In my previous post on Open Digital Platforms, I introduced the five domains of strategy that Digital Transformation is changing, I looked at how developing digital platforms can lead to significant competitive advantage. In this post, I focus on the data domain and look at the role of Artificial Intelligence (AI) and Machine Learning (ML) in turning data into assets, thereby supporting Digital Transformation.
AI and ML, terms once reserved for academia and research, have permeated their way into the lexicon of technology leaders and to some degree, the general public. We now associate them with self-driving (or drive assisted) cars, intelligent personal assistants, smart home systems, futuristic robots from sci-fi movies and the Technical Singularity hypothesis that relates to the creation of a general AI super intelligence.
While the terms AI and ML are widely used, they are generally not widely understood. They are often used interchangeably and I have frequently heard vague definitions for each, especially ML. Therefore, before we look at how they are forming the foundation of the next wave of Digital transformation, I am going to spend some time providing what I hope is a clear and simple introduction to AI, ML and how they are related. This will lay the foundation for a future post, which will look at why and how they are relevant to business transformation and make some predictions about their future application.
According to Wikipedia, AI is a field of computer science that refers to:
"the study of intelligent agents: any device that perceives its environment and takes actions that maximise its chance of success at some goal"
This definition is broad and lends itself to everything from very simple systems to incredibly complex systems being classified as AI; it also suggests that agents should be physical in nature, which isn’t necessary.
To help our understanding, let’s go from the abstract definition to a simple example that I hope most readers can relate to. Let’s take a look at AI through the lens of the classic video game, Pac-man, specifically the behaviour of the ghosts that attempt to capture or avoid Pac-man depending on whether he has eaten a power pellet.
If we relate the definition of AI to the game, the term “environment” refers to the maze, pellets, fruit, Pac-Man and of course the ghosts. By taking an “action” of changing direction, ghosts either move closer to or further away from the “goal” of capturing or evading Pac-Man. As a player or observer of a game, after watching each ghost move around the maze, it’s reasonable to come to the conclusion that their movement is in some way intelligent as opposed to purely random. We can therefore think of the game as possessing some element of AI that is specific to the game.
If however, a ghost were to find itself in a platform game such as Donkey Kong, then the specific AI that worked successfully in Pac-Man would almost certainly not work. For a start, there is no Pac-Man, pellets or maze. This highlights the difference between specific AI and more generalised AI, well at least within the highly restricted realm of computer games.
Today, systems classified as having AI have a limited capability to adapt to a new environment, actions and goals. A system that is able to overcome these limitations is said to possess general AI, an end-state that is a distant goal and has a number of associated concerns as outlined in the Technical Singularity hypothesis.
With a better understanding of AI, let us now look at why a system behaves in an intelligent way; this will introduce us to Machine Learning (ML).
Machine learning is a sub-field of AI that gives:
"computers the ability to learn without being explicitly programmed"
That doesn’t mean that computers do not need to be programmed at all, they still run a program, but in the case of ML, the brain of our AI system can be updated as part of a closed-loop learning system. That’s sounds a bit complicated, so let’s contrast two examples of how systems can possess AI by continuing our Pac-Man theme. In the original game, a ghost’s behaviour would have been programmed based on a set of expert rules, rules such as If Pac-Man’s state is powered-up (he’s eaten a power pellet) move in an unobstructed direction that maximises the distance between the ghost and Pac-Man. While the ghost’s behaviour may change over time, the rules controlling its behaviour do not. An expert defined the rules or created the AI brain that controls the behaviour of the ghost. This is an example of an expert system, another subfield of AI but as we will see is quite different to ML.
If we were to use ML to create a ghost’s AI, it would not require an expert to define a set of rules that determine its behaviour. Instead, we just need to define what the goal or goals are, define a set of actions the ghost can perform (move in each direction) and a means to sense the environment (detect a wall or where PacMan is). Then, using one of a number of different ML algorithms, we can train (develop AI) the Ghost through a series of random mutations of the rules that control its behaviour based on environmental conditions. By promoting the most successful set of rules at each stage of mutation, the ghost's AI will improve over time through a process of natural selection.
In summary, AI is a field of computer science with a focus on creating systems that interact intelligently with their environment, whereas ML is a sub-field of AI focused on developing intelligent systems without the need to explicitly define rules that determine behaviour. ML allows a system to evolve to changing environmental conditions and as we’ll see in a future post, this is critical for a system or service to move from the state of being managed to one of being optimised.