Can AI Comply to Transparency Standards?
Authored by: Robert Huntsman, Senior Director, Head of Data Science Practice - Synechron
Risk-related regulatory compliance spending is expected to reach $72 billion annually by 2019. Technology expenditures are increasing as financial institutions strive to meet evolving regulatory standards such as MiFID II and to exploit advancements made possible by improvements in processing, memory, and distributed computing.
One such advancement is the ability to apply machine learning and artificial intelligence to regulatory problems. Banks are using machine learning for trade surveillance and transaction monitoring. Regulators themselves are using machine learning to parse filings and detect anomalous patterns. Early adopters of machine learning and AI for regulatory purposes include The Securities Exchange Commission (SEC) and Financial Regulatory Authority (FINRA).
Machine learning is the use of use of algorithms and statistical methods to progressively improve the performance of computers on specific tasks without human intervention. Common problems include minimizing the distances among objects to form clusters or recognizing patterns by reweighing inputs until an optimal combination is found that correctly identifies a large percentage of outputs. Machine learning is a subset of artificial intelligence (AI), the development of computer algorithms able to perform tasks that usually require human intelligence. Deep learning is a specific type of artificial intelligence that uses neural networks to mimic the human thought and decision-making process. For regulators who require a clear audit trail, deep learning is challenging given that neural networks include hidden layers that don’t provide the level of transparency usually required by regulators.