Any science fiction movie buff will understand the difference between artificial intelligence (AI) and machine learning, given one simple analogy, State Super asset allocation general manager Charles Wu says.
“AI is a bit like the Terminator; machine learning is more The Edge of Tomorrow,” the computer engineer says.
Those who have seen Tom Cruise’s science fiction epic, which throws his character into a time loop where he relives brutal battles and death over and over, will know machine learning is about reinforcement of learning or “doing the same thing again and again until you get it”, Wu says.
In the two years since Wu and his five-strong team set up in-house machine learning to better inform the fund’s investment decisions, the process has revealed “pros and cons”. While throwing up some spurious correlations, the data crunching also has provided a starting point for the fund to do more research.
“In everything we’ve done, machine learning has supported us in taking on more risk and that’s been working well for us,” he says.
Algorithms created to crunch nine years of “structured” or numbered data have helped distil information that has helped the fund continue to buy more equities despite geopolitical noise in markets, Wu says.
State Super has been an early adopter of the technology, partly due to necessity. The fund has been cashflow negative for the last 10 years, with half its members in retirement and the other half nearing retirement. State Super applies machine learning as a management component to its DC portion of the $43 billion fund.
“State Super has shorter investment horizons and negative cash flows,” Wu says. “At other super funds that have positive cash flows, you can average down or average up; for us, we don’t have the luxury…when the market draws down, we can realise our loss at the worst possible time.”
Because of State Super’s size, it has not been in the fund’s interest to trade the market, so it has tended to take bigger, high-conviction positions and hold them for longer, making it particularly important to settle on the right investment decisions.
“(Machine learning) helps us take emotion out of the decision and that has been very important over the last two years,’’ Wu says.
While other funds have used machine learning for member communication, insurance and legal claims, State Super will continue using it to make investment decisions, albeit in tandem with its staff.
“In terms of the investment process, it’s hard to see machine learning taking a bigger role, because it is an algorithm in a black box and, in light of the [Hayne royal commission], what is needed is more transparency,” Wu says. “When you have a black box, it should be only part of the picture. We still prefer to sit back and look at what machine learning is providing and also what our smart people are telling us.
“Machine learning is part of the process as long as we understand the pros and cons.”
Charles Wu spoke to Investment Magazine ahead of his session “Machine Learning: Deep Learning or Spurious Correlations?” at the Absolute Returns Conference, to be held in Sydney, September 19-20.
