The ultimate goal of artificial intelligence scientists is to replicate the kind of general intelligence that humans have. And we know that people do not suffer from the problems of existing deep learning systems.
"Humans and animals seem to be able to learn a large amount of background information about the world, largely through observation, independently of the mission," Bengio, Hinton, and LeCun write in their paper. This knowledge supports Common Sense and allows people to learn complex tasks such as driving with just a few hours of practice.
Elsewhere in the paper, scientists say, humans can generalize differently and more strongly than ordinary iid generalization: we can accurately interpret new combinations of existing concepts, even if these combinations are highly unlikely under our education. Distribution, as long as they respect the high-level syntactic and semantic patterns we have already learned.
Scientists offer a variety of solutions to bridge the gap between artificial intelligence and human decency. One approach that has been widely discussed over the past few years is hybrid artificial intelligence, which combines neural networks with classical symbolic systems.Symbol manipulation is a crucial part of people's ability to reason about the world. It is also one of the biggest challenges of deep learning systems.
Bengio, Hinton, and LeCun do not believe in mixing neural networks and symbolic artificial intelligence. There are some who believe that there are problems that neural networks cannot solve, and that we should resort to the classic AI, symbolic approach, says Bengio in a video accompanying the ACM article. But our study shows otherwise.
Deep learning pioneers believe that better neural network architectures will eventually lead to all aspects of human and animal intelligence, including symbol manipulation, reasoning, Causal Inference, and common sense.