The future of artificial intelligence: general artificial intelligence

To gain a true understanding of artificial intelligence, researchers should turn their attention to developing a basic, potential AGI technology that can replicate human responses to the environment. understand.

Industry giants like Google, Microsoft, and Facebook, research labs like Elon Musk’s OpenAI, and even platforms like SingularityNET are all betting on artificial general intelligence (AGI)—intelligent agents that understand or learn anything that humans cannot do. task capabilities, which represents the future of artificial intelligence technology.

Somewhat surprisingly, however, none of these companies are focused on developing a basic, underlying AGI technology that replicates human contextual understanding. This may explain why the research being conducted by these companies relies entirely on intelligent models that have varying degrees of specificity and rely on today's artificial intelligence algorithms.

Unfortunately, this reliance means that AI can only exhibit intelligence at best. No matter how impressive their abilities are, they still follow a predetermined script that includes many variables. Therefore, even large, highly complex programs such as GPT3 or Watson can only demonstrate comprehension. In fact, they do not understand that words and images represent physical things that exist and interact with each other in the physical universe. The concept of time or the idea of ​​cause having an effect is completely foreign to them.

This is not to take away the capabilities of today’s artificial intelligence. Google, for example, is able to search vast amounts of information incredibly quickly to deliver the results the user wants (at least most of the time). Personal assistants like Siri can make restaurant reservations, find and read emails, and give instructions in real time. This list is constantly expanding and improving.

But no matter how complex these programs are, they still look for input and respond with specific outputs that are entirely dependent on their core data sets. If not convinced, ask a customer service bot an "unplanned" question and the bot may generate a meaningless response or no response at all.​

In short, Google, Siri, or any other current example of AI lacks true, common-sense understanding, which will ultimately prevent them from moving towards Artificial General Intelligence. The reason goes back to the dominant assumption underlying most AI developments over the past 50 years, which is that if hard problems can be solved, easy intelligence problems will be solved. This hypothesis can be described as Moravec's Paradox, which holds that it would be relatively easy to get computers to perform at an adult level on intelligence tests, but give them the perception and action abilities of a one-year-old baby The skills are difficult.

Artificial intelligence researchers are also wrong in their assumption that if enough narrow AI applications are built, they will eventually grow together into general intelligence. Unlike the way children can effortlessly integrate vision, language and other senses, narrow AI applications cannot store information in a general way, allowing the information to be shared and subsequently used by other AI applications.

Finally, researchers mistakenly believe that if a large enough machine learning system and sufficient computer power can be built, it will spontaneously exhibit general intelligence. This also proved to be wrong. Just as expert systems trying to capture domain-specific knowledge cannot create enough case and example data to overcome an underlying lack of understanding, AI systems cannot handle “unplanned” requests, no matter their size.

General Artificial Intelligence Basics

To achieve true AI understanding, researchers should turn their attention to developing a basic, underlying AGI technology that replicates human understanding of context. understand. Consider, for example, the situational awareness and situational understanding a 3-year-old displays while playing with blocks. 3-year-olds understand that blocks exist in a three-dimensional world, have physical properties such as weight, shape, and color, and will fall if stacked too high. Children also understand the concepts of cause and effect and the passage of time, as blocks cannot be knocked down before they are stacked first.

A 3-year-old can also become a 4-year-old, then a 5-year-old, then a 10-year-old, and so on. Simply put, 3-year-olds are born with abilities that include the ability to grow into fully functional, generally intelligent adults. Such growth is impossible with today’s artificial intelligence. No matter how sophisticated it is, today's artificial intelligence remains completely unaware of its existence in its environment. It does not know that actions taken now will affect future actions.

While it is unrealistic to think that an artificial intelligence system that has never experienced anything outside of its own training data can understand the concepts of the real world, adding mobile sensory pods to artificial intelligence can allow artificial entities to escape from reality. Learn in the environment and demonstrate a basic understanding of physical objects, cause and effect, and the passage of time in reality. Like that 3-year-old, this artificial entity equipped with sensory pods is able to directly learn how to stack blocks, move objects, perform a sequence of actions over time, and learn from the consequences of those actions.

Through sight, hearing, touch, manipulators, and more, artificial entities can learn to understand in ways that are simply not possible with text-only or image-only systems. As mentioned before, such systems simply cannot understand and learn, no matter how large and varied their data sets are. Once an entity acquires this ability to understand and learn, it may even be possible to remove the sensory pods.

Although at this point we cannot quantify how much data is needed to represent true understanding, we can speculate that there must be a reasonable ratio in the brain related to understanding. After all, humans interpret everything in the context of everything they have already experienced and learned. As adults, we interpret everything in terms of what we learned in the first few years of life. With this in mind, it seems that true artificial general intelligence will only be possible if the AI ​​community recognizes this fact and takes the necessary steps to establish a basic foundation of understanding.

The above is the detailed content of The future of artificial intelligence: general artificial intelligence. For more information, please follow other related articles on the PHP Chinese website!