Kathy Gibson reports from InterSystems Summit in Sandton – The true impact of artificial intelligence (AI) could be both more and less than organisations expect.
This is the word from Jonathan Tullett, research director of IDC, who explains that an AI project could be as simple as enhancing an existing business process, making it a little faster or more consistent.
“This is where a lot of the use cases are now,” he says. “And this is fine, because it is helping businesses to improve processes.”
The second phase of AI deployment moves to building with the future in mind, Tullett says.
“Organisations can use AI to figure out the processes, analyse the data, and use that data to redefine the processes in a way that is compliant and with minimum disruption.”
The ultimate goal today is to use AI to design entirely new processes.
But it is a journey, Tullett emphasises. “Focus on what you can control today, then build for more impact in the future.
“But don’t expect a silver bullet: you have to go through the processes.
“At the same time, don’t underestimate the future either – this is transformative.”
Allan Macfarlane, CEO of Mammoth, believes that AI use today is largely focused on removing complexity.
“Typically we do a lot of work to get to the place where we can make an informed decision. But with AI we can take large amounts of information, distill it down and get to being able to make a decision quite quickly.
“But we need to manage it,” he cautions. “We cannot defer decisions and outputs to the models. We should be using it in a supportive way to expedite processes, with boundaries set.”
Shaun Havelaar, chief AI officer of Q Link, agrees that AI is great accelerant.
“People always think about the most complex use cases. But we have all seen this movie before, and we know that we have to build teams to build and manage the technology, also to figure out guardrails, compliance, plugins, skills and more.”
Macfarlane adds that the biggest issue his organisation has had to contend with in AI deployment is keeping the boundaries in place.
“When agents have context they are more useful.”
There is still a degree of discomfort surrounding AI agents, Tullett adds.
“A lot of this is because we don’t trust them enough – and some of this is well-founded, and we have already seen governance and security failures.
“Building both internal and external trust comes back to a lack of governance or a lack of understanding.
“And if it’s a lack of understanding, that’s down to poor AI enablement when it comes to culture.
“It has been too easy for us to focus on the tech for AI”
However, there are areas where it would be best to remain uncomfortable, Tullett adds.
“We should not give up authority or responsibility to an agent,” he says. “You are not the child in the relationship: the agent is acting on your behalf but you are responsible.
“If you find yourself giving up responsibility, this is a sign that you should be uncomfortable.”
Havelaar, who has a background in cybersecurity, says he gets uncomfortable with what some people allow AI to do.
He cites instances of users downloading skills from public marketplaces and sometimes inadvertently introducing malware.
“The large language model (LLM) itself is not a problem: what is an issue is a lack of awareness and of guardrails.”
On the positive side, AI is quickly democratising technology – but that brings in new challenges around how companies can differentiate their own offerings.
“When customers realise you are leaning on AI to automate what you do, they probably won’t want to pay up,” says Tullett.
“What you need to do rather is use it to build a better product. And you should always be better than AI.”