The deep learning computer built by researchers outperformed humans on IQ test
For the first time ever a computer programmed to understand multiple meaning of words and sentences has beaten humans in the Intelligence Quotient test.
For ages, humans have given a lot of importance to the Intelligence Quotient (IQ); however when it comes to computers, IQ as such has very little value.
According to a recently released study, researchers from Microsoft and the University of Science and Technology of China programmed a computer to understand multiple meaning of words and sentences.
By building this deep learning machine they compared the IQ of Artificial Intelligence with that of humans and surprisingly the results showed that the machines outperformed an average human on the types of problems that have always been the toughest for computers to solve.
The test involved three categories of questionnaire:
- Logic questions which comprised of patterns in sequences of images
- Mathematical questions which consisted of patterns in sequences of numbers
- Verbal reasoning questions which dealt with analogies, classifications, synonyms and antonyms.
Huazheng Wang and pals at the University of Science and Technology of China and Bin Gao at Microsoft Research in Beijing, concentrated on this last category of test i.e. Verbal reasoning.
A review from M.I.T Technology states that whenever a natural language processing computer is posed with any questions from the verbal reasoning, then the performance of the computer is seen to be very poor and even worse when compared to the ability of an average human.
On the contrary, when researchers compared the deep learning machine and 200 human subjects at Amazon’s Mechanical Turk crowd-sourcing facility to answer the same verbal questions it was seen that the A.I powered system outperformed average human on these questions.
Earlier computer scientists used the data mining techniques to program the computers due to which the machines were able to analyze particular patterns of texts to find the links between words they contain and also determine as to how these words are related to each other. With this technique computers could accomplish the work of translation because it assumed each word has only one meaning.
When it comes to Verbal reasoning, computers are expected to understand words which have more than one meaning so as to be able to differentiate between the synonyms and antonyms which is where computers have shown a poor performance for ages and humans have always had an upper hand.
Researchers thus focused on going beyond the existing technologies and created a framework which comprised of three components as their aim was to make the computers efficient enough to solve the verbal comprehensive questions.
Very first element was the classifier which would help the machines to recognize the type of verbal question. Basically the computer needs to make out if it is an analogy or classification or synonym or antonym problem.
Next was the most intelligent move wherein the computer recognizes the multiple meanings of a word by matching it with the other related words that are contained in the dictionary.
The final step involves actual solving the problem using the definite meanings of the words based on all the data the machine collected.
Thus in all researchers developed a technique through which the computer was able to recognize the different meanings which a word can probably have.
Based on the overall education levels of the human participants the report stated: “Our RK model can reach the intelligence level between the people with the bachelor degrees and those with the master degrees, which also implies the great potential of the word embedding to comprehend human knowledge and form up certain intelligence.”
This is just a beginning step which promises a better development of the future computers.