A research team from the Graduate School of Informatics at Nagoya University has taken a big step towards creating a neural network with metamomory through computerized evolutionary experiments.
In recent years, rapid advances have been made in the design of artificial intelligence technology using neural networks that mimic brain circuits. One of the goals of this field of research is to understand the evolution of metamemory so that it can be used to create an artificial intelligence with a human-like mind.
Metamorphosis is the process by which we ask ourselves if we remember what we ate for dinner last night and then use that memory to decide if we will eat something different tonight. While this may seem like a simple question, answering it involves a complex process. Metamemory is important because it allows a person to know their own memory capacity and adjust their behavior accordingly.
“In order to explain the evolutionary basis of the human mind and consciousness, understanding metamory is important,” says lead author Professor Takaya Arita. “A truly human AI, with which a person can communicate and enjoy like a family member in the home, an AI that has a certain amount of metamorphosis, because it has the ability to remember what it has already heard or heard.
When studying metamorphosis, researchers often use a “sample delayed matching task”. In humans, participants in this task see an object, such as a red circle, remember it, and then participate in an experiment that selects some of the objects they have seen before. Reward the correct answer and punish the wrong answer. However, the subject may choose not to take the test and still earn a small prize.
A person performing this task can naturally determine using their metamemory whether they have seen or remembered the object. If they remember, they will test to get big rewards, and if they are not sure, they will avoid the risk of punishment and get small rewards instead. Previous research has shown that monkeys can do just that.
A team from the University of Nagoya, made up of Takaya Arita, a professor at the Graduate School of Informatics, Yusuke Yamato, and Rezi Suzuki, has created an artificial neural network model that performs a sample-delayed matching task and analyzes its behavior.
Although it originated from random neural networks that did not even have memory functions, the model was able to scale to the point of working like monkeys in previous studies. The neural network can examine its memories, store them, and isolate the outputs. Intelligence was able to do this without the need for any assistance or intervention from researchers, suggesting appreciation of its metamorphosis mechanism. The need for metamemory depends on the user’s environment. Therefore, it is important for artificial intelligence to have a metamemory that adapts to the environment in which it is learned and developed, “said Arita, professor of discovery. “The key is that artificial intelligence learns and develops to create a meta-memory that adapts to its environment.”
Creating an adaptive intelligence with metamemory is a big step towards building machines that have memories like ours. The team is excited about the future, “this achievement will provide a source of perception of artificial intelligence with a ‘human-like mind’ and even consciousness.” A
The results of the study were published in the online edition of the International Scientific Journal Scientific report. The study was partially funded by a JSPS / MEXT grant for Kakenhi scientific research (JP17H06383 in # 4903).
Source of the story
Materials provided by University of Nagoya. Note: Content can be edited for style and length.