Advances in generative artificial intelligence (gen AI) have opened up new commercial, social and technological opportunities. But the development of generative AI in Singapore is a double-edged sword, according to a new white paper by a consortium made up of the Monetary Authority of Singapore (MAS), financial institutions, technology companies and consultants: To tell.
Published in May 2024, the document outlines the challenges associated with Generation AI and shares a comprehensive Generation AI Risk Framework to help financial institutions use the technology in a responsible manner.
The Risks of Generative AI in Singapore
One of the main concerns highlighted in the whitepaper is the presence of bias in generative AI models: for example, temporal or historical bias could lead these models to favour perspectives from a particular time period, producing outputs that reflect outdated views.
Political bias is also a concern, as biased training data could lead AI systems to favor certain political ideologies over others, potentially resulting in content that favors one political position at the expense of other valid viewpoints. Additionally, generative AI could perpetuate stereotypes, produce biased information, and unfairly disadvantage certain demographics.
The risk of generating harmful, offensive or malicious content (toxicity) is also significant with gen AI, especially if the AI learns inappropriate language or behavior from its training data. Harmful output from gen AI has the potential to cause significant harm, including the spread of misinformation or offensive content.
Another ethical concern outlined in the report is that AI could produce convincing deepfake content, leading to misinformation.
In addition, gen AI uses public data for training, often developed by third parties, which raises several legal and regulatory risks, including copyright and intellectual property issues, as well as privacy concerns. Privacy concerns are further amplified when personal data, such as social media, is used in gen AI.
Gen AI also poses increased security risks, including data contamination, model manipulation, and prompt injection. This can be due to the sheer amount of data required for training, the dependency on third-party suppliers for models and data, and the free-form nature of the inputs accepted and outputs produced. Additionally, Gen AI can be used for malicious purposes, amplifying the scale, speed, and sophistication of cyber attacks and fraud.
The report also introduces a platform-agnostic gen AI reference architecture, providing a list of building blocks and components that organizations can use to create a robust, enterprise-level gen AI technology capability. The architecture is intended to serve as a sample framework, rather than a prescriptive set of practices, to help organizations jump-start their pilot phase and map out their implementation journey as they adopt the technology, from initial use cases to full-scale enterprise deployment.
Generative AI Platform-Agnostic Reference Architecture Sample, Source: Project MindForge, via Monetary Authority of Singapore, May 2024
Adoption of generative AI in Singapore’s banking sector still in its early stages
The whitepaper, titled “Emerging Risks and Opportunities of Generative AI for Banking,” was produced by MindForge, a consortium formed in 2023 to understand the risks and opportunities of generative AI technology specific to the financial services industry.
The consortium comprises MAS, Citi, DBS, HSBC, OCBC, Standard Chartered, the Association of Banks in Singapore (ABS), UOB, and technology partners Accenture, Google and Microsoft.
Phase 1 of Project Mindforge Conclusion Established on November 15, 2023, the consortium aims to develop a risk framework for Generation AI. In the next phase, the consortium will expand its scope to include financial institutions in the insurance and asset management industries, extend the Generation AI risk framework to the entire financial industry, and develop an AI governance handbook for the industry.
Gen AI is an AI system designed to autonomously generate new, original content. These systems have expanded rapidly over the past few years, driven by excitement following the introduction of OpenAI’s ChatGPT in late 2022 and the opportunities it offers for automating routine tasks, improving efficiency, and enhancing recommendation engines and customer experiences. In the banking sector, McKinsey Quote Next-generation AI could deliver potential benefits of US$200 billion to US$340 billion per year, equivalent to 9% to 15% of operating profits.
In Singapore, financial institutions are still in the early stages of implementing gen AI capabilities, with early use cases primarily focused on improving internal productivity, risk mitigation and insights, while emerging use cases are looking at revenue generation and improved customer interactions, according to the MindForge report.
October 2023, UOB became It is the first bank in Singapore to pilot Microsoft’s Copilot, an AI-powered assistant, to improve productivity, accessibility and collaboration, followed by OCBC Bank in November. launch It introduced the gen AI chatbot to help its 30,000 employees around the world with writing, research and ideation.
Singapore was It was one of the first jurisdictions in the world to articulate principles for AI governance, publishing the Model AI Governance Framework (MGF) in 2019. Updated in 2020, the document promotes the responsible use of traditional AI.
In response to the rise of Gen AI and the associated risks, the AI Verify Foundation and the Infocomm Media Development Authority of Singapore proposed an enhanced framework in January 2024. The MGF-Gen AI aims to address concerns about Gen AI while promoting innovation, and calls for collaboration among policymakers, industry, researchers and the public.
Featured Image Credit: Free Pick
