Analogous to the human structure of the brain, Artificial Intelligence includes layers of connections that attempt to imitate physiological neural connections. In review of human learning, we see that this process slowly builds through the levels of sensation, perception, and cognition. An infant spends the first months of life attuning its senses to the sound, sight, touch, smell, and taste of its world. Soon, it grasps the “what” of sensations, labelling them with tags that are perceptions. By the end of the first year, understandings or cognition of the world dawn, so that the infant brain has developed layers of neural connections that will continue to expand well into old age. From sensation, a child’s brain journeys through experiences into deep learning and wisdom capable only in a human brain.
Science fiction long fantasied the “machine” development of a humanoid since the mythology of Prometheus with the help of the gods to Mary Shelley’s Frankenstein. When the modern computer showed its ethereal capacity for speed of calculations and eventual cinematographic effects, it became a logical step to imitate the brain’s capacity to process information with supersonic speed. Computer scientists discovered that machines could “learn”, even though Alan Turing demonstrated that machines do not think. Neil Nie prefers the concept that computers calculate. Nevertheless, computer science pushed forward in tagging electronic computer connections as “neural connections.” Machine learning is what computers do, now known as “Artificial Intelligence”.
Machine learning embraces two types of artificial intelligence: narrow AI and AI general intelligence (AGI). AGI is the fantasy that believes a machine will eventually have the autonomy of the human brain and mind, something serious computer scientists hesitate to anticipate.
Narrow artificial intelligence is viewed as two-fold: simple machine learning that focuses on single tasks and deep learning using biologically-inspired neural networks containing hidden layers of connections that go deep into learning. Though Narrow Artificial Intelligence is termed “Weak AI”, it still has the power to be an amazing tool simulating human learning. Frank Chen is quoted in the BuiltIn Artificial Intelligence website: “Artificial intelligence is a set of algorithms and intelligence to try to mimic human intelligence. Machine learning is one of them, and deep learning is one of those machine learning techniques.”
Deep learning became the stepchild of Narrow AI as computers increased their capacity to hold big data and learn to recognize patterns through supervised learning with the use of algorithms, while unsupervised learning through self-organizing was based on Hebb’s principle that “neurons that fire together wire together”. Unsupervised learning allows for reinforcement allowing connections to strengthen deep learning. These neural networks are layers of connections that process data, assisting the computer to go deep in learning, and providing best predictions for best decisions and best results.
Key to deep learning is the stacking of algorithms, building depth upon early sets of algorithms with more algorithms to amplify the “neural connections” which build smart machines that perform tasks usually attributed to human intelligence. Deep learning simulates human intelligence in machines and through statistical techniques, and helps the computer to learn to get better at what the machine is doing.