As the crescendo peaks, AI displays a new maturity
One power network participating in the Digital Think Tank is already using AI with “a fair degree of success”, says its CIO. Hype over AI has reached a “crescendo” recently with the advent of generative AI, he adds, but the technology (rule-based or deterministic AI, which uses pre-programmed rules and algorithms to perform specific tasks) has been around for a while.
There is a new maturity: aspects of AI such as understanding language and image recognition outperform baseline homo sapiens. Could AI potentially bankrupt a utility? “I’d like to think not. Something has got to be fundamentally wrong for any type of IT initiative to have that kind of impact.
“It’s fair to say there is risk. But the risk profile is changing, because the veracity of the output of AI is growing over time, and no doubt will continue to do so.” Exploring AI should have a business case, the same as any other initiative, the CIO says. Networks must understand the value proposition, and the return.
Machine learning has been used at this network for more than five years for everything from forecasting consumption and generation to predicting vegetation growth, to analysing where customers have had a poor experience. One of the next uses of artificial intelligence will be to triage 600 or so emails each day in the connections queue using generative AI, automatically processing applications for connections.
In summary, AI is already proving its worth. “We have delivered value [using AI]. We are improving the customer experience, and we are generating internal efficiencies. We are doing so with absolute confidence that we are not creating a problem.”
When it comes to the technology, the CIO recommends:
- Always focusing on value, not hype – be realistic about what can be achieved.
- Starting small with low risk and high-value projects. (One panellist from a gas network at the Digital Think Tank agreed this was important: “We’ve been selective in picking the things that drive the small change, knowing there’s a long game we need to play. If we try and boil the ocean, we’ll never get there.”)
- Rethinking processes. AI inevitably means reengineering.
- Aiming for a minimum viable product rather than a proof of concept.
- Developing the skills and knowledge of the staff that involved.
- Having a strategy with a clear view of the risks and how to mitigate them.
The CIO of a water company on the panel agreed that their spend on AI is in no danger of sending the company down the tubes. The most financial harm he could do to the utility, he says, would involve sitting back and doing nothing. “By that I mean keeping legacy systems around. I am 100% certain that a cybersecurity incident would cost us more than anything we are ever going to do on AI. People often underestimate the risk of doing nothing.”
The innovations they are working on include a smart sewer system featuring a network of 750 sensors to detect level and flow in the sewage system to stop foul water going into rivers. The sensors dynamically calculate hydraulic capacity in pipes while a digital twin controls gates, valves and penstocks to prevent storm overflows. The system potentially enables the capture of more than 280,000 cubic metres of water, he explains.
When it comes to the technology, the CIO recommends:
- Always focusing on value, not hype – be realistic about what can be achieved.
- Starting small with low risk and high-value projects. (One panellist from a gas network at the Digital Think Tank agreed this was important: “We’ve been selective in picking the things that drive the small change, knowing there’s a long game we need to play. If we try and boil the ocean, we’ll never get there.”)
- Rethinking processes. AI inevitably means reengineering.
- Aiming for a minimum viable product rather than a proof of concept.
- Developing the skills and knowledge of the staff that involved.
- Having a strategy with a clear view of the risks and how to mitigate them.
The CIO of a water company on the panel agreed that their spend on AI is in no danger of sending the company down the tubes. The most financial harm he could do to the utility, he says, would involve sitting back and doing nothing. “By that I mean keeping legacy systems around. I am 100% certain that a cybersecurity incident would cost us more than anything we are ever going to do on AI. People often underestimate the risk of doing nothing.”
The innovations they are working on include a smart sewer system featuring a network of 750 sensors to detect level and flow in the sewage system to stop foul water going into rivers. The sensors dynamically calculate hydraulic capacity in pipes while a digital twin controls gates, valves and penstocks to prevent storm overflows. The system potentially enables the capture of more than 280,000 cubic metres of water, he explains.
Far from being a scheme that breaks the bank, this “traffic management” system is “much, much better value for money” than digging storage sites. Another scheme uses machine learning and open data (“a force multiplier for AI”) to attribute sources of pollution in rivers. Another pilot is using generative AI for treatment works optimisation, which creates an operational assistant for new operators to operate the site optimally when it comes to chemical usage. This also helps capture knowledge in the business.
The water company’s core AI principles include:
- Starting with assets, rather than people (“a safer space to learn”, says our CIO).
- Using an ethical approach to AI
- Always keeping a human in the loop.
This last principle is important to mitigate risk. “If you have a black swan event – a once-in-10-year storm coming in – no model is going to be trained for that”. For this reason, completely automating operations that need experienced people is “not something the company is pushing”.
The question of data quality
One Think Tank attendee was keen to find out how the respective CIOs dealt with foundational challenges such as data quality and whether they were dependent on external parties for AI expertise.
The senior figure at the DNO acknowledged that “early on our AI journey we had a high dependence on external resources”. “As we have learnt and developed, we have increased our internal capacity through creating teams for data governance, data science, and platform services,” he says. Meanwhile the company’s data strategy involves increasing accessibility to “erode legacy siloes that exist internally and also with a push to increased open and shared publication of our data”, while increasing data reliability, which impacts overall quality. “How effective and useful is the data for contemporary purposes, bearing in mind that some of the data is decades old?”
Interoperability is also important. “How do we advance integration?” He adds that data is seldom immaculate. “Is there such as thing as perfect data? No, never. And that’s partly what you need to consider when you are implementing AI.” If foundational data is flawed in applications where insights are being generated for customers or employees then you will “quickly run into trouble” without a reliable, clean dataset.
“Our data quality is in a pretty good place, but it is never perfect,” agrees the water company CIO. “We have a long history of working with imperfect data for machine learning and saying, ‘you know what, it’s good enough for us to gain some insight’.
“Don’t wait until your data is perfect would be my top tip!”
