Jan Lekun proposed to endow the AI systems with “common sense”
“Autonomous” artificial intelligence, capable of exploring and perceiving the world around the world as people, is necessary more effective training. This conclusion was reached by the Vice President of META and the host of AI-Specialist of the company Jan https://gagarin.news/news/crypto-lending-top-5-services/ Lekun, writes Motherboard.
Compared to people and animals, modern artificial intelligence systems lack the mind. This is the concept necessary for the development of “autonomous” ALGIRITMA, capable of studying “on the fly” by observing the real world, and not through long training to perform only one specific task.
Lekun published a study where he proposed a way to solve this problem. In his opinion, algorithms require more effective training, since so far “not very good” in planning and predicting changes in the real world. On the other hand, people and animals are able to get a huge amount of knowledge through observation and insignificant physical interactions.
Adolescents can learn to drive a car by practice for several tens of hours, and at the same time not to get into an accident. To solve the same problem, II systems need a huge amount of data.
“An unmanned car will need to move out of the cliff several times before he understands that this is a bad idea, and thousands of attempts to learn not to fall from it,” Lekun noted during a report at the University of California in Berkeley.
According to him, the difference is the presence of “common sense” in people and animals.
In the article, the scientist describes this concept as a set of models capable of helping a living creature to conclude that the difference between the likely, executive and impossible. Such a skill allows a person to explore his environment, make up for the missing information, as well as provide new solutions to unknown problems.
Lekun said that many modern training processes like learning with reinforcement “not on top”, when it comes to the correspondence of human reliability in real tasks.
“This is a practical problem, because people really need cars with common sense. We need unmanned cars, home robots and intellectual virtual assistants, ”he said.
The specialist proposed an architecture designed to minimize actions that must perform AI for successful training and fulfilling the task.
Lekun described the model of the creation of “autonomous” intelligence, consisting of five separate, but customized modules. One of the most difficult parts of architecture will become like a simulator and will evaluate the state of the world, as well as predict imaginary actions and other sequences. This will use the knowledge of the environment for different tasks.
Recall that in July, Google fired an engineer after he announced the “emergence of the mind” by AI-Algorithm LAMDA.
In January, Elon Musk promised to prevent a riot of cars against people.
In June 2021, DeepMind scientists said that there is enough training with reinforcement to achieve general artificial intelligence.
Subscribe to FORKLOG news at Telegram: Forklog AI – all news from the world of AI!