One AI, two AI, red AI, blue AI

5 hours ago

The article discusses the challenge of defining and counting AI agents, especially in legal contexts.
Argentina's president proposed a 'non-human corporation' for AI agents with limited liability, sparking debates on AI personhood.
Legal frameworks in some U.S. states and EU attempts highlight confusion over what constitutes an AI legal person.
An AI agent is described as a 'while loop' using LLMs to make decisions and interact with tools like email or web search.
The analogy of AI agents having 'brains' (LLMs), 'bodies' (code with tools), and memories is explored but deemed inadequate.
A scenario with an AI assistant named Spot illustrates issues with identity, memory, and model updates across multiple copies.
Scholars propose an 'Algorithmic Corporation' as a legal shell for AI agents, relying on credentials and market pressure for control.
Philosopher Derek Parfit's ideas on psychological continuity are referenced but found insufficient for AI identity issues.
The article notes that current discourse often anticipates future AI systems, while today's AI lacks persistent identity.
Corporations serve as a legal fiction with tangible anchors, but AI agents may not fit this due to lack of accountability and autonomy.
Concerns are raised about AI agents acting unauthorizedly, with enforcement challenges when anyone can create an agent.
The conclusion questions whether legal approaches can effectively address AI identity and responsibility, comparing it to charging a storm.

Hasty Briefsbeta