How Does Artificial Intelligence Work?
AI Approaches and Concepts
Less than a decade after he cracked the Nazi Enigma encryption machine and helped the Allied forces win World War II, mathematician Alan Turing changed history for the second time with a simple question: "Can machines think?"
Turing's paper "Computing Machinery and Intelligence" (1950) and his subsequent Turing Test established the fundamental goal and vision of artificial intelligence.
At its core, artificial intelligence is a branch of computer science that aims to answer Turing's question in the affirmative. It is an attempt to replicate or simulate human intelligence in machines. So much so that no singular field definition is universally accepted.
Can machines think? – Alan Turing, 1950
The main limitation in defining AI as simply "building machines that are intelligent" is that it doesn't really explain what AI is? What makes a machine intelligent? Artificial intelligence is an interdisciplinary science with many approaches, but advances in machine learning and deep learning are creating a paradigm shift in virtually every sector of the technology industry.
In their ground-breaking textbook Artificial Intelligence: A Modern Approach, authors Stuart Russell and Peter Norvig approach this question by unifying their work around the topic of intelligent agents in machines. With this in mind, AI is "the study of agents that receive perceptions from the environment and perform actions".
Fields of AI:
- Think humanly
- To think rationally
- Act humanely
- Act rationally
Norvig and Russell focus in particular on rational agents acting to achieve the best outcome, noting that "all the skills required for the Turing test also enable the agent to act rationally". While these definitions may seem abstract to the average person, they help focus the field as a field of computer science and provide a blueprint for imbuing machines and programs.
Four types of artificial intelligence:
Reactive Machines:
A reactive machine follows the most basic principles of artificial intelligence and, as its name suggests, is only able to use its intelligence to perceive and react to the world in front of it. A reactive machine cannot store memory and, as a result, cannot rely on past experience to make real-time decisions.
Direct perception of the world means that reactive machines are designed to perform only a limited number of specialized tasks. However, deliberately narrowing the world view of a reactive machine is not some kind of cost-cutting measure, and instead means that this type of AI will be more trustworthy and reliable – it will respond in the same way to the same stimuli every time.
A famous example of a reactive machine is Deep Blue, which was designed by IBM in the 1990s as a chess supercomputer and beat international grandmaster Gary Kasparov at the game. Deep Blue was only able to identify the pieces on the chessboard and know how each moved based on the rules of chess, recognizing the current position of each piece and determining what the most logical move would be at that moment. The computer did not make future potential moves on its opponent or try to place its own pieces in a better position. Each turn was considered its own reality, separate from any other move that had been made before.
Another example of a reactive gaming machine is Google's AlphaGo. AlphaGo is also unable to evaluate future moves but relies on its own neural network to evaluate the evolution of the current game, giving it an edge over Deep Blue in a more complex game. AlphaGo also defeated the game's global competitors when it beat Go champion Lee Sedol in 2016.
Although reactive machine AI is limited in scope and not easy to change, it can achieve a level of complexity and offers reliability when built to perform repeatable tasks.
Limited memory:
An AI with limited memory has the ability to store previous data and predictions as it gathers information and considers potential decisions—essentially looking into the past to see what might come. Artificial intelligence with limited memory is more complex and offers more capabilities than reactive machines.
Memory-limited AI is created when a team is constantly training a model to analyze and exploit new data, or an AI environment is created to train and refresh models automatically. There are six steps to follow when using AI's limited memory in machine learning: training data must be created, a machine learning model must be created, the model must be able to make predictions, and the model must be able to receive feedback from a person or the environment, this feedback must be stored as data and these steps must be repeated as a cycle.
There are three main machine learning models that use memory-limited artificial intelligence:
Reinforcement learning, which learns to make better predictions through repeated trial and error.
Long Short Term Memory (LSTM), uses past data to help predict the next item in the sequence. LTSMs consider more recent information to be most important in making predictions and discount data from the past, although they still use it to make inferences.
Evolutionary Generative Adversarial Networks (E-GAN) evolve and grow over time to explore slightly modified paths based on previous experience with each new decision. This model is constantly looking for a better way and uses simulation and statistics or chance to predict outcomes throughout the cycle of evolutionary mutations.
Theory of mind
Theory of Mind is just that – theoretical. We have not yet achieved the technological and scientific capabilities necessary to achieve this next level of artificial intelligence.
This concept is based on the psychological premise of understanding that other living beings have thoughts and emotions that influence human behavior. In terms of AI machines, this would mean that AI can understand how humans, animals and other machines feel and make decisions through self-reflection and determination, and then use that information to make its own decisions. Essentially, machines would need to be able to grasp and process the concept of “mind,” the fluctuations of emotion in decision-making, and a litany of other psychological concepts in real-time, creating a two-way relationship between humans and AI.
0 Comments