One day in the future, you may have the ability to create a virtual human being or machine using neural networks, the technology that powers Google’s artificial intelligence technology.
If you can learn to use neural networks to create such a thing, you’ll be able to create and control an army of them, one that could potentially be a real threat to human society.
That’s exactly what Google’s DeepMind, a British-based startup, hopes to do.
“We want to give a very big step in the direction of artificial intelligence,” DeepMind co-founder Demis Hassabis told me.
“I think we’re going to have the first generation of intelligent robots within five years.”
What’s an intelligent robot?
In the coming years, the most common use of AI for humans is as an agent for the development of new products and services.
For example, Google and Facebook have begun working on an AI that can automatically identify and prioritize items in a shopping cart.
For a variety of applications, AI can be used to automatically generate and customize new products or services, and it can even perform some of the most important tasks in the world: taking photos and videos, for example.
But what about when AI is used to create or control a living thing?
There are several kinds of “minds” out there, and a number of them have been working together for years to make AI more capable of producing useful results.
DeepMind and others have created machines that can create and process images, perform tasks like translation and speech recognition, and even play video games.
These machines have also been able to make predictions about the future.
A few years ago, Google released a machine learning model called DeepDream, which can “dream” by taking pictures of a scene, recording a series of frames, and then trying to predict what might happen next.
This was a promising start, but it was limited to just three images.
“You can’t really get a good picture of what’s going to happen in the next 10 minutes,” Hassabis said.
“It’s all very theoretical.”
But a number recent projects are trying to build on DeepDream and build on it, creating new kinds of machines that mimic the kinds of mental processes humans use.
These “mind machines” use algorithms to learn from images, video, or other experiences to learn more about how to perform the same tasks.
The latest model, called DeepMind Vision, is a model that can take pictures, and combine them with a neural network to create images that can then be used for video or audio analysis.
These models are able to do much more than simply learn from pictures.
They can also use data from those images to create the kinds, or the features, of objects, people, or events that make the images appealing to a human viewer.
And because the neural networks can learn from experience, it’s possible for a DeepMind machine to learn to be more “emotional” in its actions.
“A lot of AI is going to be about this very kind of deep learning,” Hassabs said.
A DeepMind model that uses images to build an emotional image.
Image: DeepMind/YouTube “The emotional aspects of a lot of our jobs, they have to be able take in what’s happening in the moment.
They have to have some sense of empathy.”
The current models that are building AI for human tasks use a kind of machine learning called deep reinforcement learning.
This means that the model tries to learn by looking at examples of what a person does, rather than trying to solve problems that require human judgment.
For instance, it might build a model to tell the difference between a dog running a race and a human playing with a toy.
This is a very human task that requires human judgment and, in theory, would be impossible to train a neural net to do better than that.
But DeepMind has a way of training its neural nets to be better than humans at it.
The model that builds DeepMind’s DeepDream Vision has a neural system called a “supervised learning” system that uses a neural graph to train its neural network.
A graph is a kind a computer program can take input data and make predictions based on how the data looks like in a network.
This kind of neural graph allows the neural network a way to learn and then apply the training data to a real-world task.
For this type of learning, a deep learning system looks at the graph it’s training on and then learns the rules for how to apply the learning to its own neural network, which in turn then can apply those rules to the next training data.
The network will be able tell what the next picture looks like based on its previous training data, and if it gets the right result, it can then apply that training to make the next image better.
In the past, this kind of learning was a big problem for neural networks.
Because deep learning is a relatively new field, it is still new