Georgia Institute of Technology: It’s Just Distributed Computing – Rethinking AI Governance
Summary of Read Full Report
Argues that the current approach to governing "AI" is misguided. It posits that what we call "AI" is not a singular, novel technology, but rather a diverse set of machine-learning applications that have evolved within a broader digital ecosystem over decades.
The author introduces a framework centered on the digital ecosystem, composed of computing devices, networks, data, and software, to analyze AI's governance. Instead of attempting to regulate "AI" generically, the author suggests focusing on specific problems arising from individual machine learning applications.
The author critiques several proposed AI governance strategies, including moratoria, compute control, and cloud regulation, revealing that most of these proposed strategies are really about controlling all components of the digital ecosystem, and not AI specifically. By shifting the focus to specific applications and their impacts, the paper advocates for more decentralized and effective policy solutions.
Here are five important takeaways:
- What is referred to as "artificial intelligence" is a diverse set of machine learning applications that rely on a digital ecosystem, not a single technology.
- "AI governance" can be practically meaningless because of the numerous, diverse, and embedded applications of machine learning in networked computing.
- The digital ecosystem is composed of computing devices, networks, data, and software.
- Many policy concerns now attributed to "AI" were anticipated by policy conflicts associated with the rise of the Internet.
- Attempts to regulate "AI" as a general capability may require systemic control of digital ecosystem components and can be unrealistic, disproportionate, or dangerously authoritarian.