Machine Learning vs Generative AI: what’s the difference?

But before that, we were told that ‘vanilla’ AI, or regular old machine learning, was also going to change the world.

"In summary, traditional machine learning models are more ‘investigative’ by nature and gen-AI models more ‘creative’. It is this creative ability of LLMs that has caught the imagination of users – and businesses – all around the world.”

What is the difference this time?

Given the mind-boggling uptake in usage of the former, it appears the generative evolution is going to be what makes the technology stick.

So, what is it about the ‘gen’ part of gen-AI that’s so convincing and why are so many companies investing in the technology?

One of the reasons for the rapid adoption is because many experts believe gen-AI is a foundational technology, like the internet, television or the wheel.

A foundational technology is any technology – a tool or a product or a service – with a broad range of applicability across industries, geographies and demographics.

These technologies impact large sections of society and have the power to change the fundamental construct of how things work and what we believed to be true.

AI has been around for over six decades, slowly working its way to this point. For much of this time it was primarily referred to as ‘machine learning’.

Whenever you use a fingerprint to log in or voice prompt to complete a transaction, that is AI at play.

In our daily lives there are many machine-learning models enhancing efficiency, improving experiences and automating processes.

But there are some clear areas where the abilities of gen-AI and ‘old’ AI distinctly diverge. Here are just four.

Creativity and originality

Gen-AI tools have given users the ability to generate entirely new data, like text, code, images, videos (the latest from Open AI-SORA) and even music.

This makes it ideal for tasks like writing, designing new products and developing customisable marketing campaigns specific to a customer or customer group.

Traditional machine learning relies on analysing pre-defined data parameters to make predictions or classifications. These older models work on data patterns and inferences based on those defined parameters – but they cannot generate new content.

Additionally, the context under which these traditional structures operate is very clearly defined and more focused on specific tasks. They are excellent where predictions are required based on input data.

Data efficiency

While this is an emerging field of research, it has been found large-language model (LLM) gen-AI can learn from smaller datasets, something traditional machine learning cannot do.

The underlying ‘transformer’ technology is fundamentally different to traditional models in the way that makes it quicker and efficient.

The tech can predict the most ‘probable’ next word through a complex algorithm that determines association of words (pixels, voice samples etc) with different contexts and assigns weights to each association.

Traditional machine learning can often require large amounts of high-quality data to be trained effectively and accurately.

The expected outcomes from this older model are clearly-defined in a way that makes them narrow in most cases – like an algorithm that produces the same output with the machine going through the same set of comparison steps. It can be expensive and time-consuming to collect, cleanse and label large volumes of data.

Explainability and transparency

Research is still underway, but gen-AI models are relatively less complex than traditional machine-learning ones, making them easier to understand and interpret.

This is important for building trust and ensuring AI systems are not making biased or unfair decisions. Although this heavily depends on the complexity of the use case and the models being used.

Traditional machine learning is often complex and opaque, making it difficult to understand how the models arrive at their decisions. The technology can require complex mathematical understanding to interpret and understand, giving rise to concerns about fairness, bias and accountability.

Adaptability and generalisation

Perhaps one of the biggest strengths of gen-AI LLMs is the ability of these models to adapt and be generalised in a way that caters to a wide variety of use cases.

’Old’ models can be fragile and limiting, meaning they may perform well on data they were trained with, but any application in different contexts or change in scope inevitably requires retraining.

In summary, traditional machine learning models are more ‘investigative’ by nature and gen-AI models more ‘creative’. It is this creative ability of LLMs that has caught the imagination of users – and businesses – all around the world.

From generating new text-based content to pictures and videos, the technical capability, ease of use and generic nature of LLMs are key reasons gen-AI is the new buzz around town.

Madhujith Venkatakrishna is Associate Director, Innovation at ANZ Institutional