Artificial Intelligence or AI is a phrase near the top of many business planning discussions. 

There are numerous urgent articles and entire books written on the need to embrace change management and introduce Artificial Intelligence (AI) into firms before the competition renders your existing business obsolete.  While embracing AI is important, it is imperative to have a realistic view of AI and its potential uses.  Over this and the next few blogs we will attempt to introduce some of the theoretical and more useful aspects of AI with some very technical and some less technical papers.  Much of the technical discussion will focus on successful deployment, a much over-looked aspect of many products.

At its core, at the very basic level, AI may be described as the use of statistics to guess the most probable answer to a question.  A useful addition to this definition is the use of natural language and human vision to make human interaction with the computers doing the statistical calculations simple and intuitive.  Although we will delve into a deeper discussion of the “I” of AI in our blog on Artificial General Intelligence, for now this is a workable definition for the AI that most people will encounter. 

To start, we can divide AI into two broad categories: robotics and Machine Learning.  Although they may overlap, we will not discuss symbolic learning, and instead focus on the machine learning.  All Machine Learning tool require data.  Lots and lots of data.  We can further subdivide this into four narrower areas: pattern recognition, speech recognition, computer vision and natural language processing.  We will talk more about these in future blogs but briefly we can help unpack how these are processed through one or two examples for each area.

Let us say you have a supply chain and want to optimize your gross margins.  You might start by analysing a particular supplier and its stock levels using some simple statistics (such as standard deviation).  For a one supplier or two or three you may be able to do this manually with a spread sheet.  But for multiple suppliers with factors such as weather and financial stability and aggregate demand thrown in, a machine is a much better and faster tool.  Using a machine, you can take in loads of data, even data such as the number of cars parked in a competitor’s parking lot and using some machine learning find otherwise hidden patterns in the data.  Patterns like these can help you determine the stock levels needed to keep your production on target.

This might have been enough for most people, but dedicated and creative scientist decided to take this further and invented computer vision.  A computer vision program does something quite clever.  It harnesses the speed and power of computers to look at every single pixel in an image and using the pixel brightness and colors determine if that pixel is part of an image or not.  Classical techniques were insufficient to do this work, so programmers invented something called a Neural Network or Deep Learning.  Deep learning replicates a simplified version of the building blocks of an animal’s nervous system – neurons.  Each neuron processes some information and when linked together become a powerful analytic tool. 

If you give a neural network with many labelled examples of data, it will be able to extract common patterns between those examples and transform it into a mathematical equation that will help classify future pieces of information.  In other words, if we show it loads of pictures of cats, the program will be able to use statistics to determine if a new picture is probably likely or unlikely to be a cat.  At present, we can identify an object among many images and even identifies objects within a particular image, but we cannot yet describe what that image is doing in language.