Humans are gradually becoming surpassed by machines in a number of fields. Yet, IQ tests have long proven that mankind still reigns supreme in terms of abstract thought. That is no longer the case, as scientists have now developed a deep learning machine smarter than your average college-educated human.
A computer has beaten humans in IQ tests for the first time since programmers made a breakthrough to teach it the multiple meanings of words and sentences.
A technological breakthrough made by programmers taking a new approach to language has helped the computers outperform humans in the verbal reasoning part of the Intelligence Quotient (IQ) test.
Questions in the test are in one of three categories: logic (looking for patterns in sequences of images), maths (finding patterns in numbers) and verbal reasoning.
Computers have previously struggled in the final category, which deals with analogies, classification, synonyms and antonyms. But researchers had a special new machine and 200 people answer the same verbal reasoning questions at Amazon’s Mechanical Turk crowdsourcing facility and the system performed better than the average human.
To do so, researchers from Microsoft and the University of Science and Technology of China built a deep learning machine that takes a new approach to the problems and outperforms the average human according to the report.
Previously, computers have used a data mining technique, taking large amounts of texts and finding links to other words to work out how the words relate to each other.
This is effective for tasks such as translating but it works by assuming a word has a single meaning which is why computers have struggled with synonyms and antonyms, something humans can determine quite simply with context.
To remedy this, programmers had to build a whole new method for their computer to use to solve verbal comprehension questions. Firstly, the computer recognizes if the IQ question is an analogy, classification, synonym or antonym question.
But the clever part is the second stage, when the computer considers the multiple meanings of a word by looking at the other related words, essentially using reason to establish context.
Finally, using the meanings it has established, the computer solves the problem.
And it’s a promising step towards further development in the future.
The report concludes: ‘While this work is a very early attempt to solve IQ Test using AI techniques, he 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.’