The seven types of artificial intelligence

Understanding the types of AI that are possible and the types that exist now will give a clearer picture of existing AI capabilities and the long road ahead for AI research

The definition of artificial intelligence (AI) and what should or not be include has changed over time. Some experts in the field gives an interesting definition: AI is everything that computers cannot currently do!. Despite the definition of AI itself is volatile and has changed over time, it could be useful to come back to the earliest (and so purest) definition of AI from when it was first coined. The official idea and definition of artificial intelligence was first coined by John McCarthy in 1956 at the Darmouth conference:

“Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine could be made to simulate it. An attempt will be made to find how to make machine to use language, form abstractions and concepts, solve kinds of problems now reserved for human and improves themselves”.

During this conference were presented some programs able to make logical reasoning, especially related to mathematics. The “Logic Theorist” program, developed by Allen Newell and Herbert Simon can be considered the first program in the AI history.

Intelligence: a multiface entity

In essence, AI is the capability of a machine to imitate human intelligence. Typically, AI systems demonstrate at least some of the following behaviors associated with human intelligence: planning, learning, reasoning, communication, perception, and, to a lesser extent, social intelligence and creativity. Let's give a quick look at these aspects (or domains)

- Planning: it explores the process of using autonomous technique to solve planning and scheduling problems. A planning problem is one in which we have some initial starting state which we wish to transform into a desired goal state through the application of a set of actions. Rather than a pre-programmed decision-making process that goes from A?to B to C to reach a final output, automated planning is complex (for example certain actions cannot happen before others) and requires a system to adapt based on the context surrounding the given challenge. The main purpose of automated planning concerns the design and execution of strategies to carry out some activities typically performed by intelligent agents, autonomous robots and unmanned vehicles. In the domain of planning we can define different sub-fields like searching and optimization.

- Learning: it is one of the fundamental building blocks of artificial intelligence solutions. From a conceptual standpoint, learning is a process that improves the knowledge of an AI program by making observations about its environment. From a technical/mathematical standpoint, artificial intelligence learning process focuses on processing a set of input/output pairs for a specific functions and predicts the output for new inputs.

- Reasoning: it plays a great role in the process of artificial intelligence. Reasoning means draw inferences (generating the conclusions from evidence and facts is termed as inference) appropriate to the situation in hand. Reasoning in artificial intelligence has two important forms, inductive reasoning and deductive reasoning.

Deductive reasoning uses a top-down approach. In deductive reasoning, the arguments can be valid or invalid based on the value of the premises. If the value of the premises is true, then the conclusion is also true.

Inductive reasoning uses a bottom up approach and it is not logically rigorous: it can provide inaccurate conclusions, as premises only provide probable support to the final output and even though the?premises are true,?they do not guarantee the accuracy of the final conclusion. While deductive reasoning helps deduce new information from logically related known information, inductive reasoning is used to arrive at a conclusion using limited sets of facts through the process of generalization. Among the sub-fields of reasoning we can mention the following: knowledge representation, automated reasoning an common sense reasoning.

- Perception: it is a process to interpret, acquire, select and then organize the sensory information that is captured from the real world. For example:?human beings have sensory receptors such as touch, taste, smell, sight and hearing. So, the information received from these receptors is transmitted to human brain to organize the received information. According to the received information, action is taken by interacting with the environment to manipulate and navigate the objects. Analysis is complicated by the fact that one and the same object may present many different appearances on different occasions, depending on the angle from which it is viewed, whether or not parts of it are projecting shadows, and so forth. With the perception, a systems become aware of their environment through the sense. Computer video ad audio processing are two typical sub-fields of the perception domain.

- Communication: NLP (Natural Processing Language), obviously the main way of communications, refers to the machine ability to identify, process, understand and/or generate information in written and spoken human communications.

Classification of artificial intelligence

The evolution (and application) of artificial intelligence can be better understood by its classification. The classification of artificial intelligence based on its capabilities features how AI is evolving. More precisely, these are the three stages through which AI can evolve, rather than the 3 types of Artificial Intelligence. Based on capabilities, the classification is the following:

1. Narrow AI (Weak AI)

2. General AI (Strong AI)

3. Super AI

The classification of AI based on functionalities highlights its technical applicability. This classification was given by Arend Hintze, an assistant professor of integrated biology and computer science at Michigan State University, in a 2016 article.

Based on functionalities, there are four types of artificial intelligence:

1. Reactive machines
2. Limited memory
3. Theory of mind
4. Self-awareness

Let us understand the different types of artificial intelligence.

The seven different types of AI

AI based on capabilities (Type 1):

Narrow AI (or Weak AI): it is a type of AI which is able to perform a dedicated task with intelligence. ANI (Artificial Narrow AI) cannot perform beyond its field or limitations, as it is only trained for one specific task: hence it is also termed as weak AI. In many cases, Narrow AI systems outperform humans in their specific domains but as soon as they are presented with a situation that falls outside their problem space, they fail. They also can’t transfer their knowledge from one field to another. Some of the most common narrow AI techniques are machine learning, natural language processing, and computer vision. IBM's Watson supercomputer is an example of narrow AI: it uses an expert system approach combined with Machine learning and natural language processing. Common examples of application using these techniques are: searching the Internet, disease detection, facial recognition, recommender systems, autonomous vehicles, predictive maintenance and analytics, robots. For example, face recognition software can recognize a familiar face, but it cannot respond to a voice command. A virtual personal assistant can respond to a voice command, but it cannot perform anything else. Much of the weak AI is the implementation of machine learning or deep learning.

General AI: That is a type of AI (AGI – Artificial General AI) that we see in movies. Also called strong AI or full AI, it has cognitive abilities that are functionally equivalent to those of a human. It can perform any task a human can and apply its intelligence to complex problems. General AI models should be able to think and understand like a human being. They need to be aware of their environment and have the capability to adapt to new situations. A suitable example of general AI is the human brain. We can say we reached general AI when we shall be able to create a machine able to work exactly like the human brain, that contains around 100 billion neurons! Fujitsu-built K, one of the fastest supercomputers, is one of the most notable attempts at achieving strong AI, but considering?it took 40 minutes to simulate a single second of neural activity, it is difficult to determine whether or not strong AI will be achieved in our foreseeable future.

Super AI: it is the AI that surpasses human intelligence and ability. It’s also known as artificial super Intelligence (ASI) or superintelligence. It’s the best at everything — maths, science, medicine, hobbies.... ASI would have a greater memory and a faster ability to process and analyse data and stimuli. Consequently, the decision-making and problem solving capabilities of super intelligent beings would be far superior than those of human beings. The potential of having such powerful machines at our disposal may seem appealing, but the concept itself has a multitude of unknown consequences.

AI based on functionalities (Type 2)

Reactive machine: it is the primary form of artificial intelligence that does not store memories or use past experiences to determine future actions. It works only with present data. It perceives the world and reacts to it. Reactive machines are provided with specific tasks, and they don't have capabilities beyond those tasks. These machines will react the same way each time if the input is not varied. An?example?of the Reactive AI is the?Deep Blue?supercomputer designed by IBM in the 1980s which won a chess competition against world champion Garry Kasparov. Expert Systems would be an example of Reactive Machines. Expert Systems are designed to have a narrow focus to very specific tasks. In fact, designing a broader view would be a detriment and reduce the Expert System’s capability in making the correct decision. In other words, each time the system encounters a specific situation, it will behave the same way.

Limited memory: it is considered to be one of the most popularly used kind of AI today. Limited memory AI is similar to reactive machines except they have a small memory that they can use to make observations over a period of time to judge the situation and give a response based on that. Limited memory AI is widely used for machine learning. Most of the current machine learning applications are based on limited memory AI.

An example of limited memory AI would be self-driving cars. Self-driving cars need at least short term memory to react properly to road signs and to observe the speeds and paths of other vehicles around them.

All the current AI systems are either reactive or limited memory devices, and they all fall into narrow AI.

Theory of mind: In psychology, "Theory of Mind" means that people have thoughts, feelings and emotions that affect their behavior. This level of AI includes all of the aspects of Reactive Machines and Limited Memory, but it adds a crucial bit of understanding: it must learn to understand that everyone (both people and AI objects) have thoughts and feelings.

Self-aware AI: Any AI from science fiction falls into this category. While it currently exists as a hypothetical concept, Self-aware AI could have incredible implications. The goal is to engineer machines that are so highly developed and skilled at imitating the human brain that they develop self-awareness. It can form memories of the past, and make predictions. Crucially, it can learn and become more intelligent based on its experiences.

Can we trust about artificial intelligence?

Despite the spreading use of artificial intelligence in our day-life, it's remains a significant challenge to develop human trust in these systems, particularly because the systems themselves cannot explain in a way graspable to humans how a recommendation or decision was reached. This lack of trust can become problematic in critical situations involving some critical fields, like finances or healthcare, where AI decisions can have life-altering consequences. To address this issue, eXplainable Artificial Intelligence (XAI) has become an active research area both for scientists and industry. XAI develops models using explanations that aim to shed light on the underlying mechanisms of AI systems, thus bringing transparency to the process. Results become more capable of being interpreted by both experts and non-expert end users alike.