Deciding When to Apply AI – Managing Your Complexity Budget

Integrating AI successfully into an investment organisation requires clear vision on where to apply it and when to stop, State Super CIO says

Artificial intelligence (AI) can be a powerful tool to drive efficiencies and innovation and increasingly more investment organisations are starting to explore the benefits to their business.

At the recent Frontier Advisors Annual Conference in Melbourne, delegates were asked to answer the question of how prepared their organisation was for AI integration at scale. No less than 63 per cent said they were running pilot projects or used AI on a small scale.

About 20 per cent said they were in the planning or exploration stages, while 10 per cent said they had already integrated AI in core processes.

Only four per cent said they had no plans to adopt AI whatsoever.

The results indicate organisations are increasingly attaching greater importance to AI adoption, but deciding where and how to apply it is a large determinant of success, according to Charles Wu, Chief Investment Officer of State Super.

State Super has been recognised as one of the most innovative asset owners in Australia when it comes to AI, but Wu admitted the fund had somewhat of a false start.

“Our journey started around seven years ago,” Wu told delegates at the conference in mid-June.

“If you wind the clock back a little bit, you may recall what happened seven years ago. [We had] negative yields and the time value of money didn’t quite work. What do we do?

“We need something that would help us as a small team … to cover the ground more. We want to develop more insights and distill more insights from the data, so we decided to bring machine learning into the practice.

“One of the things that we learned very quickly is that you need to understand what you can achieve before you jump ahead to do it, because for the next four months we failed miserably and had to roll the program back,” he said.

Having a clear vision of not just where to start, but also where to stop is crucial when it comes to integrating AI, Wu said.

“In our framework, we divide machine learning applications into four categories. Type one is automation, type two is descriptive and then there’s type three, which is predictive. And that’s where we stop. Type four is an automatic decision making process, and that will never happen,” he said.

Having determined the goal of the AI integration and developed the framework for execution, the team became more targeted in its approach to AI.

“That is when we started to engage external parties. We set up the academic oversight party. We got model verification in place,” he said.

It was then that the benefits started to come through.

“It helps us to cut through the numbers and distill insight in the way that we don’t normally have. It helps us to have a better and more meaningful conversation,” Wu said.

Back Office a Popular Target for AI Integration

During the conference, delegates were also asked, which area of their business would be impacted most by AI over the next three years.

The vast majority, 48 per cent, said their back office operations would be affected most, well ahead of portfolio management and trading activities, which received only four per cent of the votes.

But Wu believes State Super’s efforts are best spent on gaining insights into market analysis and forecasting and the team spends much of its time on understanding different signals, trying to capture non-linearities and interim market movements.

“I think it’s great to apply AI to the back office, but setting up that process in itself consumes what we call ‘complexity budget’,” he said.

“You really need to think about where you want to spend that budget, because you might go into a process trying to streamline your back office operations by spending hours and hours without getting the kind of outcome that you want.”

Market research requires thorough analysis and Wu says AI deep research tools can help in distilling vast topics into something understandable in a much shorter time frame than was previously possible, allowing the team to cover a greater number of topics.

Many large language models, such as Gemini or ChatGPT, include deep research prompt tools, which are designed to automate complex, multi-step research tasks. These tools can search the internet, analyse data, and summarise findings into detailed reports.

“For example, if I want to know what happened with the rare minerals, I just have to type in a really long prompt and 10 minutes later 20 pages come out,” he said

“If they’re too scientific, then I need you to go through the prompt [and bring it] down to the level of understanding. But [this research] process would normally take a week. Now, I can do three times the amount of the research within half an hour,” he said.

But Wu also points out there are dangers of relying too much on AI and any answer should be considered a starting point, rather than a fully formed answer.

“The risk of hallucination [is] real. Different large language models all have their pros and cons, so just be mindful of what you’re getting. I think it serves as a really good starting point, but don’t rely on it.”

Investment Opportunities in AI – What Exposures are Best?

Asked about the best way was to access AI investment opportunities over the next one to three years, 29 per cent of delegates said AI driven industries were their key focus. Another 25 per cent said data centres are the best way to approach AI investing, while 24 per cent said investments through technology giants was their preferred route.

Almost nobody voted for hedge funds as the best way to invest in the AI evolution, but Wu has a different view on this topic.

“There are a few things that we are definitely looking at in terms of either AI start-up, AI-driven industry, but why I come back to hedge fund investment at the end of the day [is because] the way we manage the portfolio hinges a lot on how we work with our managers and our partners,” he said.

“We’ve been on this journey for quite some time. Naturally that allows the conversation to be a lot richer in terms of understanding their model. So yes, the last currency manager that we appointed has a neural network model in it,” he said.

His interest in AI start-ups derives from the idea that the world is at the forefront of a number of AI-driven breakthroughs. He pointed to Google’s Deep Mind, whose AI tool GNoME has identified about 380,000 new stable materials, including inorganic crystals that might have applications to battery energy storage systems and superconductors.

“That means a lot of the things are on the verge of being changed, so I do think there’s a great opportunity there. Do we have an edge in it? No, but we’ll try to develop something. And that’s kind of how I think about investments for the next one to three years,” he said.

__________

[i3] Insights is the official educational bulletin of the Investment Innovation Institute [i3]. It covers major trends and innovations in institutional investing, providing independent and thought-provoking content about pension funds, insurance companies and sovereign wealth funds across the globe.