Researchers at New York University have created an artificial intelligence system that models goal-setting according to human logic. The finding deepens our understanding of cognitive processes and promises significant progress in AI development.
In experiments, participants came up with game tasks using objects in a virtual room. The analysis showed that goal-setting is based on two principles: physical plausibility (objects behave according to the laws of physics) and combining familiar elements into new forms. These patterns formed the basis of a new AI algorithm. Participants who evaluated games created by AI and by humans saw no difference: the AI-generated tasks were just as interesting and original.
Previously, artificial intelligence could not come up with goals on its own; people set them. Now AI analyzes principles of human thought and creates its own tasks. This is a step toward creating more creative AI that better understands human intentions.
Artificial intelligence has moved one step closer to human thought. On the one hand, this makes interaction with machines easier and makes them more flexible and useful for creative tasks. On the other, the more AI resembles humans, the more important questions of control, ethics and the limits of its autonomy become. Where will the line fall between an assistant and an independent agent?
