While our recent times and evolving technologies have tended to stoke the debate concerning the rise of AI, most of it has to do with the ‘intelligence’ (or capability) intrinsic to the machine. But how does the progressing AI measure up to actual human intelligence? Well a team of scientists from University of Illinois (at Chicago) and a research group in Hungary, have done their best to gauge the equivalency of AI to humans. The result showed that the AI scored the equivalent of a WPPSI-III VIQ, which is the standard IQ for a four-year old child, but lower than 6 to 7 years old children. As for the AI system in question here, the researchers used an open-source project called ConceptNet, which is being developed by MIT Common Sense Computing Initiative since 1990.
The ‘WPPSI’ part of the test stands for Wechsler Preschool and Primary Scale of Intelligence. It is generally utilized for estimating IQs for children ranging from the age of 2 years 6 months to 7 years 3 months, and measures children’s performance in five different fields: information, vocabulary, word reasoning, comprehension, and similarities. Now of course, the questions were not asked to the AI in a general format. Instead the researchers modified the question patterns by some programming, so as to conform to the computer’s structure and knowledge database relating to our world.
As for the results, there were some interesting insights into the ‘thought’ pattern of the machine. To that end, the scientists found out that the ConceptNet tended to do well on vocabulary and similarities, while faring poorly in word reasoning and comprehension. For example, when a comprehension-category question ‘why do we shake hands?’ was asked to the machine, the computer utilized a three-pronged concept that entailed individually searching the words ‘hand’ and ‘shake’ and then searching their combined form ‘hand shake’. From these searches, the machine came upon the wrong (yet technically justified) answer – ‘epileptic fit’.
In essence, the AI was was found to have the IQ proficiency of an average four-year old child, in spite of it using a three concepts approach. And while the researchers are quite definite about this measured end result, they are still not sure of the confused answers that conflict with the technical aptitude of the machine. However, the scientists have already put forward their notions on how to actually develop and increase the ‘human’ IQ of the computer. One of such ideas pertains to endowing the artificial intelligence with natural language processing capabilities. This would avoid or at least reduce the dependency on built-in programming. Such abilities would be similar in scope to online assistants like Siri, Cortana and Google Now.
Furthermore, it should also be noted that the AI used in this case was developed in 2012, and a lot has happened (and evolved) in this field since the last three years. As MIT Technology Review put forth their observation –
Taking Ohlsson and co’s [the research team’s] result at face value, it’s taken 60 years of AI research to build a machine in 2012 that can come anywhere close to matching the common sense reasoning of a four-year old. But the nature of exponential improvements raises the prospect that the next six years might produce similarly dramatic improvements. So a question that we ought to be considering with urgency is: what kind of AI machine might we be grappling with in 2018?