The epiphany of the Father Ai about how the AI will be able to understand the world22 октября, 2021
· You can teach AI to understand the world around the world as people understand him. The key to this will be the technique of perception of the world, similar to the human.
· Human perception is built on intuition, and so that the AI understood the world, you need to simulate intuition for AI.
· Intuition is the ability to easily conduct analogies. With its help, AI, like the human brain, will understand the world and possess insight.
· Since childhood and throughout our life, we comprehend the world using reasoning by analogy, reflecting the similarity of one object (ideas, concepts …) with another — or, in the GLOM terminology, is the similarity of one large vector with another.
· Modern theories either come from the fact that in the course of perception, the brain will gain invention («pixels»), or from the fact that the brain processes characters. GLOM argues that OA approaches are incorrect: the brain operates not by pixels and not symbol, but by large vector neural productivity (ie, these or other neurons are labeled).
· If a GLOM is based on creating a new class of models and deep learning algorithms, this can be a breakthrough to AI, able to flexibly solve problems. Such a AI will be able to understand the things that have never come across before, to extract similarities from past experience, experimenting with ideas, to generalize, extrapolate — in one word, to understand.
GLOM is a new hyper-breaking theory of AI developed by
Of all researchers, Hinton are quoted more often than three walking behind it, combined. His students and graduate students go to work in the AI Apple, Facebook and OpenAi laboratory; Hinton himself is a leading scientist in the Google Brain AI team. Virtually any achievement in the field of AI over the past ten years — translated, recognizing speech, recognizing images and games — one way or another concerns the works of Hinton.
His new GLOM theory solves the two most complex problems for visual perception systems (and the other modern AI cannot in principle):
And although Glom focuses on visual perception, Hinton expects the same ideas to be applied to the language to reproduce the syntactic analysis tree in the neural network (as long as this is an unrealistic dream).
The key GLOM ideas were recently published by Hinton in the 44th page article. These ideas are a significant revision of the architecture of capsule neural networks, invented by Hinton 4 years ago.
And although GLOM today is no more than the «new philosophy of neural networks», but this philosophy of charming and honeycomb sounds. For it promises to open before AI almost unlimited horizons.
Thanks for watching! Put husky I.