Search This Blog

Why Google Added Neuroscience To DeepMind AI Platform

Google has added a new capability to DeepMind that enables its artificial intelligence (AI) platform to take a nap at a given point of time. DeepMind explained in a blog post that the new development in its AI effort incorporates ideas from neuroscience research, a move that significantly complements the typical logic-based and theoretical mathematical approaches to AI. 
Why Google Added Neuroscience To DeepMind AI Platform

Neuroscience does this through the identification of biological computation that DeepMind believes could have a substantial contribution to helping the platform learn from its environment in a more human-like fashion. DeepMind, therefore, intends for the addition of neuroscience into its AI work to check whether existing AI techniques are functioning as they should and to produce new kinds of algorithms that can be used to create AI-based systems.

DeepMind mentioned a new finding in neuroscience research stating that the human brain is capable of revisiting its earlier neuronal activities during sleep. That means the biological brain can still store and process data from an earlier experience, failed or successful, even if it is resting and consequently make successes in the future. That inspired DeepMind to make its AI platform learn how to sleep. The company argued that while it might appear as though the concept of sleep is illogical in the context of AI as a tool to solve computational problems, the idea is to develop an algorithm that can store a host of experimental data and review those pieces of information while at rest in order to study where it flopped and succeeded in the past. In other words, even if an AI-based computer previously failed to solve a computational challenge, it can still handle training data while in offline mode and use that same information to accomplish the task in the future. The concept forms part of DeepMind’s deep-Q network algorithm that uses raw pixels and score data to learn a wide variety of Atari 2600 games.

DeepMind said that some of the challenges that AI can solve in the future with the help of neuroscience include imagination, a concept that the company introduced into two kinds of AI that it developed recently with the goal of helping to plan for the future by learning to separate useful data from irrelevant information. An update on DeepMind’s growing AI efforts will likely follow in the coming months. 

Source: DeepMind Via: The Next Web


Recent Posts Widget