Google announced a new AI architecture, which has profoundly expanded Google’s AI computing capabilities. The new artificial intelligence architecture is a single model that can be trained to do millions of things, and Google calls it closer to the mammalian brain.
Google Path Artificial Intelligence
Google’s announcement stated that the current state of artificial intelligence is to train machines to do one thing well, such as recognizing images or understanding animal sounds.
Therefore, in order to understand vision and sound, two different AI models are needed to accomplish these two tasks.
advertise
Keep reading below
The new AI architecture takes a different path, training it to generalize by learning skills that can be applied to different tasks.
Google explains how it works:
“…We want to train a model that can not only handle many individual tasks, but also use and combine existing skills to learn new tasks faster and more effectively.
In this way, what the model learns by training on one task (for example, learning how aerial images predict the altitude of a landscape) can help it learn another task-for example, predicting how floods will flow through the terrain.
We hope that the model has different functions, can be called as needed, and stitched together to perform new and more complex tasks-closer to the way the mammalian brain generalizes across tasks. “
advertise
Keep reading below
Artificial intelligence that can train multiple senses
Current AI models train for a single sense organ, such as sight or hearing, but cannot train both at the same time. Google said that such a model will be able to understand a concept in all relevant senses.
With Pathways, Google can simultaneously understand text, images, and speech in a single AI model.
Pathways works like a human brain
The human brain uses only a small amount of its processing power to complete tasks, using specialized parts instead of the entire brain network.
Google’s new Pathways AI will accomplish tasks in a similar way, which will improve energy efficiency and be able to learn and complete all tasks faster than the old model.
“There are nearly 100 billion neurons in your brain, but you rely on a small part of them to explain this sentence.
Artificial intelligence can work in the same way.
We can build a single model of “sparse” activation, which means that only small paths through the network will be called as needed. “
Ways AI will solve multiple problems
The purpose of Pathways is bigger than search. Google said it can adapt to solve many problems, including problems we haven’t encountered yet.
“…We are building a next-generation artificial intelligence system that can quickly adapt to new needs and solve new problems that have arisen around the world, and help mankind make full use of the future before us.”
advertise
Keep reading below
Google Pathways AI architecture
Google’s blog post did not state whether Pathways has been activated in any system. The announcement stated that Google now has the ability to create these systems, and they will build them.
Google also said that their team is working on designing a “next generation” artificial intelligence architecture.
Google made it clear that they are “creating” this new artificial intelligence architecture:
“This is why we established Pathways. Pathways will enable a single artificial intelligence system to generalize thousands or millions of tasks, understand different types of data, and do this with extraordinary efficiency…”
advertise
Keep reading below
Google has been publishing research papers on machine learning techniques that can learn generalization, one of which is called Fruit tart.
FLAN can train a machine to learn how to complete a task and apply what it learns to other tasks, very similar to the content described in the article on Pathways.
If this new Pathways AI architecture has not yet appeared in some form, it may be very close to realization.
Citation
Read Google’s announcement on the new AI architecture
Introduction to Pathways: Next Generation Artificial Intelligence Architecture
Read the new Google research paper on FLAN
Google researches and develops better machine learning
advertise
Keep reading below
