Researchers Developed Artificial Intelligence (AI) That Thinks Like A Baby to Learn How to Use Common Sense

Scientists have longed to create an artificial intelligence (AI) that thinks like a human. Although there have been recent issues about an AI gaining consciousness and hiring a lawyer, another team of researchers claim to have created an AI that can think like a baby.

Researchers from DeepMind said their AI "intuitive physics," has common sense. The new AI called Physics Learning through Auto-encoding and Tracking Objects (PLATO) was trained using a set of animated balls moving and demonstrated positive learning results.

Teaching an AI "Intuitive Physics" A.K.A. Common Sense

Having common sense seems intuitive to humans even as a baby, unlike AI, which has yet to demonstrate this knowledge despite mastering complex games, such as chess and poker. If AI gets to learn common sense then it will give insights into how it works and how it arises in humans.

In the study titled "Intuitive Physics Learning in a Deep-Learning Model Inspired by Developmental Psychology," published in the journal Nature Human Behavior, a team from DeepMind takes a step toward advancing by incorporating common sense into machines.

According to Scientific American, they developed an intuitive physics model that integrates inherent knowledge that babies are born with into an AI. Also, the team created a testing method almost similar to the cognitive tests used in human babies.

The AI was previously trained to identify facers and balls, but they have not yet learned to predict what will happen to objects when they move or bump into each other. The team presented a model called PLATO that focuses on whole objects and trained it using 300,000 videos so that it could learn how balls move.

The goal is to teach PLATO what violates the laws of intuitive physics-based on object permanence, solidity, continuity, unchangeableness, and directional inertia. Like an infant, the AI acted "surprised" when the ball moved in another direction where it did not expect it to go. Also, PLATO has become better at predicting possible versus impossible scenes than other AI.

The new study shows that teaching AI intuitive physics is also like a baby first displaying awareness of what an object is and then learning the physical rules that govern its behavior by watching it move. DeepMind researchers noted that as of now, their work is not ready to advance to robotics, self-driving cars, or any other AI applications as more research and training are needed.

Importance of Intuitive Physics in AI

The intuitive understanding of humans in their physical world is crucial to everyday life. From the way, people fill their cup of coffee, stack items, and navigate around a slippery patch of ice, intuitive physics is at play. Numerous aspects of cognition, such as spatial thinking, temporal prediction, and working memory, contribute to this mental machinery.

Some psychologists believe that physical reasoning rests on something extra, like a devoted set of mental resources that process information from cognitive systems, while others think that it is merely the sum of its parts that humans apply to the physical world.

Intuition is only natural for humans. On the other hand, it will take a long time before AI can achieve common sense reasoning. However, given the interest in intuitive physics for machines, the Center for Brains and Minds said that AI models could someday learn it.

Experimental evidence suggests that neural models could be used to understand the cognitive and neural underpinning of intuitive physics in humans that can also provide AI applications with better inference and reasoning abilities.

Check out more news and information on Artificial Intelligence in Science Times.