The development of artificial intelligence gaming methods
The complexity of the human brain is impressive in itself, but if we assume that the body that defines a person has reached this peak through generations of development, this fact is even more stunning. Neuroscientists, through new types of research, will learn more about the brain every day. In particular, at the Medical University of Wisconsin-Madison, researchers study how neural networks are developed, teaching computers to play video games.
The team describes how they created simple systems called “animats” and observed their changes through the developmental generations in the article “Evolution of Integrated Causal Structures Animats Exposed to Environments of Increasing Complexity” growing complexity ”). Using a process not unlike natural development, a team of researchers studied how artificial intelligence becomes more complex, in the hope that the data could better explain the complexity of our brain.
Each "animat" possessed two sensors, two engines and four internal computers to coordinate memory, movement and sensations. For these basic nervous systems, a task with a puzzle game was defined, where the goal is to catch the blocks falling from the top of the screen. Thus, each "animat" would have to learn how to determine where the block is going to land in order to successfully complete the action. In view of the further development of analogies, the coding of the "animat" could also randomly modify and make some systems better than others. In this case, the researchers copied these successful systems and passed on to achieve the desired result. Digital natural selection, so to speak.
')
The team also increased the complexity of the games to match the growing complexity of the "animats" and to provoke their further development. After more than 60,000 generations of systems, not only did the descendants adapt to more complex tasks, they themselves became more complex. Their once elementary neural networks have now become much more complex, creating more connections to make up for the limited size of the brain. And this experience is not far from the result that today is our most complicated human brain.
“Animats” and their intellect do not become larger, but they integrate more for better functionality and efficiency. The team says that their experiment demonstrates how and why the process called “Integrated Information Theory” works. In fact, Mr. Tonoi himself suggested that this theory is based on the nature of conscious organisms to adapt to mental problems in the same way as physical problems - by changing themselves.
also for those interested not in translating the overview, the original work of Mr. Tononi and his team.
And the translation for you was prepared by the developers from the VirCities project - sign up for the alpha test!
The team describes how they created simple systems called “animats” and observed their changes through the developmental generations in the article “Evolution of Integrated Causal Structures Animats Exposed to Environments of Increasing Complexity” growing complexity ”). Using a process not unlike natural development, a team of researchers studied how artificial intelligence becomes more complex, in the hope that the data could better explain the complexity of our brain.
Each "animat" possessed two sensors, two engines and four internal computers to coordinate memory, movement and sensations. For these basic nervous systems, a task with a puzzle game was defined, where the goal is to catch the blocks falling from the top of the screen. Thus, each "animat" would have to learn how to determine where the block is going to land in order to successfully complete the action. In view of the further development of analogies, the coding of the "animat" could also randomly modify and make some systems better than others. In this case, the researchers copied these successful systems and passed on to achieve the desired result. Digital natural selection, so to speak.
')
The team also increased the complexity of the games to match the growing complexity of the "animats" and to provoke their further development. After more than 60,000 generations of systems, not only did the descendants adapt to more complex tasks, they themselves became more complex. Their once elementary neural networks have now become much more complex, creating more connections to make up for the limited size of the brain. And this experience is not far from the result that today is our most complicated human brain.
“Animats” and their intellect do not become larger, but they integrate more for better functionality and efficiency. The team says that their experiment demonstrates how and why the process called “Integrated Information Theory” works. In fact, Mr. Tonoi himself suggested that this theory is based on the nature of conscious organisms to adapt to mental problems in the same way as physical problems - by changing themselves.
also for those interested not in translating the overview, the original work of Mr. Tononi and his team.
And the translation for you was prepared by the developers from the VirCities project - sign up for the alpha test!
Source: https://habr.com/ru/post/365183/
All Articles