The Finance Shelf
Library/IMF Fintech

Generative Artificial Intelligence in Finance: Risk Considerations (IMF Fintech)

Research paperIMF FintechAugust 202324 pages36 min readOriginal source
Key number
100 million
ChatGPT 2-Month Active Users p.6
Key number
1.75 trillion
WuDao 2.0 Parameter Count p.16

Summary

  • ChatGPT acquired more than 100 million active users within two months of its launch on November 30, 2022.p.6
  • BERT was introduced by Google AI in October 2018 with 340 million parameters and achieved a GLUE score of 86.5.p.16
  • Microsoft released Turing-NLG in February 2020 featuring 17 billion parameters and a GLUE score of 92.8.p.16
  • OpenAI launched GPT-3 in November 2022 with 175 billion parameters and a GLUE score of 80.3.p.17
  • GPT-4 was released by OpenAI in March 2023 with 175 billion parameters and achieved a GLUE score of 93.3.p.17
  • WuDao 2.0 was introduced in June 2022 by Beijing Academy of Artificial Intelligence with 1.75 trillion parameters and a GLUE score of 94.2.p.16

Key insights

  1. Generative AI distinguishes itself from standard AI/ML technologies through its core ability to create new content using large language models.p.6
  2. Publicly available GenAI systems create major privacy vulnerabilities by automatically using input data from user engagements for model re-training.p.8
  3. GenAI aggravates embedded bias due to its reliance on broad web text datasets containing human prejudices and vulnerability to search engine optimization tools.p.9
  4. Hallucination risks in GenAI produce plausible-sounding but wrong outputs, presenting threats to financial risk assessment and consumer protection.p.10
  5. Financial service providers are using GenAI to code synthetic data generators for customizing fraud detection, product delivery, and regulatory reporting.p.11
  6. The massive scale of training data and multi-layered neural architectures make output explainability an ongoing challenge for financial adoption of GenAI.p.12
  7. Emerging cybersecurity risks in GenAI include increased susceptibility to data poisoning, prompt injection attacks, and universal jailbreaking.p.13
  8. Excessive reliance on GenAI market assessments can amplify herd mentality, mispricing risk, liquidity dislocations, and systemic financial shocks.p.14

Topics and entities

GlobalAI in financeBankingCybersecurityRegulationCapital OneChatGPTCitadelConsumer Financial Protection BureauGoldman SachsGoogleGPT-4JPMorgan ChaseMicrosoftMorgan StanleyOpenAIWells Fargo

Summary and figures generated from the document and checked automatically: every number shown appears on the cited page. The full document is available from the publisher. Publishers can request removal.