Artificial Intelligence (AI) is a field of computer science that explores machines learning, thinking, acting and solving problems humans naturally excel at. Well, that’s one way of defining it, however there are a ton of definitions. This post will explore some of the ways we define AI.
At a broader level, AI also encompasses knowledge representation, perception, optimisation, self organising systems, and complex algorithms based on genetics, evolution and survival of the fittest. Furthermore, AI draws inspiration from other natural systems like human organ operation, and other forms of intelligence such as animal, herd, and swarm intelligence to uniquely solve problems.
Weak, which is more commonly known as ‘Narrow AI’. It focuses on solving a single narrow task. It addresses specific application areas such as playing strategic games, language translation, self-driving vehicles, and image recognition.
Strong, which is sometimes called ‘Artificial General Intelligence’. It refers to a future AI system that exhibits intelligent behavior at least as advanced as a person across the full range of cognitive tasks. This definition doesn’t give any thoughts to the system exhibiting consciousness (which it may inherit or evolve to have!)
Think like a human: We take inspiration from the the massive distributed connectedness of the human brain and neuron firing mechanism, as the biological basis for neural networks, cognitive architectures, bayesian inference and massively parallel processing.
Act like a human: To see and understand real world entities, and how they relate to each other through video capturing and computer vision techniques. To communicate in human language, understanding people’s intentions and emotions through natural language processing techniques. Also the ability to store knowledge, reason with it, and continuously learn.
Think rationally: Solving problems through inductive or deductive logical reasoning. Inferring a good solution to a constrained problem with multiple circumstances and outcomes. Also, optimising a decision to get maximum benefit from it, for instance, what is the best next move I can make in this chess game?
Act rationally: Embodying a rational, intelligent system into an agent or robot that can use it’s own sensor data to achieve goals through perception, planning, reasoning, learning, communicating, decision-making, and acting.
Logical reasoning: Solving problems through logical deduction or induction.
Knowledge representation: Representing knowledge about the world through entities, events, and their relation.
Planning and navigation: Setting and achieving goals through sequences of actions.
Communication: Understanding written and spoken language.
Perception: Deducing things about the world from visual images, sounds and other sensory inputs.
Expert Systems: Emulating human expert decision-making through reasoning about knowledge stored in a large set of conditional rules.
Natural Language Processing: The ability to understand natural human language.
Computer Vision: The ability to understand and make sense of the word through visual data.
Evolutionary Computation: Global optimisation methods inspired by Darwinian evolution.
Machine Learning: The ability to learn without being explicitly programmed.
Robotics: Machines that can deduce things about the world from visual sensory, sounds and other sensory inputs.
Symbolic AI which uses logical reasoning based on abstract symbols. It’s commonly referred to as ‘old AI’ and was popular from 1950 - 1980, with its greatest contribution being expert systems.
Connective AI which builds structures inspired by the human brain. Neural Networks and all of their different architectures encompass most of this space
Evolutionary AI uses methods inspired by Darwinian evolution; genetic algorithms, genetic programming, and particle swarm optimisation are the most popular
Bayesian AI uses probabilistic inference. This includes gaussian processes, hidden Markov models and bayesian belief nets
Analogistic AI extrapolates from similar cases seen previously, resulting in simple algorithms that are easy to understand. k-nearest neighbour, naive bayes, and k-means clustering are all examples.
This is just a few different dimensions to view A.I. We use a number of these ways to think about A.I, and to explain to our clients what it all is. Some ways resonate better with certain people. For me, I like to think of A.I through the five different mindsets. I’m a bit of an evolutionary as my research is in the field of Particle Swarm Optimisation (an Evolutionary Computation method based on swarm intelligence). However, I think the future of A.I will use more and more ideas from the combination of connective, evolutionary and bayesian AI.