Artificial intelligence

Narrow intelligence and general intelligence - the 2 main categories of artificial intelligence (AI).

Artificial intelligence (AI) can be broadly divided into 2 categories:

  • Narrow intelligence

  • General intelligence

Narrow intelligence, also known as Weak AI, exists, whereas General or Strong intelligence is something machines could only dream about - if they had it!

Narrow intelligence

Narrow intelligence is AI that is focused on performing 1 main task. All AI Systems that currently exist have narrow AI although they may appear to be smarter than they are. An example of this is Alexa from Amazon which has a limited pre-defined range of operations that it can carry out; it does not possess any intelligence or self-awareness capabilities although we may have the illusion that it does.

General intelligence

General intelligence makes reference to machines that can perform many tasks, be cognitively aware of what they are doing and be able to self- learn and adapt. Examples of General intelligence include HAL from 2001: A Space Odyssey, R2-D2 from Star Wars and data from Star Trek. That all these examples come from Science Fiction should give you a clue that General Intelligence is a long-term aspiration (at least for some) rather than anything you will be able to put into use anytime soon. Perhaps because of the prevalence of General intelligence in Science Fiction we often equate AI with `human like’ intelligence. In fact, there is no reason to suppose that this is the case and indeed there are many indications that machine intelligence will be quite different to human intelligence.

Further references to AI throughout this guide will refer to Narrow intelligence.

AI methods

AI isn’t a single thing, but a collection of different methods that aim to meet the general objective of performing actions using human or animal-like intelligence. Even which methods are, or are not, AI is disputed. Here are summarised methods that are often included within the AI family.

Symbolic AI

Up until the late 1990s, AI was dominated by an approach now called symbolic AI. Back then it was just AI, a time when both AI and the definition of AI were simpler. Symbolic AI is based on reasoning, typically through the application of a branch of mathematics called first-order logic. This approach was successful in certain areas such as Expert Systems - systems using rules to provide advice and guidance - but suffered from many limitations. It gave much promise but often great disappointment too.

Symbolic AI uses simple statements to provide basic knowledge. For example:

A is within B.

B is within C.

From this a Reasoner, a programme capable of making inference from such statements based on logic, could infer that.

A is also within C.

One problem with Symbolic AI is that it is built on rigid lines and struggles to match a human’s ability to deal flexibly with the complex ways we understand the world. For example, if asked to think about a bird most people will think of something small and feathered, which flies and makes tweet, tweet sounds. None of this is true of all birds but people don’t struggle with this. It is however a major problem for Symbolic AI.

Symbolic AI is by no means dead and gone, it is still used today in many areas. 1 common use is in the development of Ontologies–formal descriptions of topic areas, enabling machines to make more sense of data about those topics. For example, an ontology on the structure of an Army would enable a machine to understand that a Division is made up of Brigades, Battalions of companies and so on.

Other techniques that have been labelled as AI by some include Agent-Based Modelling and Genetic Algorithms.

  • Agent-Based Modelling is a method that creates ‘Agents’–discrete bits of code–that interact with other Agents through a set of rules. This approach is often used in building computer simulations of complex areas.

  • Genetic Algorithms is an approach that seeks to solve problems through an evolutionary approach that simulates the process of natural selection.

The current high

AI has been through several highs and lows and is currently undergoing a high. This high is due to the application of machine learning (ML) techniques which we discuss later in the biscuit book.