After undergoing several tests and validations, the model was able to assess the basic psychological thinking abilities of the AI model
AI systems are rapidly evolving to become man’s new best friend. With this technological innovation, AIs can now write poetry and help doctors perform surgery. But one thing they can’t do — much easier than everyone else — is use common sense.
To beat this difficulty, researchers at IBM, MIT, and Harvard have created a model called AGENT (Action-Goal-Efficiency-coNstraint-uTility), which claims that a new ‘common sense test’ can produce smarter machines.
By going through several testing and validation processes, the benchmark was able to assess the basic psychological capability of the AI model. This means that it can provide a sense of social awareness, interacting with people in real-world settings. The model is based on a large-scale 3D animation dataset inspired by experiments on cognitive development in children.
According to IBM, the animation depicts someone interacting with different objects under different physical constraints. “The videos include various trials, each of which includes one or more videos devoted to introducing the agent’s typical behavior in a particular physical environment, combined with test videos of the same agent’s behavior in the new environment, which are labeled ‘expected’ or ‘surprising,’ given the agent’s behavior in the respective introductory videos.”
The court seeks to demonstrate common sense. The AI model should have built-in insights into how people plan. This would be accomplished by combining the basics of physics and ‘price/reward ratios.’ Although the results are not yet perfect, AGENT is a promising diagnostic tool for developing and evaluating common sense in AI.
The company claims that in the future, AGENT could help significantly reduce the need for training, allowing businesses to save time, computing energy and money. Since robots do not yet understand human consciousness, the development of such tools can measure how close we are to the level of AI development.
