Yan LeKun proposed to endow AI systems with 'common sense' | Crypto_NFT
Yan LeKun proposed to endow AI systems with "common sense"
“Autonomous” artificial intelligence, able to explore and perceive the world around them like people, needs more effective training. This conclusion was made by the vice president of Meta and the leading AI specialist of the company Jan LeKun, writes Motherboard.
Compared to humans and animals, modern artificial intelligence systems lack intelligence. This is the concept needed to develop "autonomous" AI algorithms that can learn on the fly by observing the real world, rather than through lengthy training to perform just one specific task.
LeCun published a study where he proposed a way to solve this problem. In his opinion, algorithms require more effective learning, since so far AI is “not very good” at planning and predicting changes in the real world. On the other hand, humans and animals are able to gain vast amounts of knowledge through observation and non-essential physical interactions.
Teenagers can learn to drive a car with dozens of hours of practice and still not get into an accident. To solve the same problem, AI systems need a huge amount of data.
“An unmanned vehicle will need to drive off a cliff several times before it realizes that this is a bad idea, and thousands of more attempts to learn how not to fall off it,” LeCun said during a presentation at the University of California at Berkeley.
According to him, the difference lies in the presence of "common sense" in humans and animals.
In the article, the scientist describes this concept as a set of models that can help a living being draw a conclusion about the difference between probable, feasible and impossible. Such a skill allows a person to explore their environment, fill in missing information, and provide new solutions to unknown problems.
LeCun stated that many modern training processes like reinforcement learning are “not up to par” when it comes to matching human reliability in real-life tasks.
“This is a practical problem because people really need machines with common sense. We need self-driving cars, home robots and intelligent virtual assistants,” he said.
The specialist proposed an architecture designed to minimize the actions that an AI must perform in order to successfully learn and complete the task.
LeCun described a model for creating "autonomous" intelligence, consisting of five separate but customizable modules. One of the hardest parts of the 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 allow using knowledge about the environment for different tasks.
Recall that in July, Google fired an engineer after he announced the "birth of the mind" in the LaMDA AI algorithm.
In January, Elon Musk promised to prevent machines from rioting against people.
In June 2021, DeepMind scientists stated that reinforcement learning is enough to achieve artificial general intelligence.
