Business Intelligence: Has AI called checkmate on human decision-making in Financial Services?
When an unstoppable force meets an immovable object?
In 2018, two very different chess AI programmes played one another. One, called ‘Stockfish 8’, had been programmed with 400 years of human chess grandmaster games. The other, ‘AlphaZero’, was programmed with the rules of chess and allowed to play itself for four hours, with no human input, relying only on machine learning to develop its strategies. The chess and AI worlds held their breath as a titanic tussle looked guaranteed.
After 1000 games, it became apparent that AlphaZero’s machine learning had crushed Stockfish’s human grandmaster expertise, with AlphaZero losing only 6 times (1). Not only was it convincing, but commentators were in awe of the way that AlphaZero had played. Rather than the dry and mechanical style that is commonly associated with chess AI, experts remarked that AlphaZero’s style of play was “unorthodox”, “creative” and “highly dynamic” (2). One claimed, “it’s like discovering the secret notebooks of some great player from the past” (1), and another observed that the moves appeared to be “alien” (3). But what gave AlphaZero such an advantage?
Machine decision-making can beat human decision-making due to the elimination of bias
Being self-taught meant that AlphaZero was unconstrained by conventional wisdom and, perhaps more importantly, human bias. As such, it was able to create a superior, human beating strategy. AlphaZero made materiality sacrifices early in the game to win strategic gains that were recouped in the longer term. Commentators noted on some of AlphaZero’s more aggressive moves that “no top chess player would take such a big risk” (4). The idea that ‘material’, i.e. the value of the chess pieces on the board, provides an advantage in chess is one that “underpins the modern game” for humans (1). The idea that increased materiality is linked to an increased chance of survival can also be seen in investments. Organisations with a larger balance sheet, such as FTSE100 companies are deemed more likely to survive challenging circumstances and are deemed less risky than other smaller cap companies.
Should human decision-making be replaced by machine decision-making in Financial Services?
There is competitive advantage to be gained where human bias is identified and removed, which AI can help expose and facilitate better decision-making. A large number of FS organisations are already seeking to take advantage of these opportunities. Of respondents to an FCA survey last year, two thirds had examples of live machine learning and 52% of firms have a machine learning strategy in place, with the majority of adopters being in the insurance and banking spaces. (5) There are further examples where banks have improved their Anti-Money Laundering investigation decision-making time by up to 200% through AI by providing greater insight from data and allowing humans to focus on the most suspicious cases. (6) In many instances, AI has become the first line of defence against financial crime with humans being used to perform oversight checks where required.
Machine decision-making is helpful, but there are pitfalls
Although there are benefits from using AI to improve decision-making, FS organisations should be wary that human bias is not programmed into AI solutions or ingrained in the underlying data being used. For example, Apple’s co-founder, Steve Wozniak complained that Apple’s ‘sexist’ credit card offered him significantly more credit than his wife, despite having no separate bank accounts or separate assets (7), creating reputational harm and drawing the attention of New York’s financial regulator. The algorithms used to offer credit had bias ingrained within them and examples such as this should act as a warning to FS organisations of the need to ensure that the decision-making strategies of their AI products are transparent and free from human bias. (8)
In an imperfect world where information is often lacking, human intervention is still important
AlphaZero represents a significant step forward for AI and indicates that AI can make better decisions when it is not influenced by human bias. However, this example is limited to a game based on ‘perfect information’, where each player is fully informed of events that have previously occurred. Unfortunately, this is not always true of the real-life conditions in which Financial Service organisations operate. As one expert puts it, most AI solutions are only, “good at doing the one thing they’re programmed for” (4). Although AI may be very powerful at solving problems and processing data in a perfect information scenario, it lacks the understanding of context, long-term goals and general problem-solving abilities which favour human decision-making.
Can AI be combined with human decision-making?
In short, yes. Although AI technology has a relatively confined set of specific use cases for now, it can be utilised as a powerful tool to help organisations remove human bias and make better decisions. A human-AI team would have the advantage of the immense, unemotional calculating powers of the machine to drive insight, whilst also having the context-specific, generalist advantage of human problem-solving. Humans can use the insight that AI provides and can even learn from the new knowledge it produces. (1)
Augmented Analytics can drive greater insight and better decision-making
To maximise the benefit of available AI technology to help make informed, bias-free and data-driven decisions, Financial Service organisations should instead augment human insight with AI insight. This is known as Augmented Analytics, where AI provides insights such as correlations, exceptions, clusters, outliers and even predictions detected in data sets. Humans can investigate these bias-free AI generated insights to gain a better understanding of data and therefore optimise decision-making. As well as bespoke solutions, more generalist Augmented Analytics capability is already available as a feature with market leading Business Intelligence tools such as Power BI and Tableau. These tools can scan a data set for insights and highlight these for a human to review and investigate further, potentially identifying findings that a human may not have detected.
Augmented Analytics is still several years away from mainstream adoption (5). However, FS organisations have in some cases bucked the trend, being quick to see the benefits that this kind of insight can drive. Leaders in this space, such as Monzo, are already providing augmented analytics tools that help customers understand if they may be running out of money based on previous spend. In addition, Monzo’s Business Intelligence tool, ‘Looker’, enables 85% of business intelligence queries to be self-served, driving far greater insight from its data by non-technical business users (10). As an industry, FS is the biggest spender on AI services outside of technology firms (11) and appears to be embracing the trend towards commoditisation of data that has been effectively employed by tech firms such as Facebook, Alibaba and Google. (13)
The use of AI technology is only set to expand as a powerful tool to help organisations remove human bias, improve insight and make better decisions from ever-increasing data pools. Given that Finance is a, “wonderfully pure information-processing business” (12) FS organisations are in a unique position to harness the potential of their data. A key differentiator for the success and survival of FS organisations will be the ability to use data to drive effective decision-making. Those FS organisations that can effectively combine AI and human insight into decision-making through augmented analytics will pull ahead of the competition.
Checkmate to the Human-AI Team.
(2) Game Changer: AlphaZero’s Groundbreaking Chess Strategies and the Promise of AI” Matthew Sadler, Natasha Regan (Feb, 2019)
(12) Humans Need Not Apply: A Guide to Wealth and Work in the Age of Artificial Intelligence” Jerry Kaplan (Aug, 2015)