To learn more, check out my publication with apress on this topic.
The definition of Artificial Intelligence has evolved ever since its first reference in 1956 at the Dartmouth conference. From emulating how the human brain works to solving focused, complex problems, doing all that a human can do like see, hear, communicate, act, learn, perceive, think, decide, demonstrate emotion and compassion or interact with the environment and responding in a context. The recent AI breakthroughs with vision, language recognition, and self-driving vehicles changed the way we perceive AI today. Given below is a simple and informal definition of Artificial Intelligence.
Artificial Intelligence (AI) as a broad umbrella term can be defined as field of computer science that involves enabling computers to behave like humans or perform tasks that usually requires human intelligence.
The purpose of Artificial Intelligence systems has been evolving. In this section, we will cover different types of Artificial Intelligence systems that have been categorized based on its core purpose. We can also observe how these different types of AI systems signify a step towards building smarter systems.
The diagram below lists different types of Artificial Intelligence:
1. Reactive Machines AI was the first kind of AI t. These machines do not have memory and do not use information from experience. In these machines, the current context is directly perceived as it is and is put into action. This makes the computer behave precisely the same way every time it encounters a situation. The benefit of this is a reliable and consistent outcome. Examples: Deep Blue (a chess-playing computer developed by IBM that won against Kasparov in the game of chess).
2. Limited Memory AI machines,on that other hand, look into the past and use them as a preprogrammed representation of the world and apply to the current data set. For example, in self-driving cars, the data regarding the lane markings, speed limits or road directions, current speed of the vehicle, and relative neighboring car speeds, decisions on when the car should change lanes are taken.
3. Theory of mind AI machines are intelligent machines that use advanced technologies that have more to do with understanding human emotions. The “theory of mind” is a psychological term that refers to the fact that living beings have feelings and thoughts and that they determine their behavior.
4. Self Aware AI machines: These machines are an extension to Theory of Mind AI They can configure representations; this means we will have devices that are conscious and aware given a context. This is also called Human Aware AI or Human Interaction AI. There are no prototypes built on these machines.
Type of AI
Another way of categorizing AI systems is based on the degree of complexity of the problem on hand; Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI), and Artificial Super Intelligence (ASI).
Artificial Narrow Intelligence (ANI) is about solving a problem against a given request with a narrow range of abilities. A capability like Siri in smart phones can be considered an example in this case. This is also called Weak-AI, Artificial General Intelligence (AGI) on the other hand is referred to as Strong-AI and refers to a machine that is as capable as humans. Pillo Robot is an example where the robot can diagnose an illness and administer pills as well. Artificial Super Intelligence (ASI) is about machines that can do beyond what humans are capable of. Alpha 2 robot for a first attempt towards this where a robot can manage a smart home and operate things at home. It potentially could be a member of the family. Most of the existing AI today is ANI. AGI and ASI are at an evoluationary stage.
The diagram below represents the core functions and features of an Artificial Intelligence system at the center and related sub-fields that support implementing these functions.
The applications or sub-fields of Artificial Intelligence are:
1. Natural Language Processing
2. Robotics
3. Machine learning and Deep learning
4. Expert Systems
5. Speech or Voice Recognition
6. Intelligent Automation and
7. Computer Vision
Each of these sub-fields is interrelated, and any real-world implementation usually includes one or more sub-fields. In the next section, we will look into a brief definition of each of these sub-fields with real-world examples and related technology tools wherever applicable, before taking a deep dive into Computer Vision.
Natural Language processing
Natural Language Processing, also referred to as NLP, refers to an area of specialization in computer science that deals with analyzing and deriving useful or meaningful information from natural language or human language. At a high level, this requires employing formal techniques like tokenization, relationship extraction in the context for a specific business case, word classification, and sentence detection. For a language, the syntax that refers to basic rules the language follows and semantics that refers to its meaning play an important role. The complexity comes with the fact that the purpose of text can be ambiguous and can change with the context. For example, the word “saturation” could mean different when used with colors against the background of human behaviors.
The diagram below refers to critical areas of current applications of NLP.
NLP is used in a wide variety of disciplines to solve a variety of problems. A brief list of applications is; Searching, which refers to identifying specific elements of text within a bigger context of the content. Machine translation, which is about translating text from one natural language to another, Summarizing text described across more substantial content in documents, blogs, etc. Named-Entity Recognition (NER) to extracting names of locations, people, and things from the text, Information grouping that is about categorizing a text based on its content and context. Sentiment analysis usually used to perceive and provide automated help or feedback on how a product is doing in the market, like a book, a movie, etc. Answering queries or support that is used medicine or service; for example, chatbots and Speech-recognition that helps analyze and understand the context automatically from a conversation with humans.
Some essential NLP techniques with examples are listed below:
Robotics
Robotics is a computer science discipline that deals with the design, programming, engineering, and development of physical Robots or machines that are built to execute tasks that are usually done by humans.
Adoption of Robotics was initially targeted for jobs that are hazardous for humans like welding, riveting, mining, cleaning toxic wastes, or defusing bombs among others or those that need high precision or have a low tolerance for human errors like long surgeries in the medical field.
While Robots have been around and evolving for several decades, it is only now that the use of robots in day-to-day activities is picking up. With the advent of IoT and Big Data, the assimilation of a large number of streaming data points and analysis is not a challenge. For example, if we look at a simple sensor [LJ1] on an autonomous vehicle, it processes hundreds of thousands of data points every millisecond or second to assess if a move by the car is safe and aligned to reach the target destination within the stipulated time.
Machine learning
Machine learning is a way of building intelligence into a machine that will be able to learn over time and do better using its own experience. It deals with a pattern search mechanism that is all about filtering the relevant details from the universe of more information or the environment.
Machine learning algorithms that are constructed this way handle building intelligence. The goal of a learning algorithm is to produce a result in the form of a rule that is accurate and precise to a maximum extent.
The following figure depicts various subfields of Machine learning.
Supervised learning is all working to a known expectation, which means what needs to be analyzed from the data is defined. In the case where there is no clear target in mind or specific problem to solve, the learning is referred to as Unsupervised learning. The goal, in this case, is to decipher the structure in the data first and identify potential output attributes. As an example, to train a pet pup rewarding him every time he follows instructions works very well. He figures out quickly what behavior helps him earn rewards. A learning methodology that focuses on maximizing the rewards from the result is referred to as Reinforcement learning.
Deep learning is an area of Machine learning that focuses on unifying Machine learning with artificial intelligence. For a face detection requirement, a profound learning algorithm records or learns features such as the length of the nose, the distance between the eyes, the color of the eyeballs, and so on. This data is used to address a classification or a prediction problem and is very different from the traditional shallow learning algorithm. In Chapter 2, we will cover some specific deep learning methods that are used in Computer Vision.
Expert Systems
Expert Systems also referred to as ES, are one of the most significant research domains of AI that were first related to Stanford University. These systems primarily focus on solving complex problems in a particular area, at a level of exemplary human intelligence or expertise. Expert systems are highly responsive, reliable, accurate, and highly performant. While they cannot replace a human when it comes to decision making, they are used as an advisory to humans. They can help in diagnosis, explanation, prediction, justification, or reasoning. Any Expert System includes three core components, a Knowledge Base, an Inference Engine, and a User Interface.
Expert Systems are used heavily in many domains, some examples of usage are fraud detection, identification of suspicious transactions and stock market trading in the financial field, critical ailment diagnosis and deduction of the root cause for an ailment in the medical domain and predicting potential behavior of a system by monitoring its current status against the patterns derived from earlier monitoring reports.
Speech and Voice Recognition
Speech recognition technology enables computers to recognize spoken words, which is then converted to text for analysis. A natural progression in processing includes the application of NLP techniques on the extracted text. Voice Recognition is a subset of Speech recognition with one of the goals of identifying a person based on the voice. Today, many electronic products like mobile phones, TVs, electronic gadgets support speech recognition that enables smart and automatic operations based on simple instructions. There are advanced services like Siri, Alexa, and Google Assistant from technology giants like Apple, Google, and Amazon, among others, that are breaking barriers in simplifying living and day-to-day activities.
Intelligent Process Automation
Automation has evolved from running repetitive and mundane tasks to dealing with complex cases and optimizing overall the way humans execute tasks. Robotic Process Automation (RPA) is the application of technology that allows a user to configure a ‘software robot’ (BOT) to capture and interpret existing applications for processing a transaction, manipulating data, triggering responses, and communicating with other digital systems in an efficient way and scale to more substantial work and request loads on a need basis.
Intelligent Process Automation (IPA) has more cognitive capabilities against RPA when used in conjunction with NLP (Natural Language Processing), Machine Learning, Computer Vision, and other subfields. The diagram below depicts all that constitutes an Intelligent Automation system.
Computer Vision
Computer Vision, also referred to as Vision, is the new cutting edge field within computer science that deals with enabling computers, devices, or machines, in general, to see, understand, interpret or manipulate what is being seen.
Computer Vision technology implements deep learning techniques and, in a few cases, also employs Natural Language Processing techniques as a natural progression of steps to analyze extracted text from images. With all the advancements of deep learning, building functions like image classification, object detection, tracking, and image manipulation have become more straightforward and accurate, thus leading the way to exploring more complex autonomous applications like self-driving cars, humanoids, or drones. With deep learning, we can now manipulate images, for example, superimpose Tom Cruise’s features onto another face. Or convert a picture into a sketch mode or watercolor painting mode. We can eliminate the background noise of a photograph and highlight the subject in focus, or even with most shaky hands. A stable picture can be clicked. We can estimate the closeness of, structure and shape of objects and determine the textures of a surface too. With different lights or camera exposure, we can identify objects and recognize an object that we have seen before.
For more details, get your copy of my book published by #apress:
Comentários