Page 492 - ISC PROCEEDINGS 21.4

gold) based on each individual's financial goals and risk tolerance. Furthermore, by
collecting and processing data from economic news, global financial reports, and industry
or corporate financial statements, AI generates high-precision short-term forecasts for
exchange rates, interest rates, and stock markets. These insights serve as a reliable
reference for both customers and banks in their decision-making processes.
Open Banking and Ecosystems
AI acts as the catalyst, connecting banking operational platforms more effectively
than traditional banking methods. Embedded Finance, Banks provide intelligent APIs to
integrate lending and installment services directly into shopping, ride-hailing, and travel
applications. Cross-Data Analysis, AI analyzes data from partners within the ecosystem to
understand the holistic "big picture" of a customer's life, thereby offering the most
precise insurance packages or consumer incentives.
Table 1. Core Functional Differences in Banking Operations
Criteria Traditional Banking Banking in the AI Era
Based on physical Based on behavioral data and
Lending Decisions
documentation (Paper-based) predictive analytics
Customer Support Business hours, manual waiting 24/7, instantaneous response
Based on self-adaptive
Fraud Detection Based on fixed rules
machine learning
Business Strategy Product-centric Customer experience-centric
Source: Synthesized by the author from academic research
2.1.3. Emerging challenges and risks for Banks in the AI Era
Security risks and high-tech crimes
While the AI era empowers banks with superior security, it simultaneously provides
cybercriminals with sophisticated "weaponry." (1) Deepfake Attacks, criminals leverage AI
to forge facial features or voices to bypass biometric authentication barriers (eKYC) and
hijack accounts. (2) Adversarial Attacks (Data Poisoning): Conversely, attackers may
intentionally inject "noisy" or malicious data into the system to deceive AI models. This
causes them to make erroneous decisions, such as miscalculating the credit risk level of a
fraudulent profile.
The "Black Box" Problem and Transparency Issues
Modern AI models, particularly Deep Learning, often operate with extreme
efficiency but lack explainability in their decision-making mechanisms. When a loan
application is rejected, banks struggle to provide specific, justifiable reasons as required
by regulatory authorities. Furthermore, model risk can arise when input data is flawed or
errors occur during the training process. This may cause the model to generate
"hallucinations," leading to mass-scale biased or erroneous decisions that human
operators may not detect in time.
Algorithmic Bias and AI Ethics
Since AI processes historical data, it may inadvertently "learn" and perpetuate
human biases embedded within those datasets. This leads to two critical scenarios.
Discrimination, algorithms may unintentionally deny services to specific groups based on
gender, geographic location, or ethnicity, simply because historical data patterns reflect
past inequalities. Privacy Infringement, the extensive collection of "alternative data" (such
as social media behavior and daily lifestyle habits) for AI analysis may lead customers to
feel surveilled, potentially violating their personal privacy rights.
Infrastructure and human resource challenges

491