Why is common sense crucial for creating AGI?
Common sense generally refers to reasonable, sensible, and commonly accepted ideas, beliefs, or principles in society, culture, and daily life. People don’t need to explain too much, nor do they need to argue, because common sense is the shared understanding among people in a specific field.
Unlike common knowledge, common sense is more subjective and influenced by individual experiences and cultural background. It is based on what people believe to be logical. Common knowledge, on the other hand, is often based on facts and is widely known by everyone. It can often be verified through various sources and is more objective in nature.
Imagine a newborn baby encountering hot water for the first time. Placed in a room with a basin of hot water, they have no prior knowledge of its temperature. When the baby touches the hot water and experiences pain and discomfort, potential reactions such as crying or pulling away showcase an instinctual understanding of potential harm. This experience is crucial for the development of common sense in humans. While everyone is expected to acquire this understanding, it is not an innate trait; instead, it is something individuals accumulate throughout their lives through various experiences.
As the baby grows, caregivers take on the responsibility of educating about everyday things. For example, when giving the baby a bath, caregivers feel the water with their hands to ensure it’s not too hot. Through this repeated action by caregivers, the baby learns how to test water temperature and associates hot water with the risk of burning. Caregivers also use words like “hot” or “warm” , helping the baby learn language and making sure they understand about temperature.
Through experiences, guidance from caregivers, and their own continuous exploration, the baby gradually learns about the common knowledge of hot water and develops a sense of safety. This common knowledge and common sense contribute to a broader understanding of the world, helping them cope with different situations to ensure safety.
In the current development of Artificial Intelligence, creating a machine capable of emulating human common sense is essential for building AGI, and scientists are still working hard in this direction. The crux of the matter lies in the fact that a baby’s common sense knowledge is built upon direct experiences, generalizations, and inferences drawn from those experiences.
Conversely, machines lack the ability to learn from experiences in the same manner. They depend on explicit data and programming to recognize and respond to potentially perilous situations, with their responses often rooted in statistical patterns rather than the kind of common sense reasoning employed by humans.
Therefore, artificial intelligence excels in relying on explicit knowledge, but in terms of acquiring common sense through experience, it still lags far behind humans.
To enable AI systems to comprehend and reason about the world similarly to human common sense, which mean to develop at the level of AGI, it is important to guide machines in understanding context, making conscious decisions, and deducing consequences from various situations. To achieve this, companies need to consistently integrate vast amounts of data, conduct intricate computations, and enhance machine learning algorithms. Not only is this process time-consuming, but it also requires substantial financial investment.
To provide examples, in scenarios such as driving, it is essential for humans to apply common sense knowledge by:
· Coming to a stop at a red traffic light.
· Yielding the right of way to pedestrians at crosswalks.
· Signaling turns to indicate changes in direction.
· Reducing speed in adverse weather conditions like heavy rain, snow, or fog for safety.
· Maintaining a safe following distance to prevent rear-end collisions.
· Refraining from passing a school bus when its stop sign is extended.
· Adhering to speed limits and adjusting speed according to varying road conditions.
· Pulling over for emergency vehicles displaying lights and sounding sirens.
To replicate human common sense in the context of autonomous driving, it’s crucial not only to teach machines to follow rules and recognize static objects like traffic signs but also to understand dynamic elements such as the behavior of other vehicles, pedestrians, and cyclists. These dynamic elements pose a significant challenge for machines to adapt to unexpected situations.
Consider an autonomous vehicle that encounters a rare and unexpected scenario on the road; this scenario needs to be input into the system. Machine common sense in driving should enable vehicles to adapt to dynamic situations in real-time, such as predicting the behavior of a pedestrian who suddenly steps off the sidewalk. Machines must continuously process a vast amount of real-time data and make quick decisions, mirroring human adaptability.
In some situations, common sense might involve ethical decisions, such as choosing between safety and efficiency. Determining when it’s appropriate to prioritize safety can be a difficult challenge for machines to emulate without human-like ethical and moral judgment.
Below are situations that require machines to possess common sense knowledge, and they continue to pose ongoing challenges for AI researchers:
· A pedestrian suddenly darting across the road without obeying pedestrian signals.
· Drivers using eye contact or hand signals for communication in complex traffic situations.
· Dealing with extreme weather conditions such as flash floods or blizzards.
· Navigating unexpected obstructions like fallen trees or wildlife on the road.
· Managing a sudden brake failure that inevitably leads to a collision with either a group of pedestrians or a school bus full of children.
· Adhering to unique cultural norms, such as in the Netherlands where using the horn is reserved for genuine emergencies, and drivers often use lights instead of honking at night.
· Navigating through unfamiliar, remote, or unpaved roads.
Automation Levels of Autonomous Cars
Level 0: There are no autonomous features.
Level 1: These cars can handle one task at a time, like automatic braking
Level 2: These cars would have at least two automated functions.
Level 3: These cars handle “dynamic driving tasks” but might still need intervention.
Level 4: These cars are officially driverless in certain environments.
Level 5: These cars can operate entirely on their own without any driver presence.
Source: Sales International
As of now, the majority of autonomous vehicles in use or under development are at Level 2 or Level 3 automation. The more advanced levels (Level 4 and Level 5) with full autonomy and common sense reasoning are still in the developmental stage and not available in today consumer market.
From newborn to adulthood, humans accumulate extensive knowledge, gaining a deeper understanding of the world through their experiences. While scientists working on neural networks diligently strive to understand how to transfer this human-like comprehension to machines, contemporary AI systems are still primarily tailored for specific tasks, excelling within predefined domains by efficiently processing data and making decisions. However, the absence of common sense may limit their ability to adapt to new situations or comprehend nuances as humans do.
Whether AI qualifies as “intelligence” depends on the specific criteria used to define intelligence. Researchers and engineers primarily focus on developing AI for particular purposes, achieving notable progress in these specialized areas, with less emphasis on AI that requires the need for common sense reasoning.
