It sounds counterintuitive, but artificial intelligence (AI) is nothing new. From research labs in the 1950s to businesses during the 1980s, machines have been simulating and augmenting human decision-making for decades.
Given this, it is perhaps a surprise that AI has captured the world’s attention so thoroughly in recent years. Then again, perhaps not: what is new today is not only the appearance of a new technology in the form of generative AI, but a wide array of practical applications for AI.
Brian Finnegan, digital solutions architect at Presidio, said the early history of AI had seen plenty of bumps in the road.
“AI has been around for years, since the 1950s, but back then it was very bespoke. The early experiments were built around hardware; one of the early examples was simulating rats in a maze. It was limited to universities and research agencies like DARPA,” he said.
Next came experiments like the simulated psychotherapist Eliza, followed by robots capable of sensing their environment. But things started to stall in the 1980s.
“There was what Gartner would call the ‘trough of disillusionment’,” he said.
Indeed, so bad was AI’s reputation for over-promising and under-delivering that companies with AI-based products ditched the term for euphemisms such as ‘expert systems’.
However, despite the disillusionment, practical applications for what we now recognise as AI did exist.
“The labelling can be important,” Finnegan said. “Expert systems were used for things like credit sorting in the 1980s.”
A lot has changed since then, but one thing has remained constant: more and better data means a greater ability to make decisions.
“The more data you have the better you can do it, with the extreme being the current generation of large language models,” he said.
This can be seen in action with the recommendation engines used by the likes of Amazon and Spotify, which can look at both an individuals’ selection history and also wider trends.
Indeed, contact centres have been quick to realise the benefits of augmenting staff with AI and the sector continues to invest in this in order to reduce wait times and make things easier for both staff and customers.
“Around the early 21st century, what we saw in the call centre business, what we’d now call the contact centre business, was that the early chatbots and robotic process automation came in.
“There is a lot more data now, and it’s not only voice recording. There’s also customer data and social media sentiment analysis,” he said.
Across industries, the ability to collect, store and process data has improved.
What Presidio increasingly does is work with clients, ranging from mid-market to enterprise, to devise and implement data and AI strategies. This is crucial, as any plan to use AI is utterly dependent on the quality and accessibility of data, Finnegan said.
“The quality of data varies. I was talking to some customers recently who were looking to see what shape it was in, in the life sciences, and they were pleasantly surprised. They had half a dozen systems that were key, but were expecting to have a dozen to 20. Then you have other organisations that just don’t know how many critical systems they have,” he said.
“If your data is not in good shape you’re not going to get much out of ML [machine learning] or AI.”
First steps include identifying the system of record for data, while another question might be which systems will be more important in the future.
“We see a lot of work around getting the data out of the systems. There’s a lot of skill needed there, not just on the data side but also on the business problems.
Then you have the issue of machine learning operations: you don’t just get your model and let it run, you have to monitor it. In fact, productionising machine learning is very difficult, even if producing a pilot is very easy,” he said.
Finnegan said that Presidio’s goal was not to throw technology at companies but to show them how they can benefit, and that means making sense of their data.
“There are plenty of vendors out there who will say you need an ‘x’ to solve a data problem. The answer there is we already have an ‘x’. You need to ask: is throwing more technology at the problem going to solve it? If you can’t get any insights out of your data today, there is no point in thinking you can leap straight to machine learning or generative AI.”