Weak Gen AI use cases: A Computer Weekly Downtime Upload podcast

The distinction between AI in general and GenAI (generative AI) appears to be blurred, and with the huge drive by the industry to push out AI-enabled products and services, there is a risk that IT and business decision-makers find they deploy GenAI in wholly unsuitable application areas.

Leinar Ramos is a senior director analyst at Gartner. The analyst firm has been looking at whether the use of AI materially changes the performance of the use cases business leaders identify when they begin an AI project.

Discussing the immense hype surrounding AI, Ramos says: “There are a lot of shiny objects in the world of AI and a lot of hype particularly around generative AI, which is at the top of the Gartner hype cycle.”

He believes organisations looking at AI initiatives are too narrowly focused on generative AI. “I hear this in my conversations with clients - not just from the IT side, but also from the business,” he says.

Ramos believes many organisations equate AI with generative AI. This is a problem, according to Ramos. “If you use generative AI for the wrong use cases, then you're way more likely to fail,” he warns.

Another risk, he says, is missing out on the many opportunities the AI space presents, if the focus is only on GenAI. “GenAI is really just one tool in the toolbox and the reality is that we have all kinds of different techniques that really have nothing to do with Generative AI.” For instance, graphs, simulation and optimisation and machine learning should be considered when looking at AI initiatives. “But if we only look at all of this through a generative AI lens, then we get distracted and we lose focus. That's what we see happening,” he adds.

Broadly speaking the use cases for GenAI - such as image, music, video and text generation - are all trained in a specific way to learn from a dataset and generate new data. This, he says, is a little bit different to the type of machine learning where a model is trained to recognise patterns in data then used to make a prediction.

“Contrary to to some of the other AI techniques, with GenAI, typically organisations are not really training these models from scratch.” He says the models are usually trained with lots of data. Then the business customises the models to meet its requirements.

Within enterprises, Ramos sees the big opportunity for GenAI is in knowledge discovery, giving employees the ability to ask questions using internal document repositories.  However, Ramos warned: “GenAI is not a silver bullet. There are lots of use cases for which it is really not a good fit.”

For instance, decision management, says Ramos, typically needs to be an explainable process. “As we know, GenAI models are not very reliable; they hallucinate. I get very concerned when I hear GenAI being used in HR to read CVs. It’s very risky to use GenAI for critical decisions. They really haven't been trained or designed for that sort of explicit decision making where we need to carefully balance things out.”

Rather than thinking of how GenAI could be deployed, he believes IT leaders should look at supervised machine learning, which can be more explicitly designed in rule-based systems, where there is a clear decision flow.