A software based on artificial intelligence has passed an eighth American school test (comparable to the third year of high school), according to an article in the New York Times. It is the first time that artificial intelligence has passed a test of this level.
It’s been a few years since hundreds of computer scientists entered a competition to create artificial intelligence that can pass a test of this level, but the only one to pass seems to have been the Allen Institute for Artificial Intelligence.
In fact, this Seattle Institute has created a new artificial intelligence system that seems to have passed the scientific test by correctly answering more than 90% of the questions. The software, called Aristo, is designed to mimic the logic of human decision-making.
And it is perhaps precisely for this reason that he managed to overcome not only the questions that made a “simple” information search possible (something that even Google can do now if the questions are very simple), but also questions that needed a real reasoning, essentially the classic and simple “problems” that primary or secondary school students have to solve, issues that, however, require the use of logic.
The standardized scientific tests used in schools are increasingly being used to assess the level of artificial intelligence and the manufacturers themselves see them as excellent benchmarks for understanding the progress and level their software achieves. These types of tests are considered more important than the classic tests based on games such as chess or backgammon.
The latter may, in fact, be governed by the rules to learn, but a scientific test, a series of questions that also includes the use of logic, is more difficult to overcome. Jingjing Liu, one of the Microsoft researchers who has also worked on various Allen Institute initiatives based on artificial intelligence, seems to be cautious and openly declares that it is not yet possible to compare such technology with real human students of the third degree: their ability to reason, at least for the moment, is still superior.
However, the progress that has been made with Aristo can already be used in the short term in a range of different services, ranging from the answers that an Internet search engine can provide to the various tasks that a digital assistant can perform. However, the progress made in artificial intelligence, especially in neural networks that can understand the natural language thanks to models built on the basis of huge amounts of data, does not seem to deny it.