Table of Contents
An interdisciplinary branch of science, artificial intelligence is focused on the development of machines with the capability of performing tasks through Human intelligence. It refers to the process of simulation of human intelligence in machines. The systems are specially trained to mimic human behaviour and action and programmed accordingly. Learning, reasoning, and perception are the goals of artificial intelligence. AI is used in several industries like; healthcare, finance, etc. have been efficiently applying AI.
Exploring the different types of AI will provide a clear view of the existing types and the challenges associated with the AI in the future types.
How is AI classified?
The main purpose of artificial intelligence is to mimic the human intelligence process. Therefore the criteria used for the classification of AI are the degree to which an AI system can replicate human capabilities. Therefore, the models are considered to be the more evolved types of AI if they can perform more human-like functions with similar efficiency. On the other hand, those types of AI which have limited performance and functionality, are considered a less evolved type of AI.
Mostly artificial intelligence can be broadly divided into two categories: based on capabilities and based on functions.
Types of Artificial intelligence
I). Type 1 AI: Based on capabilities
1. Weak AI or narrow AI (Artificial Narrow Intelligence, ANI)
- When any dedicated tasks are to be performed with intelligence, that’s where narrow AI comes. It is the most common type of AI in the world.
- As the model can only perform a task for which it is trained, the narrow AI is also termed as the Weak AI. It is unable to perform beyond its field.
- One of the best examples of a narrow AI is Apple Siri, which works on a set of predefined functions.
- Another example of a narrow AI is the IBM Watson supercomputer that combines machine learning and natural language processing with an expert systems approach.
- Examples of narrow AI include playing chess, speech recognition, etc.
2. General AI (artificial general intelligence)
- Any intellectual task similar to humans can be performed by this type of AI.
- The idea behind the development of the model lies in the fact that a smarter system should exist which is capable of thinking like a human and smart.
- At present, there is no existence of any type of such system. However, researchers are focused on the development of such a system of AI.
3. Super AI (Artificial Super Intelligence)
- This type of AI is an outcome of the general AI where the system would be able to perform any task far better than the humans through the ability of cognitive properties.
- Characteristics of super AI include planning, learning, solving a puzzle, making adjustments, etc. All by itself.
- The development of a super AI system is still a challenge and is a hypothetical concept of AI.
II). Type 2: Based on the functionality
1. Reactive machines
- It is the simplest form of artificial intelligence that performs basic functions. These are also the oldest forms of AI having a limited capability.
- No type of learning is involved in this type of AI. The model generates some output in reaction to some input. There is no storage of any input and hence there is no ability to “learn”.
- The model is based on the ability of the human mind to respond to various stimuli. There are no past experiences that are to be used for deciding the present actions.
- For automatic response against a limited set of inputs, these types of AI models can be preferred.
- The reactive machines can only function against the task for which they are programmed. Beyond that, the machines fail to perform as they have no knowledge or concept about the world.
- One of the characteristics of these types of AI models is that the machines will always behave in the same manner as they are programmed irrespective of the time and place of execution of tasks.
- No growth is associated with the reactive machines, only stagnation in recurring actions and behaviors.
Artificial intelligence examples can be found in IBM’s Deep Blue, IBM’s chess-playing supercomputer, which is a game machine that beat Grandmaster Garry Kasparov in 1997. The machine can identify the pieces on a chess-board and have the capability of predicting its next move. It then chooses the optimal move from the set of possibilities. This machine uses its present knowledge without any concept of the past.
2. Limited memory
- The limited memory type of AI comprises models that derive knowledge from previously learned information, stored data, or events.
- In addition to the capabilities of reactive machines, limited memory is capable of making decisions from learning from historical data. This type of AI involves the process of storing previous data or previous predictions. These data ultimately help in making better predictions.
- The models are trained with high volumes of training data. These data are then stored as a reference model in the memory of the system which it uses for solving future problems.
Application of this type of AI can be found in virtual assistants, chatbots, etc.
The application of limited memory can be explained by the concept of self-driving cars.
- Self-driving cars look into the past like observing the speed and direction of other cars. This is not achieved at one time, but requires the task of identifying specific objects over time.
- The above-mentioned information along with lane markings, traffic lights, road curvature, etc. are already pre programmed into the cars. With this information, self-driving cars can decide when to change lanes, or avoid getting hit, etc.
- The information is transient and is not saved as the car’s library of experience.
The limited memory type of AI is applied in three different kinds of models.
- Reinforcement learning
This type of model is applied in machine learning to predict future outcomes through interaction with the environment. It consists of cycles of trials and errors. Examples of reinforcement models include teaching the computer how to play chess.
- Long short term memory (LSTM)
LSTM models help in predicting the next outcome in a sequence. Therefore, items in the past are considered less important than the current items.
- Evolutionary Generative Adversarial Networks (E-GAN)
This type of model keeps on evolving showing the process of a growing thing. It doesn’t follow a definite path every time instead it gets modified. These modifications might lead to the prediction of a better or the least resistance path. The simulation process of the model E-GAN somewhat resembles the evolution of humans on earth.
Working system of limited memory type
This type of model works in two ways
- the model is continuously being trained on the new data
- The AI environment of the model provides the opportunity for the automatic training of the model and its renewal over the model behavior.
The above-mentioned two types of AI have been practically found in abundance. However, the next two types of AI exist as a theoretical concept or work under progress.
3. Theory of mind
- Theory of mind represents the machine learning models that have the ability of decision-making process equal to that of a human mind but done through machines.
- Researchers are currently engaged in the innovation of the conceptual type of AI, “Theory of mind”.
- This type of AI interacts with the thoughts and emotions of a human. These models will include the understanding that the thoughts and emotions of people affect the behavioural output. This ultimately influences the thought process of the “theory of mind”.
- One of the important factors of human interaction is social interaction. Therefore the hypothetical machines will have to have to identify, understand, retain, and remember emotional output and behaviours while knowing how to respond to them.
- With the information gained from people, the machines will be able to apply and adapt it to their learning. As a result, they will know how to communicate with and treat different situations.
- A highly advanced form of AI.
The other type of models at present show one way relationships like commands given to Alexa or shouting at Google maps when it shows a wrong direction. Yet, the AI model doesn’t seem to respond to angry behavior. Instead, it bows down to the commander every time. An example of this type of AI model is the robot “Sophia” created by Hanson Robotics. The humanoid bot has the ability to see and respond to interactions showing different facial expressions.
Theory of mind is a bit advanced and will prove to be better companions. These types of models are seen to be in their beginning stages.
- This type of AI represents the final stage of AI which has not been practically developed yet but has its presence only in the stories. These types of machines are still a hypothetical concept of Artificial intelligence but when developed will be smarter than the human.
- The AI model of self-aware is a step further than the theory of mind will have self-guided thoughts and reactions
- The models will evolve to a point where the system attains a state of self-awareness. It’s one of the ultimate AI research
- The models will not only have emotions with those it interacts with but will be having their own beliefs and desires.
- Although the model can lead to the progress of civilization, it might also result in catastrophic situations. With the attainment of self-awareness states, the machines will be having the ideas of self-preservation. This might lead to a situation where AI will take over humanity through the plotting of schemes by this type of AI.
The main assumption behind the development of different types of AI is that human intelligence can be represented in the form of symbolic operations which could be programmed by a digital computer. AI examples have shown to what extent the models of AI can perceive the real world. With the further development of hypothetical concepts of AI models, there might be a need for more developed machines to support the complexity of human thought.
If you’re interested to learn more about Artificial Intelligence, check out IIIT-B & upGrad’s Executive PG Program in Machine Learning & AI which is designed for working professionals and offers 450+ hours of rigorous training, 30+ case studies & assignments, IIIT-B Alumni status, 5+ practical hands-on capstone projects & job assistance with top firms.