Nilson Onion Model in Artificial Intelligence
A famous researcher and planner of AI, Nils J. Nilsson, in his book "The Quest for Artificial Intelligence: A History of Ideas and Achievements" characterized the various components of AI in terms of the “Onion Model”. Hence, the model is named as Nilson Onion Model. It is a conceptual model that describes the various levels of abstraction involved in designing intelligent systems.
The Nilson Onion Model is often used in the field of artificial intelligence (AI) by engineers to perform complex and expert jobs. The model is particularly useful for designing systems that can learn from their environment and improve their performance over time. At the core of the Nilsson Onion Model is a set of basic capabilities that all intelligent systems should possess, including perception, reasoning, and action. These capabilities are organized into layers, with each layer building on the capabilities of the previous layer.
The different layers in the Nilson Onion Model in AI are as follows:
- Perception Layer: This layer involves the sensors and other input devices that allow the system to perceive its surroundings. For example, in an autonomous vehicle, the perception layer would include cameras, LIDAR sensors, and other devices that enable the vehicle to sense its environment.
- Learning Layer: The learning layer is responsible for enabling the system to learn from its environment and improve its performance over time. An example of the learning layer in action would be a chatbot that uses machine learning algorithms to learn from user interactions and improve its responses over time.
- Reasoning Layer: The reasoning layer processes the information gathered by the perception layer and uses it to make decisions and take action. An example of the reasoning layer in action would be an AI assistant that uses natural language processing to understand user requests and then uses a decision-making algorithm to determine the best course of action.
- Application Layer: The application layer represents the specific tasks that the intelligent system is designed to perform. An example of the application layer in action would be an AI system that is designed to detect fraudulent transactions in a financial system. The application layer would include the specific algorithms and techniques that allow the system to carry out this task.
The onion model is a graph-based diagram template for describing and expanding or extending the relationship between several concepts. The name is a metaphor for the layered shells that become visible when you cut open an onion.
The components of the Nilson Onion Model are shown in the figure below:
The Basic Elements of AI
The inner ring depicts the basic elements of the AI framework while the outer ring is composed of the application areas.
- Heuristic Search
Heuristics is an operationally effective part of the information to perform the search in a problem space. Their purpose is to minimize search on average. Heuristics are used to make the guess. The basic idea of heuristics search is that rather than trying all the possible search paths, the focus should be on those paths which are nearer to the goal state. Different heuristics search algorithms are :
Hill Climbing
Best First Search
A * Algorithm
- Modelling and Knowledge Representation
Early on, AI researchers discovered that intelligent behaviour is not so much
due to the methods of reasoning, as it is dependent on the knowledge one has to
reason with. (As humans go through life they build up tremendous reservoirs of
knowledge.) Thus, when substantial knowledge has to be brought to bear on a
the problem, methods are needed to efficiently model this knowledge so that it is readily
accessible. The result of this emphasis on knowledge is that knowledge
representation is one of the most active areas of research in AI today. The needed
knowledge is not easy to represent, nor is the best representation obvious for a given
task. The knowledge can be represented in the form of binary numbers, mental images, written text, and other symbolic forms.
Various knowledge representation methods include Tables, FOPL, Semantic Networks, Scripts, Frames, and conceptual dependency.
Knowledge Modelling refers to the concept of representing knowledge in such a form that it can be easily accessed and retrieved when required.
- Common Sense Reasoning and Logic
AI researchers found that common sense (virtually taken for granted in humans) is the most difficult thing to model on a computer. It was finally concluded that common sense is low-level reasoning, based on a wealth of experience. In
acquiring common sense we learn to expect that when we drop something, it falls,and in general what things to anticipate in everyday events. How to represent common sense in a computer is a key AI issue that is unlikely to be soon solved.
Another area that is very important in Al is logic.
- How do we deduce something from a set of facts?
- How can we prove that a conclusion follows from a
given set of premises?
Computational logic was one of the early golden hopes of AI to provide a universal problem-solving method. However, solution convergence proved to be difficult with complex problems, resulting in diminishing interest in logic.
Logic is now enjoying a revival based on new formulations and the use of heuristics to guide solutions. Logic is the method of reasoning based on new formulations and the use of heuristics to proceed to a solution.
- AI Languages and Tools
In computer science, specific high-level languages have been developed for different application domains. This has also been true for AI. Currently, LISP and PROLOG (Programming in Logic) are the principal AI programming language. The LISP (List Processing Language, developed in the late ’50s by John McCarthy) has been the prime language. Examples of LISP are AutoCAD, Remote Agent Software, etc.
The Outer Ring is Composed of Application areas of Artificial Intelligence which are discussed as:
Based on these basic elements, Nilsson identified four principal AI application
areas (shown in the ring of the Figure above)
- Natural Language Processing (NLP)
NLP is concerned with the basic interaction of computer and Human Languages. It deals with natural language front ends to computer programs,
computer-based speech understanding, text understanding and generation, and related applications. The goal of NLP is “to accomplish human-like language processing. The main goal of NLP is to enable computers to understand, interpret, and generate human language. NLP involves a range of techniques and methods, including machine learning, deep learning, statistical modeling, and rule-based systems.
NLP has many applications, including language translation, sentiment analysis, speech recognition, chatbots, text classification, and information retrieval. Some common examples of NLP in action include virtual assistants like Alexa, language translation services like Google Translate, and spam filters that analyze email content to determine whether it is spam or not. One of the key challenges in NLP is the complexity and variability of human language
- Computer Vision
Computer Vision is concerned with enabling a computer to see, identify or
understand what it sees, locate what it is looking for, etc. Computer vision is the science and technology of machines that see. It is concerned with the theory of making artificial intelligence systems that obtain data from images. Examples include Detecting events, Controlling Processes, organizing information, etc.
- Expert Systems
Expert Systems is perhaps the hottest topic in AI today. How do we make a
computer act as if it was an expert in some domain? For example, how do we get a
computer to perform medical diagnosis or VLSI design? It is the system that behaves like a human being. The knowledge is represented in the form of rules. Examples include MYCIN,DENDRAL.
- Problem Solving and Planning
There is a need for general-purpose problem-solving methods to hack the problem which does not have experts. In order to solve such problems, the best problem-solving strategies along with knowledge representation are required. To do the problem solving, planning is needed to –
- Choose the best rule to apply using heuristics.
- Detect when the solution is found.
- Detect the Dead ends etc.