Deep Reasoning: Is this the Next Era of AI?
Deep reasoning is said to bring us all one step closer to artificial general intelligence. It is said to be the next level beyond deep learning.
Artificial intelligence (AI) is omnipresent.
Everything that you hear and see today about AI is due to deep learning.
- Autonomous vehicles
- Language recognition
- Text generation
- Computer vision
- Deep learning robots
- Adding colors to black and white images
- Customer experience
Thanks to this new category of algorithms that has proved its power of mimicking human skills just by learning through examples. Deep learning is a technology representing the next era of machine learning. Algorithms used in machine learning are created by programmers and they hold the responsibility for learning through data. Decisions are made based on such data.
Some of the AI experts say, there will a shift in AI trends. For instance, the late 1990s and early 2000s saw the rise of machine learning. Neural networks gained its popularity in the early 2010s, and growth in reinforcement came into light recently.
Well, these are just a couple of caveats we’re experienced throughout the past years.
AI is a technology that aims at increasing efficiency and productivity for organizations that are using them.
But can AI solve or answer questions like — “what is the size of that ball which is placed at the left side of the cupboard down on the floor?”
Pretty much doubt it. A six-year-old kid could perhaps answer this simple question by just looking at the image with these elements in it. And they can’t be answered out of a deep learning method.
The crux: you need more than just deep learning
Why? Because deep learning models are only great in terms of understanding relationships between inputs and outputs, and that’s just about it. Be it reinforcement learning or supervised learning, the input and the expected output have been defined so that the model understands. This works great when it comes to placing tasks like classification, but if we wish for AI models to make decisions based on the theory called common sense i.e. abstract reasoning. Now for this, you might need to enable them with reasoning power.
This is called “deep reasoning.”
What is deep reasoning?
Deep reasoning is a field that goes far beyond deep learning. It helps machines understand implied relationships between different things. So, for example, if you say — all the animals drink water, and elephants are animals. In this sentence, the implicit relationship explains that elephants drink water, but it hasn’t been explicitly explained.
Surprisingly, humans are good with this kind of relational reasoning power — how things relate with each other, etc. But when it comes to enabling the computers, it is a tedious task.
Is there a way to explicitly offer computers the ability to reason?
According to DeepMind researchers, they were able to conclude how they made a deep learning model answer questions like the one asked about the animals. As a result, the answers given were 96 percent accurate.
However, they had to make use of three networks.
1. Convolutional neural network (CNN) to process the image: these are exceptional at identifying features present in an image. One of the reasons why researchers used CNN to extract different objects from an image.
2. Long-Short-Term-Memory (LSTM) network to process the question: they have good memory power thus they understand sequences. This is perfect when you need to deal with questions and language. Ideally, understanding a language at the beginning of the sentence can impact the meaning of the end of the sentence as well.
3. Relation Network (RN) for a better understanding of how different objects relate with each other: when this model processes the question and the image, it gets better in understanding the relationship between the object and the image.
Deep reasoning is perfect when trying to explain AI the abstract relationship between multiple things.
