MIT Delivers Comprehensive Database of 700+ AI-Associated Risks

The Massachusetts Institute of Technology (MIT) researchers have taken a proactive step in addressing the potential pitfalls of artificial intelligence (AI). They’ve created a “living database” documenting 777 risks linked to AI, extracted from 43 different taxonomies. This initiative aims to provide an accessible and continually updated overview of the complex AI risk landscape.

The AI Risk Repository

The AI Risk Repository was born out of the recognition that adopting AI comes with inherent dangers. AI systems can exhibit biases, propagate misinformation, or even become addictive. More alarmingly, they could be exploited to create new weapons or, in a worst-case scenario, spiral out of control. To effectively manage these potential risks, a comprehensive understanding of them is crucial.

MIT’s FutureTech Group at the Computer Science & Artificial Intelligence Laboratory (CSAIL), in collaboration with other teams, developed the AI Risk Repository to fill this knowledge gap. Their review of existing AI risk frameworks exposed significant shortcomings, with even the most thorough frameworks overlooking around 30% of the identified risks.

The Repository serves as an accessible overview of the AI risk landscape, providing a regularly updated source of information and a common reference point for various stakeholders, including researchers, developers, businesses, evaluators, auditors, policymakers, and regulators.

Structure of the Repository

The Repository comprises three key components:

1. The AI Risk Database: This captures the 700+ risks extracted from the 43 existing frameworks, complete with quotes and page numbers for reference.
2. The Causal Taxonomy of AI Risks: This classifies how, when, and why these risks arise.
3. The Domain Taxonomy of AI Risks: This categorizes the risks into seven domains and 23 subdomains, encompassing areas like discrimination & toxicity, privacy & security, misinformation, malicious actors & misuse, human-computer interaction, socioeconomic & environmental harms, and AI system safety, failures and limitations.

A Tool for AI Governance

Experts view the Repository as an invaluable resource for leaders establishing AI governance within their organizations. It provides a ready-made catalog of AI risks, saving organizations the effort of identifying and categorizing them independently. The Repository’s convenient Google Sheet format allows for easy customization and adaptation to specific organizational needs.

Limitations and Future Growth

While acknowledging the Repository’s current limitations, such as its reliance on 43 taxonomies and potential for errors or biases, the researchers emphasize its significance in highlighting the substantial range of AI risks. They view it as a foundation for a more coordinated and comprehensive approach to defining, auditing, and managing the risks associated with AI systems.

The Repository is expected to evolve and grow over time, incorporating potential mitigating measures and best practices. This “living work” aims to foster a proactive approach to responsible AI use, ensuring control over data usage, technology functions, and deployment contexts.

The MIT AI Risk Repository represents a significant stride towards understanding and addressing the multifaceted risks posed by AI. By providing a comprehensive and accessible resource, it empowers various stakeholders to make informed decisions and navigate the complex AI landscape responsibly.