Demystifying artificial intelligence: A primer for business decision makers

While there is no denying the opportunities that artificial intelligence can bring to businesses, it is important to separate the hype from the help.

While there is no denying the opportunities that artificial intelligence can bring to businesses, it is important to separate the hype from the help.

Artificial intelligence, or AI, has been one of the favourite buzzwords in enterprise technology for a few years now. Often hailed as a solution for enabling huge efficiency gains, even to the point of rendering entire industries obsolete it has, on many occasions, been hard to separate the hype from the help. 

For the most part, this is down to the fact that the term has become so broadly applied and overused that it often ends up losing its true meaning. This has led to many misconceptions and speculations about AI, which can be difficult to separate from the legitimate concerns and opportunities.

That is why, in this guide, we seek to explain what AI is in the context of modern business – and what it is not. 

What is artificial intelligence?

The term artificial intelligence is itself something of a misnomer. In fact, writer and researcher at Microsoft Kate Crawford goes so far to claim that artificial intelligence is neither artificial nor intelligent. Firstly, it draws enormously from the natural resources and human labour needed to build the necessary computing capacity and software. Secondly, it has no ability to discern things without extensive human training. To that end, it has no true intelligence of its own.

There have been several attempts to describe artificial intelligence and its evolution. Swedish philosopher Nick Bostrom defines three levels: Artificial narrow intelligence (ANI) refers to the solutions that are widely in place today to carry out repeatable tasks autonomously. Artificial general intelligence (AGI) refers to AI models that can learn, perceive, and function much like a human being. Artificial superintelligence (ASI) marks the pinnacle of AI research, at which point humans themselves become practically obsolete – or in other words, a technological singularity. Of course, AGI and ASI still belong in the realms of science fiction.

Further adding to the confusion is the many related terms that are often used interchangeably or confused, such as machine learning, computer vision, neural networks, and deep learning. All these are, however, subsets of artificial intelligence. For example, computer vision refers to a computer’s ability to process and make sense of visual data, such as images and videos. Deep learning relies on a constant stream of data to train an algorithm to perform a given function. Both areas of AI play a fundamental role in innovations like self-driving vehicles.

What artificial intelligence is not, at least in its existing or near-future forms, is a replacement for human expertise and decision-making. Rather, it is a way to process and make sense of data at massive scale and at speed to augment people’s capabilities. In other words, it makes people perform their jobs better. While automation will reduce the need for manual labour in some cases, just as technological innovations have always done, it is also believed that AI will create more jobs than it displaces.

Artificial intelligence in the service delivery chain

Contrary to the hype, artificial intelligence, according to its popular definitions, is not actually a new technology. In fact, it is not really a technology at all, but rather a subset of computer science that focuses on using algorithms to automate tasks by emulating the intelligent traits of the human mind. To that end, AI is an evolution of well-established numerical techniques like probability, statistics, and calculus, albeit adapted to the scale of today’s compute power capabilities.

As is the case with any computing system, the primary goal of AI is to automate repeatable routine processes. However, AI goes further than older solutions to emulate intellectual tasks like problem-solving, decision-making, and even making sense of human communication. This is important throughout the entire service delivery chain, where AI is used in everything from personalised advertising to handling customer support queries to processing invoices. AI lets businesses carry out such processes at machine speed by transforming raw data into insights that drive decision-making.

AI in scalability and automation

The need for computers was first realised in the 40s in data-heavy industry sectors like finance and government operations, where the sheer volume of records became practically impossible to manage manually. Today’s data sets are many orders of magnitude larger, and the amount of data businesses handle continues to double every two years. Moreover, as populations and consumerism increase, demand has long reached the point where routine business operations must be automated to meet the challenges of scale. AI makes this possible, since anyone with the same code can replicate tasks and workflows and automate decision-making where the necessary data is available.

AI beyond the IT department 

In the old days, innovations in IT were entirely handled by a dedicated IT department. In recent years, the lines between business and IT have blurred to the point that all knowledge workers must collaborate across computing environments. After all, AI offers substantial value across the entire service delivery chain, even though it might not be explicitly understood by anyone outside the IT department. For example, AI can automate low-level decision-making in many routine business workflows, such as supply and demand forecasting, personalised marketing, procurement and supply chain management, and more. By analysing data at massive scale, it can turn that raw data into actionable, visualised insights that almost anyone can understand.

Understanding the limitations of AI

Most notable among the limitations of AI is that it cannot be used without extensive data to train the model in the first place. Moreover, that data must be properly prepared to feed into the AI model to ensure accurate decision-making and reduce the risk of concept drift. More advanced AI models are trained to determine if and when they need more data. For example, if a model exhibits a low confidence factor when trying to identify the contents of an image, it might flag it for manual review. As such, AI models must be continuously trained and refreshed to retain their relevance and usefulness in constantly changing circumstances. This is why the human role is extremely important.

The human role in artificial intelligence

The most important thing to remember about AI is that a model will only ever be as effective as the people tasked with training it. Keeping humans in the loop throughout both initial training and iterative algorithm updates thereafter is vital for maintaining its usefulness, otherwise the model will become inaccurate. It is often said that computers do not make mistakes, but this is not really true. For example, much like a poor teacher might teach the wrong information or bad habits to a student, so too can insufficiently trained people teach an AI to carry out a task poorly. In the end, an AI can only do what it is programmed to do based on the data it is provided and the specific rules defined to harness it. 

Collecting and preparing data

Collecting data is easy enough, and many organisations have an ample supply already. Every digital touchpoint, from websites to self-service kiosks to internet-connected smart devices, collects a trail of data. Most data is unstructured, which means it has not been organised in a database format that a computer can easily make sense of. For AI to deliver on its promises, it must be trained using structured data, at which point it will be able to make sense of unstructured data like images, videos, social media posts, and email content. Data preparation for AI is where the bulk of the hard work is, yet it is an area that is often overlooked. 

Understanding key performance indicators

In the context of AI, key performance indicators, or KPIs, measure the performance of a model. It is essential that business decision makers clearly define their KPIs before deploying any AI or other solution. For example, KPIs in customer service and support might include the volume of support tickets, mean team to resolution (MTTR), and total cost of ownership (TCO). Other KPIs, such as customer experience, are a little harder to measure, and they typically need to draw from other metrics to provide meaningful qualitative or quantitative insights. Let us say the customer support team wants to leverage AI to reduce ticket volumes. In this case, the AI will need to be trained to help customers resolve common problems without needing human intervention.  

The need for technology alignment

For AI to deliver, people cannot be left behind. Furthermore, alignment with business goals is vital for a respectable return on investment. Although there may be some value in deploying a generalist AI model in certain cases, the best systems draw from the unique environment of a given organisation. After all, every organisation’s data sets are unique, which is why the best AI models are trained using that data. For example, if a business wants to deploy AI for supply and demand forecasting, then the model will only be effective if it is trained using the historic data specific to that business. In other words, AI lives in the business, not the IT department.