Transparency concerns linger around artificial intelligence despite proposed industry benefits

Speakers at this year’s TradeTech conference were keen to extol the virtues of artificial intelligence and machine learning, but perceptions around transparency and trust must be addressed.

Artificial intelligence (AI) has become one of the key watchwords in asset management technology in recent years, with institutional and boutique firms alike having made significant investments in this space.

Speakers at this year’s TradeTech conference were also keen to talk up how AI and its various subsets can optimise trading processes and its theoretical benefits to the wider industry once further development efforts have been made.

Vanguard’s head of investment operations for the EMEA region, Sean Kennedy, outlined the various forms that AI-based technology can take, including sub components such as machine learning, robotic process automation (RPA) and deep learning.

“The real value we have seen in application in other industries has been to look across functional areas of the full life cycle, in our case trading. So we are spending time at the moment looking at the ways to apply machine learning to optimise the entire life cycle,” he said.

“What we are seeing at Vanguard and also the whole industry is that what have traditionally been middle office functions are becoming more closely integrated with the front office, so those lines are becoming very blurred. The opportunity there around new technologies is to start weaving them together through machine learning to drive optimisation.” 

Sanoke Viswanathan, chief administrative officer at JP Morgan CIB, highlighted the institution’s use of natural language processing – the application of computational techniques to the analysis and synthesis of natural language and speech – in its research space, for functions such as sentiment analysis and news analytics.

“Probably the emerging area where there is a lot of time spent but not a lot of yield is what we call auto-decision making; robo-trading or robo-hedging, coming up with automatic in ways in which to answer client queries and the like. That’s an application taxonomy that I find resonates well with end users because that’s the way people can decide how they want to deploy these techniques,” Viswanathan said.

However, despite the proven and potential benefits that AI and its technology subsets can offer, there are still areas of concern that must be addressed before the industry can be completely comfortable with the technology.

“We are not satisfied with the level of fundamental research in AI that is focused on financial markets. In discussion with clients, on the types of issues we are dealing with, there isn’t enough core research going on in areas such as market simulations, time-series predictions and things like that. So we want to set up a research capability that is focused on that,” Viswanathan said.

Kennedy highlighted one of the main obstacles to the further adoption of AI technologies as a lack of transparency and trust in how these systems operate, highlighting the difference between shallow and deep learning techniques.

“Shallow learning is essentially creating small models or computations that you can go back and review, or even watch in real-time to see the input and output, and essentially justify the output through being transparent,” said Kennedy.

“In deep learning you lose transparency. Information goes in, a bunch of computations take place and the system trains itself to learn, and out comes the output, which in its ideal state is used to drive decision making. That’s where I see the majority of hesitation, which can be challenging. 

“There are plenty of applications of the technology that are far more advanced than we use in this industry, but getting regulators, clients or even users internally to trust that type of output and march forward using it seems to be the real challenge.”

Speaking at a separate session at the conference, T. Rowe Price’s global head of systematic trading and market structure, Mehmet Kinak, suggested that AI was a “great buzzword” for the industry but he had yet to find any organisation that developed a good system.

“Machine learning on the other hand is interesting, like a broker wheel for example. It incorporates a lot of transaction cost analysis (TCA) and data into the trading decision,” he said.

“Machine learning I believe will take over, but I’m not sure about AI. Even on a wheel, data is driving that approach and I’m agnostic to brokers I’m using because it’s just data and it’s like I can’t see them.”