Traditional AI is well-suited to tasks such as organising data in spreadsheets or automatically tagging photos. Its strengths, compared with GenAI, lie in consistency, auditability and cost-effectiveness, making it ideal for repetitive tasks that require close human oversight. However, it tends to be limited in scope and flexibility, Teather says.

On the other hand, GenAI, despite its versatility, can struggle with consistency and accuracy, sometimes generating convincing but false information, known as hallucinations. Critically, the huge costs of running GenAI tools also pose challenges for large-scale use.

Amy Rushby is Co-Founder and Director of product at fintech company Carmoola, which leverages AI primarily for operational efficiency improvements and to deliver a better customer experience. Natural language processing (NLP) powers Carmoola’s customer support chatbots and automates parts of its customer communications. It enables the company to handle routine enquiries promptly. However, Rushby says that 'because language is nuanced, humans need to intervene when responses fall short of delivering the empathy and context needed in more sensitive conversations'.

Even if we don’t realise it, NLP is already part of everyday life, powering search engines and prompting chatbots for customer service with spoken commands, voice-operated GPS systems and question-answering digital assistants on smartphones such as Amazon’s Alexa and Apple’s Siri.

NLP is a subfield of computer science and AI that uses ML to enable computers to understand and communicate with human language. It combines computational linguistics of the rule-based modelling of human language, together with statistical modelling, ML and deep learning (DL).

ML AI systems are capable of self-improvement through experience, without direct programming, and recognise speech and images, as well as forecast market trends. DL is a subset of ML and involves many layers of neural networks. It is often the technology behind voice control in consumer devices.

However efficient these tools may appear; AI is only as good as the data it is trained on. The information we give AI programmes is the only way they can learn so if inaccurate data is entered, then the tool will produce incorrect outputs.

“When using any type of AI, a key consideration is recognising where human oversight is necessary,” Rushby says. “For instance, AI models can sometimes produce unexpected or biased results if they are trained on incomplete or skewed data. That’s why we ensure there is always a person in the loop to validate outputs and make final decisions when it comes to customer interactions and critical risk assessments.”