Types of Narrow AI and why we need it?

Different types of narrow AI technologies

 

Symbolic artificial intelligence

Symbolic AI is also known as good-old fashioned AI (GOFAI), was dominant area of research for most AI’s history. Symbolic AI is the explicit embedding of human knowledge and behaviour combined with computational programs.

Symbols are things we used to represent something that we can remember for. Many concepts and tools you find in CS are the result of these efforts. Therefore, the symbols they used plays crucial role in AI.

An example of Symbolic AI tools is object-oriented programming which allows you to define classes, strings, Boolean etc. The best example of OOP programming is ‘Python’.

 

Machine Learning

The another type of narrow AI is machine learning. A developer or a ML engineer creates a model and then ‘Trains’ them by feeding with some example or data so that it could predict the outcome around that example or data. It creates also some algorithms and perform prediction which can be represent mathematically and graphically.

For instance, a ML algorithm trained on thousands of bank transaction with their outcome(legitimate or fraudulent) will be able to predict the new bank transaction that whether they are fraudulent or not.

It comes in many different levels and variety. They are:-

Deep Learning

Deep Learning is a subset of machine learning, an AI which can change its way by recognizing the behaviour of software has developed.

Today, Deep Learning has become a very useful and pivotal application which we use every day such as  content recommendation systems, translation app, image and facial recognition systems, chatbots and many more. It has also advanced in some special domain like healthcare, education and self-driving cars.

 

Natural Language Processing

Natural Language Processing and Natural Language Generation has removed many of the barriers between the interaction of human and a machine. Not only it is interacting and understanding but also creating new opportunities and accomplish tasks that were impossible.

It uses ML and DL algorithms to analyse the human behaviour and voice in a smart way. It does has any predefined rules instead it learns from the examples and data of how human talk and behave. It is train to recognize the thousands of text samples, words, sentences, paragraphs etc. which are labelled by human.

For example, based on the context of conversation it can determine if the word “cloud” is reference to cloud computing and also the mass of condensed water vapour in the sky.

Reinforcement Learning

Reinforcement Learning is also a subset of machine learning, which is a classical approach to creating AI required programmes to manually code every rule that defined the behaviour of the software.

The best example of RL is Stockfish which is an open-source AI chess engine that has been developed with contribution from the hundreds of programmers and chess experts who have turned their experience into the game rules. The AI model will peruse the data and find the similarities between the moves made by the chess winners and experts and will predict the upcoming moves.

Computer Vision

Computer Vision is the field of the computer science that focuses on replication how the computers can gain the high-level understanding the complexity of human vision systems and enabling computers to identify the objects from the digital images or videos in the same way human do. In other words, it understands and automate tasks that human visual system can do.

It has been to a great leap in recent years and has able to surpass the human in some tasks, thanks to the artificial intelligence and innovations in deep learning and neural networks. It is used in understanding of the surroundings of  self-driving cars via cameras that feeds data to the software of CV.

 

Why Narrow AI?

Narrow AI is good and can easily perform the single-handed tasks. They can outperform human in some basic tasks but not all of them. The tasks that it can perform has a limited range.

Both Symbolic AI and machine learning learns from the parts of human intelligence, but they fall short of bringing together the necessary pieces to create an all-rounder human-like AI. This prevents the outcome of moving AI beyond narrow artificial intelligence.

Manipulation symbols in human thinking process plays a big part. But the human mind does more things than manipulating symbols. There are skills that we acquire in our childhood such as walking, running, eating, tying shoelaces, brushing teeth etc. are things we learn subconsciously and without doing any kind of symbol manipulation in our minds.

Symbolic AI require precise and minute information on every task it must accomplish and it can only function according to the defined rules.

On the other hand, machine learning algorithms are good at replicating the kind of behaviour of human that can’t be done with the symbolic reasoning. Recognition of faces and voices are the kind which comes under deep neural networks, can be recognized as extraordinary in this period if we compare it with previous technology.

But again, the mind’s learning process cannot be simplified into pure pattern-matching. For example, we recognize an image of cat because we’ve seen it in our lives. Now, the machines to recognize the image can be difficult as it involves a lot of symbolic manipulation like it has four legs, a tail, furry body etc.

The lack of symbol manipulation can lead to poor AI results. It limits the power of deep learning and other machine learning algorithms. CNNs (Convolutional Neural Networks) are used in computer vision needs to be trained on thousands of images of each type of object they must recognize, but sometimes it often fails to recognize when they encounter the same objects under new lighting or from different angle.

Machine learning are also strictly bound to the context of their training examples, which is why they are narrow AI. For example, Game-playing AI systems such as AlphaGo, OpenAI five must be trained on tens of millions of matches or thousands of hours’ worth of gameplay before they can master their respective games. These are games more than a human can play in their lifetime.