The right tool for the job

Without data, a company does not exist. This was one of the answers Lotar Schin, AI centre of excellence lead at Hungarian commercial bank, OTP Bank made during a panel discussion at the recent Gartner Data & Analytics Summit in London.

He was one of three panel members at a session run by Cloudera, exploring the challenges moving AI from pilot to production.

Computer Weekly has had a number of conversations with Leinar Ramos, senior director analyst at Gartner, on the difficulties enterprises face delivering business value from AI initiatives. It is relatively easy to get a pilot project running. But research from the analyst firm has found that almost half (49%) of the organisations it polled say that their biggest challenge is estimating and delivering value with AI-based initiatives.

Ramos recently produced a Gartner paper looking at when not to use generative AI (Gen AI), which, for many, seems like a silver bullet that can solve all manner of business problems. But it has clearly not been designed to run many of the tasks some people are trying to do with it.

There is an awful lot of industry hype around AI and Gen AI. But no one should be swayed by the cleverness of OpenAI’s latest GPT-4o to sound human-like. Clearly it is a breakthrough and exceeds the capabilities of older technologies like Alexa, Google Assistant and Siri. However, an AI is only as good as the data it has been trained on. If something never existed before, there will be no historical data on which an AI can draw upon to generate an answer.

Hallucinations and dreams

While AI researchers use the word “hallucination” to describe rogue AI responses, as far as we know, AI systems do not dream up original ideas. Consider  the world before the Covid-19 pandemic. In a recent conversation with Computer Weekly, Grant Farhall, chief product officer at Getty Images and iStock points out that overnight, everything about how we see the world changed. Prior to the pandemic, it is hard to imagine how an AI trained on pre-pandemic data, could have generated images that reflected society under lockdown measures.

What these conversations and discussions illustrate is that we should not get swayed by AI hype. There are plenty of sound and mature techniques that can and should be deployed to get value from enterprise data.

“If you can solve the problem with the simplest tool, then use it,” is Schim’s advice.

And if this happens to be an AI tool, it needs to be laser-focused on helping us do our jobs. Farhall urges people to “build muscle memory” when using such tools, to ensure they complement they way we work.