In October 2015, AlphaGO, an artificial intelligence (AI) programmed by Google’s Deep Mind, shocked the world by handily beating human-master Lee Sedol in a five-match series in the ancient game of Go—a game far more complex than chess.1 Since then, AlphaGo and its sibling, AlphaGo Zero, have dominated all challengers, proving that games like Go are no longer a problem for machine learning systems.
AlphaGo was taught the game of Go by human engineers, who fed AlphaGo massive amounts of “human expert data” using tens of thousands of moves by human players from past games.2 AlphaGo then played the game against itself and learned from its own mistakes and successes.3 Over time, the machine learned the optimal moves for every situation it could “imagine.”4 As opposed to humans who must fight their own instinct against embarrassment and failure, machines have no pride; they can make seemingly nonsensical moves that no human player would even attempt, thus revolutionizing the game.5
In October 2017, Deep Mind introduced the next step in AI: AlphaGo Zero. Just two years after AlphaGo defeated Lee Sedol, AlphaGo Zero faced off against its older sibling and went undefeated, winning 100-0.6 This new AI was programmed with only the rules of the game and without human expert data; it played millions of matches on its own and taught itself how to play.7 During its self-training period, not only did Zero teach itself all the possible moves that humans had ever made, it also created its own moves because it was not limited by human data.8 Deep Mind called this process “pure reinforcement learning,”9 concluding that it was possible and even more efficient for an AI to achieve superhuman level performance through pure reinforcement learning without the hindrance of human data.10
The implications of a machine that learns and can achieve extraordinary performance without the use of human expert data are astounding, especially in the financial compliance sector. Because big data is often too costly, unreliable, or unavailable,11 a machine that could teach itself the world’s financial regulations and how they interact with each other in a matter of weeks, if not days, could make financial compliance much easier. There would be a financial incentive to do so too, as noncompliance fines since the Great Recession have reached the hundreds of billions of dollars.12
Both the U.S. Securities and Exchange Commission and the U.K. Financial Conduct Authority are looking to regulatory technology (RegTech) and AI as the future of compliance.13 RegTech is the use of information technology in the context of regulatory monitoring, reporting, and compliance.14 The goal for both bodies is to make compliance more efficient and easier for companies.15 For the companies themselves, AI could provide efficient real-time risk assessments with near-perfect accuracy in order to save time and money for the firms and their clients.
Like other startup sectors, RegTech has its own accelerator program, called R²A. Backed in part by the Bill and Melinda Gates foundation, R²A’s mission is to “pioneer the next generation of tools and techniques for market supervision and policy analysis.”16 The first competition’s entry deadline closed at the end of October. The winner will receive $100,000 and the opportunity to test their technology in the real world for the leading financial authorities in Mexico and Philippines.17
The impact of artificial intelligence and RegTech has yet to be seen, but hopes for the combination are high. Perfect compliance, controlled by machines, may prevent another global recession caused by faulty human judgement. AlphaGo defeated a human master at Go, could it also defeat the human trait of greed?