Did we just get one step closer to the singularity? Chinese researchers claim they’ve developed an AI that scored higher than the average human on the verbal portion of an intelligence test. According to the authors, the verbal section of an IQ test is more difficult for an AI to complete accurately than the quantitative section because the verbal portion requires knowledge of the nuances of polysemous words (words with multiple meanings) and the relationships between them. Usually, computer programs utilize word embedding technologies that allow the AI to learn just one vector per word, which is insufficient to pass this sort of test.
In this study, the researchers claim to have invented a new “knowledge-powered” type of word embedding that allows the AI to adjust its strategy depending on what type of question is being asked and to take into account the relationships between different words. Using this new method, the AI was not only able to satisfactorily complete the IQ test, but were able to score slightly higher than the average human.
The authors wrote in their paper: “The results are highly encouraging, indicating that with appropriate uses of the deep learning technologies, we could be a further small step closer to the human intelligence.”
These results are, in fact, groundbreaking, as the most successful AI IQ test so far measured the program as having the intelligence of a human four-year-old, while this AI was measured as equally intelligent to most humans. But, that being said, there are several qualifications to consider. First, many experts believe that the IQ test itself is flawed, as it only measures a normative type of intelligence and doesn’t take into account other types of mental ability, such as creativity or emotional intelligence. Some dispute it as a measure of intelligence altogether, as intelligence is a difficult concept to define and test bias has been demonstrated against certain marginalized groups. But even if it’s not exactly a measure of intelligence, most experts contend that it is correlated with intelligence, so these results should still be taken very seriously.
This study was also published on an online database, and has not yet been accepted to a scientific journal. But still, experts not involved in the study have praised the findings, even as they caution that this technology is in its infancy, and we’re not anywhere near Hal 9000 yet. Hannaneh Hajishirzi, an electrical engineer and computer scientist at the University of Washington in Seattle, told Business Insider that the researchers in this study “got interesting results in terms of comparison with humans on this test of verbal questions, we’re still far away from making a system that can reason like humans.”
Robert Sloan, a computer scientist at the University of Illinois at Chicago, similarly acknowledged that this AI was a small step forward, but claimed that there was no guarantee the program would be able to perform as well as a human on an open-ended test, as opposed to multiple-choice. In the field of artificial intelligence research, Sloan said, “the places where so far we’ve seen very little progress have to do with open dialogue and social understanding.” For example, a normal human child would be expected to understand that if he or she sees an adult lying in the street, then he or she should call for help. “Right now, theres no way you could write a computer program to do that.”