Interpretation of the development status of artificial intelligence machines: the need to overcome the boundaries between the four types of AI in the future

According to the American Life Science Network, in the field of artificial intelligence (AI) research, the consensus of the industry on the latest breakthrough is that the distance between intelligent machine distances with emotional awareness is still far away. Machines have done better than humans in understanding voice commands, distinguishing images, driving cars, and playing games. So, how long does it take for the machine to walk naturally among humans?

The latest AI report released by the White House is skeptical about this dream. According to the report, in the next 20 years, we may not be able to see that machines have a wide range of applied intellectual abilities or transcend human intelligence. But in the future, in the implementation of more and more tasks, the performance of the machine will reach or exceed the human level.

As an AI researcher, I am very happy that my research has become one of the most important areas of the US government. But the White House report is almost always focused on what I call "the most boring AI" and refuses to accept AI into different kinds of opinions. At the same time, the report does not explain how AI evolution can help develop a constantly improving AI system, how computational models can help us understand how human intelligence evolves.

The White House AI report highlights tools known as "mainstream AI," including machine learning and deep learning. These techniques can play the intellectual game "Jeopardy", defeating the human Go master in the most complicated game in history. Today's intelligent systems do handle massive amounts of data and quickly make complex calculations. But they lack an important element that will be the key to our future creation of emotional machines.

In addition to teaching machine learning, we have more work to do. We need to overcome the boundaries between the four different types of AI that separate the AI ​​from us.

Most basic AI systems are purely responsive machines that do not have the ability to form memories or the ability to draw on past experience to help make current decisions. IBM's supercomputer "Deep Blue" defeated human chess master Garry Kasparov in the late 1990s, and it is the perfect representation of a responsive machine.

"Dark Blue" recognizes the pieces on the board and knows how to move the pieces. It can even predict the next move for itself and its opponents, and choose the best solution among many possibilities. However, "dark blue" has no concept of "past" and no previous memory. In addition to occasionally using the specific rules of chess, such as against repeating the same steps three times, "Deep Blue" almost ignores what happened before. Its focus is on the pieces on the current board and making choices for the next move.

This kind of intelligence involves the world directly perceived by the computer and acts where it can be seen. It does not depend on the internal concepts of the world. In groundbreaking papers, AI research expert Rodney Brooks believes that we should only make such machines. His main reason is that people are not good at programming the exact world of simulation used by computers.

The smart machines we are currently developing either do not have the concept of such a world at all, or only the intelligent machines that perform specific tasks have very limited concepts. The design innovations of “Deep Blue” did not expand the range of possible steps that computers considered. Instead, researchers have found ways to narrow their horizons to prevent them from pursuing certain potential future steps, depending on how they evaluate the results. Without this ability, "dark blue" may require a more powerful computer to defeat Kasparov.

Similarly, Google's AlphaGo has defeated the master of human Go, but it can't assess all the potential steps in the future. Its analysis is more complicated than "dark blue", using neural networks to assess game progress. These methods do improve the AI ​​system's ability to perform better when playing a particular game. But they cannot be easily changed or applied to other situations. These computerized imaginations do not have the concept of a broad world, which means that they cannot surpass the functional limitations of specific tasks and are therefore easily fooled.

In addition, they cannot participate in the world in an interactive way, which is why we think that the future AI system has the most important capabilities. Instead, each time these machines encounter the same situation, they execute the same strategy pre-programmed. This may be good news in ensuring the credibility of the AI ​​system, such as the fact that you want your driverless car to be a reliable driver. But if we want the machine to really engage in interaction and make a real response to the world, this can be terrible. These simplest AI systems never feel bored, interested or sad.

This type of AI can briefly review past experience, and driverless cars can already do this. For example, they look at the speed and direction of travel of other cars. They are not able to complete this action right now, but need to identify specific targets and monitor them for a while.

These observations are added to the pre-programmed simulation world of driverless cars, including lane markings, traffic lights, and other important elements such as highway curves. It also includes when the car decides to change lanes to avoid being hit by a nearby car.

But these simple information about the past is only short-lived and will not be stored in the experience library that the car has already learned. The experience library is equivalent to the driving experience accumulated by human drivers. So how do we build an AI system that can fully simulate the world, remember their experiences, and learn how to handle new situations? Brooks believes that it is difficult to do. My research method draws inspiration from Darwin's theory of evolution, and it is possible to build a simulated world by machine itself to make up for the shortcomings of human beings.

This may be an important gap between the AI ​​machines we have built and the AI ​​machines we will build in the future. Future AI machines will be more advanced, they will not only build the simulation world themselves, but also simulate other objects and entities in the world. In the real world, this is called "the theory of mind", that is, understanding human beings and creatures in the world have thoughts and emotions, and these thoughts and emotions can affect their behavior.

This is crucial for human beings to form society because they allow us to communicate in society. If you don't understand each other's motivations and intentions, and don't consider other people's understanding of yourself or the environment, the best case is that collaboration is very difficult, and in the worst case, there is no possibility of collaboration.

If AI systems want to be in the middle of humanity, they must be able to understand that each of us has different ideas and feelings and expects to be treated ourselves. To do so, they must adjust their behavior accordingly.

The final stage of AI development is to build systems that can form the simulated world that represents them. In the end, our AI experts not only need to understand their own consciousness, but also build a machine with self-awareness. In this sense, the fourth type of AI is an extension of the "mental theory" machine represented by the third type of AI, and it is also the reason why consciousness is called "self-consciousness."

A conscious existence can be aware of themselves, understand their internal state, and be able to predict the feelings of others. We assume that someone screaming at the back represents anger and impatience, because we also feel the same when we do that. Without the theory of mind, we cannot make these inferences.

It may be a long way to create a self-aware machine, and we should focus on building AI capabilities that understand memory, learning, and making decisions based on past experience. This is an important step in understanding human wisdom. This is also important if we want to design or develop machines that go beyond these categories.

Monocrystalline Solar Panel

Using latest technology and expertise knowledge, we are indulged in providing Monocrystalline Solar Panel. Our Monocrystalline Solar Panel is highly appreciated amongst clients for its excellent features such as reliable operation and dimensional accuracy. Clients can buy this product from us at industrial leading price.

Monocrystalline Solar Panel,Monocrystalline Silicon Solar Panel,High Transmission Rate Solar Panel,Photovoltaic Cell Solar Panel

Delight Eco Energy Supplies Co., Ltd. , https://www.cndelight.com