top of page

AI, ML, GenAI, LLMs – what’s the difference?

Find out the difference between AI, Machine Learning, GenAI and Large Language Models

AI and Machine Learning in Telecoms

It seems like everyone is talking about AI these days. Jonathan Abrahamson, Chief Product and Digital Officer at Deutsche Telekom, in an article for TM Forum Inform, explained the huge impact it was having on the telecoms market.

“It’s difficult to envisage a world where this doesn’t change everything and certainly in the context of what we do as a telecommunications company,” he said.

But confusion still abounds about what Artificial Intelligence (AI) is exactly, what it isn’t, and how it’s different to Machine Learning (ML). Since Subtonomy uses ML / AI, we thought it was time to clearly explain what these are and how we use them in telecoms customer support.


The right type of AI

Part of the confusion surrounding AI is that the term is used both generically and specifically, and appended to lots of different things. It’s particularly confusing in the customer support domain because, as analyst Teresa Cottam explained in her recent report for the TM Forum on AI in customer service, “Digital customer service is being transformed by the application of multiple types of AI which work hand in hand”.

Let’s look at a few of these different types to see if we can make things clearer.


  • Artificial intelligence (AI)-

is the category term for a new field of technologies that are creating more intelligent machines or software.


  • Machine learning (ML) –

is a type of AI that uses statistical algorithms to learn from data in order to make predictions or decisions, without being explicitly programmed.


  • Natural language processing (NLP) –

enables systems to understand and capture information from (unstructured) human speech, whether it’s spoken or written.


  • Generative AI (GenAI) –

is a form of AI which generalizes patterns from large datasets to produce new content that looks and feels like it was created by a human.


  • Conversational AI –

a form of AI that imitates human interactions, recognizes speech and text inputs, and translates their meanings across different languages.


  • Predictive AI –

uses algorithms to anticipate likely future events based on historical data to assist with data-driven decision-making.


And LLMs?

Large language models (LLMs) are a type of GenAI that specialize in text generation. Generative AI itself is a category that encompasses a far wider scope of content generation capabilities beyond just text, such as images, audio, computer code, and more. An LLM works by analyzing vast amounts of online articles and books in order to understand how language is constructed and used. It then uses this insight to create new text-based content. In contrast, image-making tools – another type of GenAI - detect recurring patterns in the works of painters or photographers to create images in the same style.


How does GenAI work?

GenAI uses two main techniques – machine learning and deep learning – within two core components. The first of these is known as a ‘generator’ and the second as a ‘discriminator’, which work together in a system called a Generative Adversarial Network (GAN).


  • The generator creates new, artificial content based on its analysis of a large dataset.

  • The discriminator examines the generator’s output and determines whether it resembles the original data closely enough.

  • The generator tries to improve its creations based on the feedback it gets from the discriminator.


This feedback-loop way of learning, mimicks how humans learn and results in an ongoing process of refinement that produces higher-quality, more convincing, artificial content.


GenAI learns to do this by adjusting settings called ‘parameters’. The more parameters the model has, the more detailed and specific the output will be. That’s what the ‘large’ refers to in ‘large language model’ – where the model has millions or even billions of parameters to generate outputs. This process of adjusting parameters through discovering patterns in datasets is called ‘training the model’. And this is why the quality of the data set matters a lot. Without high-quality data, the model might be trained to become specific, but still not be accurate.


Want to know how these techniques are being used in customer support? Check out our next post.


Tags:

bottom of page