The terms, AI and automation are sometimes used interchangeably. It’s certainly the case that a lot of AI-based applications are designed to automate manual tasks. However, not all automation relies on AI.

Take a corporate accounting package, for example. Let’s say you can programme it to follow certain predetermined instructions – e.g. to extract a narrow range of data points to make certain calculations and populate a report. This is automation – but it isn’t necessarily AI.

Contrast this with more advanced office of finance functionality. With this, the software has the ability to make predictions and decisions based on the data it is presented with. The solution isn’t just following narrow, pre-defined rules; it’s making correlations and uncovering insights that are often impossible to pick up on through human intervention alone.

Machine learning is the key concept that makes the difference between automation and actual AI. With this, the machine isn’t just reliant on explicit programming instruction. Rather, it relies on algorithms to make predictions and recommendations by processing data. The more data the machine processes, the more accurate it becomes.

Particularly relevant in areas such as performance management and sales forecasting, machine learning takes you beyond descriptive analytics (describing what just happened), and opens up the possibility of predictive and prescriptive analysis: anticipating what will happen next, and providing recommendations on what to do.

One step up from machine learning, deep learning gives you the potential to process a much wider range of data sources for more complex use cases. It is based on multiple, interconnected layers of calculators known as neurons, forming a neural network. Vast amounts of data can be fed into the machine, and processed through those layers. The network can then make a determination about the data, learn if this determination is correct, and then use this knowledge to inform future determinations. Use cases include image classification (e.g. visual diagnostics), facial and voice recognition.

Think about all of the data generated across your sales and customer service channels, operations, finance, and other key areas of your business. Much of this falls into the category of Big Data: i.e. large, diverse sets of data generated through the usage of devices, software, and networks.

With AI, the more data you feed into an application, the more accurate the solution becomes. As such, the success of any AI initiative is always going to be intrinsically linked to your ability to harness and process the right data, at the right time. Millennium Consulting’s Big Data White Paper explores this in more detail.

Generative AI – i.e. the ability of a machine to generate new output based on training data – isn’t new. However, when ChatGPT’s latest model, GPT-4 arrived last year, it was instantly recognised as a big deal. Previous generative AI models could produce serviceable – if not always very impressive – output. GPT-4 stepped things up several levels.

Like its immediate predecessors, the solution is trained through a neural network utilising massive amounts of data, combined with powerful natural language processing (NLP) capabilities. However, the latest version is much more effective at completing tasks that demand advanced reasoning or creativity, including the ability to understand and respond to nuances of conversation. It can also comprehend user intention more accurately, it allows users to adjust tone and style depending on needs, and is also a lot better at avoiding ‘hallucinations’ and nonsensical output.

In terms of practical application, generative AI is no longer confined to providing basic responses to simple requests. It can troubleshoot, advise, diagnose, solve complex problems – and even come up with new ideas: potentially disrupting a vast range of not just routine activities, but also senior and specialist job roles.