After engaging in Peter Moss’s exciting keynote speech, “Garbage in, Garbage Out—Why Regtech is Only Half the Answer” (listen to the Podcast here), discussing the crucial importance of data as a foundation for regulatory compliance, the audience at the recent RegTech Summit New York on November 15, 2018 prepared themselves for a leap into an altogether different discipline: with a panel discussing the potential and promise of AI and machine learning within the regtech industry.
Moderated by Dale Richards, President of Island 20 Ventures, panelists discussed business use cases, how AI and machine learning could be used to augment systems, and barriers to the widespread use of AI and machine learning.
The panel agreed upon working definitions for machine learning and AI when Mary Jane Ajodah, Vice President of FinTech Strategy & Partnerships in the Chief Digital Office at BNY Mellon, distinguished the terms. Ajodah explained to the audience that most machine learning techniques, or training a machine to learn through inference, could generally be taken on by a data science team; while in AI, the machine can learn independently.
The concepts behind the “Garbage in, Garbage Out” address drove some of the discussion, as panelist Ali van Nes, SVP and Senior Director of Regulatory Solutions at Factset, emphasized the point that without organized data underlying AI, it becomes difficult to implement AI. Viktoriia Samatova, Vice President and Head of Research and Product Development at Quantextual Research at State Street, commented that data can be both unstructured and biased, which can impact the quality of the model. Nes identified the weak links for firms as data management and plugging into existing systems.
“If you can’t solve those problems, the best regtech in the world is not going to deliver what you’re looking for,” noted van Nes.
Overbond CEO Vuk Magdelinic proposed that AI could be applied not simply as a trading vehicle, but to fill gaps in datasets.
“There are a lot of problems, especially in very gappy data sets, [such as] unknown correlation sets, so just coming in with a quant model with statistical assumptions and coming in trying to validate all your hypotheses is limited,” explained Magdelinic. “The AI application does give you a pickup in precision and that’s the primary motivation to go through some difficulties now with big data and modeling and the iterative approach that AI requires.”
Currently, van Nes foresees an issue due to a lack of common standards and interpretations across the enforcement of regulations. Due to the lack of standard definitions amongst regulators, it is difficult for AI to solve for the problem.
“If standards do start to bubble up, we’ll have an ecosystem that’s more ripe for AI,” she predicted.