One of the smartest computers in the world has recently passed an intelligence test. This test was a set of tasks that are commonly found in IQ tests. Surprisingly, the ConceptNet 4 artificial intelligence system, created by scientists from the Massachusetts Institute of Technology, showed very impressive results for a soulless machine: the computer coped with the task at the level of a four-year-old child.
The results, of course, are impressive, but Robert Sloan, the lead researcher of this project, hastened to explain that not everything is as wonderful as it seems:
“If an ordinary child had shown such results, then most likely the question would have arisen that something was clearly wrong with the development of his mental abilities.”
The basis of the entire test was to determine the ability of the machine to compare concepts and properties that are contained in the gigantic complex of semantic connections ConceptNet 4, but without affecting the significance of the reason why one or another choice was made. That is, based on the data embedded in it, the AI knows that a saxophone is used in jazz. And can answer questions related to it. Or an evil example: a robot has a knife - a person is dead. The information about the result of such a consequence is not relevant within the framework of the question.
Sloan says that the machine copes well with tasks on vocabulary or searching for similarity of objects. However, she is not able to draw even simple logical conclusions. That is, the AI knows, for example, that water freezes at a temperature of 0 degrees Celsius, but can not answer the question of whether the ice is cold. In other words, questions of the why category are baffling him. And the next task of scientists is precisely to teach artificial intelligence to come to logical conclusions.
The article is based on materials .
Comments
Post a Comment