The Limits of AI in Financial Services
Isabella Loaiza, Roberto Rigobon · 2025
Read The Limits of AI in Financial Services by Isabella Loaiza, Roberto Rigobon (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-22035.txt.
Introduction
Isabella Loaiza and Roberto Rigobon wrote a 2025 essay on the limits of artificial intelligence in finance. AI excels at data-heavy tasks, they argue, but it falters with tiny data sets, judgment calls, and questions of trust and ethics. Their framework, named EPOCH, lists five human strengths machines cannot copy: empathy, presence, opinion, creativity, and hope.
In banking those traits carry client trust, inclusion, and new ideas. The authors expect AI to reshape jobs rather than erase them, much as cash machines changed the work of bank tellers. The real task, they say, is training staff to pair machine speed with human judgment. The essay speaks to managers weighing how far to automate.
Catalog record
| CREATOR | Isabella Loaiza, Roberto Rigobon |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-22035
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Limits of AI in Financial Services Isabella Loaiza, Roberto Rigobon Abstract AI is transforming industries, raising concerns about job displacement and decision-making reliability. AI, as a universal approximation function, excels in data-driven tasks but struggles with small datasets, subjective probabilities, and contexts requiring human judgment, relationships, and ethics. The EPOCH framework highlights five irreplaceable human capabilities: Empathy, Presence, Opinion, Creativity, and Hope. These attributes are vital in financial services for trust, inclusion, innovation, and consumer experience. Although AI improves efficiency in risk management and compliance, it will not eliminate jobs but redefine them, similar to how ATMs reshaped bank tellers’ roles. The challenge is ensuring professionals adapt, leveraging AI’s strengths while preserving essential human expertise. Keywords: AI, Financial Services, Human Capabilities 1 Main Almost every sector faces disruption from the rise of Artificial Intelligence, driven by two core fears. The first is substitution—the concern over which jobs will remain once machines and algorithms can perform most tasks. The second is AI’s judgment—an existential fear reminiscent of The Matrix, where an advanced AI might determine that humans are more harmful than beneficial to global progress. These anxieties have shaped public debate across industries, and the financial system is no exception. In our paper “The EPOCH of AI” by Loaiza and Rigobon [ LR24 ] , we take a different approach to the question of labor market substitution, and implicitly, also dispel the fear of AI’s judgement. This approach – as we highlight in the paper – is complementary to a very large existing literature, but it is distinct in one important dimension: instead of analyzing which tasks and jobs the current technology can substitute, we ask what capabilities will AI never be able to substitute. Artificial Intelligence is fundamentally built on universal approximation functions—an idea that mathematicians and engineers have studied for centuries. The origins of function approximation trace back to Euclid and Archimedes (c. 300 BCE), who used geometric methods to estimate areas and volumes. In the 14th century, Madhava of Sangamagrama (India) developed power series expansions, laying the groundwork for what would later become the Taylor series. Then, in the 1660s, Isaac Newton introduced polynomial approximations and the binomial expansion, providing the foundation for Brook Taylor’s formalization of the Taylor series in 1715. The goal was always the same: to approximate complex functions with simpler ones. The 18th and 19th centuries saw major advancements in approximation theory, including Fourier series for periodic functions, Chebyshev polynomials, and Carl Friedrich Gauss’s least squares approximation. These methods broke down complex functions into more manageable components, forming the backbone of numerical analysis. Today, modern AI models—from Random Forests, to Neural Networks to Large Language Models—are all forms of universal approximators. They take highly intricate processes and approximate them using vast collections of simpler functions, enabling machines to learn, predict, and generate with remarkable accuracy. As highlighted by Loaiza and Rigobon [ LR24 ] , universal approximation functions have inherent limitations when: The dataset is small; the model must extrapolate far beyond the training data; the process allows for multiple acceptable outcomes; the desired outcome is a relationship rather than a recommendation or action; the process relies on subjective probabilities rather than realized probabilities. Although these challenges are technical, they align with core human abilities. We refer to these as human capabilities, encapsulated in the EPOCH framework: • Empathy and Emotional Intelligence • Presence, Networking, and Connectedness • Opinion, Judgment, and Ethics • Creativity and Imagination • Hope, Vision, and Leadership In the financial system, many critical aspects of financial services are deeply reliant on EPOCH, highlighting the irreplaceable role of human expertise in areas where AI falls short. This is not the first time a technology threatens the financial system – nor will it be the last. In fact, the introduction of ATMs is one of the most talked about examples in every single business school to highlight how technology does not destroy jobs, it can also enhance them. Before ATMs, bank tellers’ jobs centered around money exchange with customers – either to accept deposits or withdrawing them. The fear was that ATM’s were going to substitute bank tellers. In practice, ATM’s substituted the most time-consuming activity the tellers were doing, freeing them to be able to perform other services. Instead of declining, the number of bank tellers in the US soared. The key for the bank tellers was to invest in tasks and education on complementary activities to the technology – instead of investing time and effort in those aspects that competed with it. Therefore, the question was, what can the teller do that is complementary to the ATM? That is exactly our approach, what the financial sector can do that is complementary to the AI tools that are likely to appear in the future? Financial services offer many benefits. Some are clearly AI driven. 1.1 Benefits and Challenges of AI in Financial Services Liquidity provision. Liquidity provision ensures that funds are readily available for transactions, lending, and investment without causing market disruptions. AI can enhance this process by predicting liquidity needs, optimizing capital allocation, and automating market-making activities. Machine learning models analyze vast datasets in real time to forecast liquidity shortages, adjust asset allocations, and detect potential liquidity crises before they unfold. High-frequency trading algorithms and AI-driven market makers efficiently provide liquidity by continuously adjusting buy and sell orders, reducing spreads, and stabilizing financial markets. Risk Management. AI-driven models assess credit risk, detect fraud, and manage market fluctuations. AI can significantly enhance this activity by analyzing large datasets in real time to detect emerging risks, predicting market trends, and automating decision-making processes. Machine learning models can evaluate creditworthiness, monitor transactions for signs of fraud, and assess portfolio risk by simulating various market scenarios. Portfolio Investment. The financial system already offers robo-advisors! The greatest advantage is to be able to provide real-time data analysis that improves aspects such as tax harvesting and other services. Efficiency. For example, algorithmic trading involves using automated, pre-programmed strategies to execute trades in financial markets at high speeds and volumes. AI significantly enhances trading by detecting patterns, and identifying trading opportunities faster than human traders. Regulation. Almost every bank today uses machine learning tools to detect money laundering. Compliance with regulation has been already enhanced by AI - detecting hidden patterns in the data has been and will be a game changer for compliance. These are all activities where AI has and will continue to have major impact. These tasks are clearly at risk, but the most fundamental question is, are the jobs at risk? Or will these jobs be enhanced as it was the case with bank tellers? There are other aspects in financial services that cannot be substituted by machines - Activities and tasks that are fundamental to the provision of the service. This is the topic we would like to spend most of our time on. Trust : According to Psychologists trust is the belief in the reliability, integrity, and ability of another person or system. It is the foundation behind one person’s feelings of confidence in another one, and it is the fundamental building block of cooperation and connection. It requires repeated positive experiences, consistency, transparency, and quite importantly, the emotional safety