A Simple Definition
At the most basic level, Artificial Intelligence is the study of how to program computers so that they can simulate the intelligence of people and animals. More formally, it is the theory and development of computer systems able to perform tasks that normally require human intelligence.
These tasks include:
Learning and adaptation
Understanding language
Perceiving images and patterns
Problem-solving
Making decisions
The Core Idea: Simulating Human Intelligence
The point of creating an AI is not to write a set of unchanging instructions telling the computer what to do in every situation. This is just standard programming. You can think of it as an early form of AI, too, in which humans hard-code a set of instructions into a computer program.
What’s more interesting is AI that can learn from experience the way humans do. A human might be able to teach a computer rules and facts, but computers are often better at recognizing patterns and relationships.
To put it simply:
Traditional Programming: You tell the computer what to do. (“If the user presses the backspace key, delete the character to the left.”)
Artificial Intelligence: You teach the computer what to do. (“By studying this data set, you’ll learn to associate these traits with cats, dogs, and bears.”)
The Different Types of AI
AI can be generally divided up into three types based on capabilities
1. Artificial Narrow Intelligence (ANI)
2. Artificial General Intelligence (AGI)
3. Artificial Superintelligence (ASI)
1. Artificial Narrow Intelligence (ANI)
It is AI that has been created and trained to do a narrow task. ANI, currently, is the only form of AI that exists. Some examples might include
• The voice recognition system on your phone
• An algorithm that recommends products on Amazon or Netflix
• The Facebook friend recommendation algorithm
• The visual recognition systems in self-driving cars
These systems might be able to outperform humans on a certain domain, but they can’t do anything outside of that domain. In other words, the AI that can beat a world-class chess player at chess is absolutely terrible at every other game.
2. Artificial General Intelligence (AGI)
AGI is a machine that can outperform humans at any intellectual task. AGI would have human-level common sense and would be self-aware. This is science-fiction material (think C-3PO or Data from Star Trek), but it is the ultimate goal of many AI researchers.
3. Artificial Superintelligence (ASI)
It would be a form of AI that is much better than humans at everything. It would have significantly improved cognitive abilities and knowledge in all domains, including mathematics, linguistics, and social skills.
This is pure speculation, but many people have speculated about the existential risk that ASI could pose.
How Does AI Work? Key Subfields
AI is a broad area of computer science, and most of the systems that you probably think are powered by AI are actually powered by a few key subfields. Let’s review them.
Machine Learning (ML)
This is the workhorse of modern AI capabilities. ML is a statistical framework that allows for computers to learn new information and improve their performance on a given task.
When you train an ML algorithm, instead of explicitly programming a set of rules for the computer to follow, you give it a set of examples (images of cats and not images of cats, for instance) and you let it find patterns in the data to classify new examples. ML algorithms can be very good at pattern recognition. We can think of them as building their own lookup table that maps inputs to outputs.
Deep Learning (DL)
This is a subfield of ML that attempts to mimic the human brain using an interconnected set of “neurons.” This set of neurons is called a “neural network” and the entire network can learn complicated patterns in data by reviewing tens or hundreds of thousands of training examples.
DL is responsible for many of the AI applications we use today, including things like language translation or image captioning.
Natural Language Processing (NLP)
This refers to the field of computer science and linguistics that deals with enabling computers to understand and generate natural language.
NLP techniques are used in Google Translate or your word processor’s grammar correction tool.
Computer Vision
This area of CS enables computers to understand and interpret visual data. CV algorithms are responsible for things like facial recognition or the filters on your social media photos.
Applications of AI
Most people are actually interacting with Narrow AI systems every day. Some common examples might include
• Recommendation systems on YouTube, Netflix, and Spotify
• Navigation systems like Google Maps or your ride-share app
• Social media algorithms and content filters
• Smart assistants like Siri or Alexa
• Medical diagnosis systems
• Credit card fraud detection
• Autonomous cars
In short, AI is already playing a major role in our lives, and these systems are only going to become more prevalent in the future. While ANI is the only type of AI that currently exists, researchers are already working on AGI, which would be significantly more powerful and complex. However, most current systems are built using the statistical framework of ML and DL algorithms.
