaluation results are highly encouraging and indicate that, with appropriately leveraging the deep learning technologies, we could be a further step closer to the true human intelligence," wrote the researchers.


SOURCE  The Observer

By 33rd Square

Embed