Was It a “Common Sense” Test that Kept AI Machines from Becoming Intelligent

What Separates AI From an Idiot Savant Is Common Sense! Why is Ai Common Sense So Important?

AI systems of today are swiftly growing to replace humans as our species' closest friend. We now have AI capable of creating poetry, award-winning whiskey, and assisting surgeons during incredibly precise surgical procedures. The one thing that they are unable to accomplish is to employ common sense even though simple.

In contrast to intelligence, common sense is something that most people possess naturally and innately that helps them get by in daily life. It is something that cannot truly be taught. G. K. Chesterton, a philosopher, stated in 1906 that "common sense is a wild thing, barbarian, and beyond rules." Of course, algorithms—which are simply rules—are what power robots. Therefore, robots cannot yet employ common sense. But contemporary research in the area has enabled us to measure Artificial intelligence's fundamental capacity for psychological thinking, moving us one step closer. Artificial intelligence and AI common sense matter a lot.

Hector Levesque quoted "Without a common sense understanding of the world, the AI systems, even the most advanced ones, will remain somewhat like idiot-savants."

What is Common Sense?

Consider this: How would an automatic car know that a snowman standing on the sidewalk won't try to cross it? People utilize common sense to understand that is not going to happen.

Why is it so challenging for us to provide common sense knowledge to intelligent agents? As demonstrated in the preceding case, we apply this knowledge automatically and naturally. Frequently, we do it without even being aware of it.

All of our background information about the physical and social worlds that we have gathered throughout our lives can be summed up as common sense. It encompasses factors like how we perceive physics (causality, hot and cold), as well as how we anticipate people will act.

So why does it matter if we teach AI common sense?

In the end, common sense will improve AI's ability to assist us in resolving problems in the real world. Many contend that AI-driven solutions frequently fall short when applied to real-world situations where the challenges are unpredictable, ambiguous, and not governed by rules, such as when diagnosing Covid-19 therapies. Better customer service, where a robot can assist a dissatisfied customer instead of sending them into an unending "Choose from the following" loop, could result from injecting common sense into AI. It may improve the responsiveness of autonomous vehicles to unforeseen accidents on the road. Information on intelligence signals of life or death might potentially be helpful to the military.

So why haven't scientists been able to crack the "common sense" code thus far?

Referred to as "AI's dark matter" The future of AI depends on the elusive and crucial common sense. In truth, giving computers common sense has always been an aim of computer science; in 1958, John McCarthy, a pioneer in the field, released a paper titled "Common Sense Programs" that examined the use of logic as a means of storing data in computer memory. But we haven't made much progress in realizing it since then.

In addition to social skills and logic, common sense also includes a "naive sense of physics," or the understanding of some physical principles without the need to solve physics equations, such as why it is improper to place a bowling ball on a tilted surface. For us to plan, estimate, and organize, it also incorporates a fundamental understanding of abstract concepts like time and location. According to Michael Witbrock, an AI researcher at the University of Auckland, "it is information that you should have."

All of this means that common sense cannot be simply defined by rules because it is not a single definite thing.

AGENT

What then is AGENT? A sizable collection of 3D animations called AGENT was motivated by research on the cognitive growth of young children. The animations show a person interacting with various objects while facing various physical limitations.

This is exactly what has been developed by IBM, MIT, and Harvard researchers under the name AGENT, which stands for Action-Goal-Efficiency-coNstraint-uTility. This benchmark is capable of assessing an AI model's fundamental capacity for psychological reasoning following testing and validation. As a result, it can foster social awareness and allow users to engage with others in the real world.

The agent's activities in the "test" movies must then be evaluated by a model based on the behaviors it learned in the "familiarisation" videos. The model is then tested against extensive human-rating trials using the AGENT benchmark, in which participants assessed the "shocking" test movies as being more surprising than the "anticipated" test videos.