Understanding the Different Types of AI - A Comprehensive Overview!

The concept of AI has always been a fascinating one. Since its inception to date till now, the domain of computer science has undergone various changes. From basic weak AI to expert systems, there have been interpretations of AI.

However, a lot of people confuse the different technologies of AI with the different types of AI. To clear this confusion, we have written this article.

Therefore, let’s learn about the different types of artificial intelligence below…

What are the Types of AI? - Main Types of Artificial Intelligence

With the development of the different types of artificial intelligence systems and the inclusion of artificial intelligence companies, the types of AI have split into multiple sub-domains. Right now, we are utilizing several of them, however, some of them are still hypothetical in nature.

Therefore, let’s check out the types of artificial intelligence with examples (as many as possible)...

1. Reactive AI

Reactive AI is a subset of artificial intelligence or the type of AI that reacts and responds to the input it consumes from the sensors and takes actions based on it. Unlike any other smarter AI systems that are based on rule-based principles, a reactive AI simply reacts to the data fed to it. It is because it doesn’t have any memory, therefore, it doesn’t have the capability of learning from its past experiences. It only responds in the manner it has been trained for.

A common example that is often given in the realm of reactive AI is IBM’s Deep Blue. This system defeated the chess grandmaster Gary Kasparov. The team behind the development of Deep Blue doubled the number of chess chips used for enhancement. This gave the reactive AI to predict 100 to 200 million moves per move made by Gary Kasparov. This was a defining moment of AI and is often included within the history of AI development with golden words.

Below is the famous video where Gary Kasparov resigns to IBM’s Deep Blue during the gameplay:

Some other examples of Reactive AI are:

• AlphaGO, a computer program developed to play the board game GO developed by DeepMind (a company backed by Google)
• Self-driving cars by Tesla
• Spam filters used for detecting spam emails.

How Reactive AI works?

Here are the steps behind the working principle of Reactive AI. Let’s go through them one by one:

Note: The process defined below explains the working principle of Deep Blue.

Step 1: In this step, a commander is established. This commander analyzes the entire board to figure out patterns and other data i.e. the position of the Chessman, the alive chessman of deep blue, and different positions to play on the grid.

Step 2: At the second level, a consultant comes into play that provides a general solution and alarms the system for any incoming situation.

Step 3: The commander takes the hint from the consultant, and weighs the level of the situation, thereby, taking the best possible action.

Step 4: These hints are used by a third entity as the micromanager and turn into actual commands.

Adding to it, a reactive plan can be constructed in several ways. Let's have a look at them too down below.

Different Types of Reactive Plans

• Condition-Action Rules: This approach works on If-Else conditions. If the conditions are fulfilled then the step is taken, otherwise, reverted.
• Finite State Machines: This approach works on determining the state of the agent. The state of the agent changes at each level to determine the best action possible, whenever, a condition is fulfilled.
• Fuzzy Approaches: It is the combination of the above two approaches. As per this, condition, action, and states are nothing but boolean quantities that are either “0 or 1” or “Yes or No”.
• Connectionists Approaches: In this, there are multiple units that are connected with each other in an artificial neural network. Each unit is subjected to an abstract activity that induces a behavior into another unit. The unit that induces an activity is known as an activity inducer. 

2. Theory of Mind

The AI that follows the principle of the Theory of Mind is identical to the theory of mind used in psychology itself. As per psychology, the theory of mind deals with the attribution of different mental states such as thoughts, beliefs, intent, perception, etc. 

Likewise, Theory of Mind AI is designed to understand the mental state of the user i.e. his/her beliefs, intentions, emotions, desires, etc. This type of AI is already in development and has huge potential to transform the way we interact with computers.

It utilizes a variety of AI techniques such as natural language processing, machine learning, and computer vision. The intention behind the creation of this AI is to understand the meaning behind human language and not simply mimic it.

For this machine learning is used to understand the data to improve its performance with time. Adding to it, the technology also requires computer vision in order to make meaning out of what the system perceives.

Some potential applications of the Theory of Mind could be:

• Virtual Assistants: It can be used for creating virtual assistants that are as intuitive as a human. A virtual assistant that surpasses the capabilities of ChatGPT by ten folds. The application can also be great for psychological support.
• Robotic Interactions: It will help in creating robots that will have almost human-like interaction capabilities.
• Remote Education via AI: The only thing that a remote education via presentations lag is an intuitive teacher. Sometimes these edutech companies have access to phenomenal teachers, however, an AI-based one can be tweaked to make it more interesting.

However, the Theory of Mind is still a hypothetical concept because of the following constraints such as:

• Data: It would require tonnes of data to learn which can be difficult as well as expensive to collect.
• Complexity: Creating human-like behavior via an algorithm is a complex task and isn’t anything available as of today.
• Interpretability: For TOM (Theory of Mind) AI to truly behave like a human, it needs to understand at a subconscious level. This becomes almost impossible in today’s time considering there isn’t technology available that can comprehend this task.

It would still take some time to turn TOM into reality. However, it is a promising technology with a much wider range of applications in comparison to the technology out there. Adding to it, with the current state of development in this department, it is most likely possible that the system developed will reach the determined level of sophistication.

3. Self Aware AI

"Philosophers haven't really settled on a definition of consciousness yet," he said, "but if we mean self-awareness, and these kinds of things... I think there's a possibility that AI could one day be." says the CEO of DeepMind (backed by Google) as stated in a Futurism article.

Out of all the different types of AI, a Self Aware AI would be one that would be aware of its own existence. The CEO of DeepMind has said it so well that philosophers are struggling to find consciousness, the possibility of AI becoming sentient one day is nothing short of a miracle.

Although, if we see through the filters of reality, it is a far-fetched hypothetical concept that isn’t possible with what we have today. On the contrary, there are research projects by OpenAI (the creators of ChatGPT), AI project in Consciousness and Self Awareness Lab at the University of Sussex, Project Alexandria by Allen Institute for Artificial Intelligence, etc. that are working to achieve it.

Key Components of Self-Aware AI

Out of all the different types of artificial intelligence systems in order to become a self-aware AI, the developed system needs to have several attributes that are mentioned below:

Consciousness

The machine needs to be aware of its own consciousness. Being conscious means having the capability to introspect queries, responses, thoughts, beliefs, etc., and derive an answer from them. 

Consciousness is the capability to perceive information from the external world and process it internally. For humans, it is by virtue of the brain that we feel conscious. To a certain degree to it won’t be wrong to describe it as feeling alive. 

Right now, there are only two methods that are established in order to create something like that:

• The first would be to use Integrated Information Theory. It is a framework for understanding how a certain system feels conscious and ultimately goes back to the biology of things.
• The second would be to create a neural network as dense as that of a human brain and leave it to develop for different experiences to finally attain consciousness.

Self-Identity

The striking feature of human consciousness is to self-identify itself. If we look at our own reflection, we are able to understand that the person standing is me. In fact, even if we are standing with a bunch of people, the story is the same.

In order for this AI types to be truly Self Aware, it is first required to identify itself in different conditions. This seems like an easy task from a human perspective but creating something like that truly identifies itself would be one of the biggest breakthroughs, both in technological and philosophical senses. 

Capability of Expressing Emotions and Motivation

Showing and understanding emotions, and feeling motivated to do something is another key aspect that makes us human. The capability to feel emotions is what often determines us from machines. It is important to take it into consideration because the majority of the time humans take actions or reach an understanding via expressing emotions. Also, it is the motivation that compels us to achieve a task.

There can be several potential benefits of developing a Self Aware AI which are:

• Collaboration between self-aware AI and humans to innovate new things and solve novel problems
• Self Aware AI can contribute in terms of creative tasks such as painting, creating art pieces, and creative writing, and can contribute even in the field of science & technology.
• A self-aware AI will be able to help us in advisory and decision-making in various facets of knowledge and experience.

However, there are some potential implications too such as:

• It could lose control
• It can have a superiority complex
• It could malfunction and become dangerous
• It can be used for malicious purposes as it can be manipulated just like humans

4. Artificial Narrow Intelligence

Often referred to as “Weak AI” or “Narrow AI”, artificial narrow intelligence (ANI) is great for performing a single task. The knowledge acquired by a narrow AI isn’t applied to other associated tasks.

Unlike the different types of artificial intelligence, this one is specific in nature and similar to reactive AI doesn’t try to mimic human intelligence. Instead, it is completely focused on a single task. In fact the majority of systems today make use of this type of AI.

Some great examples of this types of artificial intelligence are:

• Facial Recognition System: This system is used by smartphone companies and companies like Google and Facebook to detect people in pictures.
• Chatbots: Some of the most common examples of chatbots or virtual assistants that use this types of AI are Siri, Google Assistant, Alexa etc.
• Self-driving Vehicles: Tesla Cars are one of the most popular examples of this types of AI. Adding to it, this particular technology is also being used in drones, boats, factory robots, etc.
• Predictive Maintenance: This system is developed to predict which system might fail and which may prolong. This is done by the machine using data from the sensors.
• Recommendation engine: The best example of this type of Artificial Intelligence system would be Netflix, Amazon Prime, Hulu, etc.

Some limitations of this types of AI technology are:

• Narrow AI can only perform simple problems and can not be included in complex reasoning.
• These systems require large datasets to perform accurately.
• It can be difficult to interpret narrow AI.

5. Artificial General Intelligence

Artificial general intelligence is a hypothetical concept that aims to achieve the consciousness level of the human mind. In terms of sentience and capability, it is much closer to self-aware AI. 

Unlike the categories of AI, this type of artificial intelligence won’t be dependent on rule-based principles. Instead, it will be capable of learning from its own environment and designing its own way of understanding just like humans. 

Right now, this AI type doesn’t exist. However, we have weaker AIs that are capable of handling complicated tasks with much ease. For example, Watson, ChatGPT, Bing AI, etc.

The benefits of General Artificial Intelligence are:

• Capability to solve complex problems without human intrusion
• High level of automation across industries
• It can help humans in providing answers to complex queries

6. Limited Memory

Limited memory AI is an AI type that stores its previous experiences and data in its memory. Based on that it derives better insights, predictions, and actions. A limited memory AI uses ML technology to learn and evolve complex tasks. 

There are several different kinds of AI technologies that utilize limited memory. These are:

• Long Short-term Memory: This type of AI is used via an artificial neural network. This AI type not only feeds on a single data point rather it feeds the entire. This makes it capable of predicting what should come next.
• Reinforcement Learning: In this method, limited memory is used by the system for trial and error. Upon different iterations an AI model undergoes, it generates insights that are collected by the limited memory and makes the AI model robust and dependable.
• E-GAN (Evolutionary Generative Adversarial Network): This type of AI model works on the principle of mutation. Therefore, this technique helps an AI model to evolve with every cycle and become better with time.

Some examples of this category of AI are:

• Autonomous Vehicles
• Recommendation Engines
• Prediction Models

7. Artificial SuperIntelligence

Nick Bostrom of the University of Oxford defines superintelligence as, “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”.

To further define it in a much simpler way, it is the artificial intelligence that surpasses human intelligence that will be considered super intelligence. As of now, this type of AI is another hypothetical concept. However, it has been an inspiration for multiple sci-fi movies. 

Some of the most popular AI superintelligence systems are:

• Skynet (Terminator)
• Ultron (Marvel)
• HAL 9000 (2001: Space Odessey)

Wrapping Up!

Till its inception, there have been several iterations in the types of AI. However, the things that fascinated people around the topic of AI, are still hypothetical. There are multiple companies like OpenAI, DeepMind, etc. are constantly trying to blur the boundaries between AI and human intelligence. However, it would be interesting to see when it becomes a reality.
 

• What are the types of neural networks in artificial intelligence?

  There are several types of neural network in artificial intelligence such as:

  • Perceptron model
  • Multilayer perceptron
  • Convolutional Neural networks
  • Feed Forward Neural Networks
  • Radial Basis Function neural network
  • Recurrent neural networks
  • Sequence to sequence model
  • Modular neural network
  • Long Short- Term Memory
• What type of AI is ChatGPT?

  ChatGPT is a generative AI that is a combination of limited memory AI and narrow AI.

• What is the difference between types of AI and technologies of AI?

  The types of AI are basically the categories of AI and what they inherently are such as limited memory, reactive AI, general AI, etc. However, the technologies of AI are deep learning, neural networks, machine learning, natural language processing, etc.

• What are the types of artificial general intelligence?

  Artificial general intelligence is a theoretical concept, therefore, there are no types of AGI. However, there are similar theoretical concepts such as the Theory of Mind AI, Self Aware AI, etc.