Stochastic Flocks and the Critical Problem of 'Useful' AI

3 days ago

AI is experiencing an 'agentic turn', moving from generating text to autonomously producing and executing code.
Agentic AI systems are like 'stochastic flocks'—multiple LLMs working together—creating more complexity and inscrutability.
Despite improvements in usability, core issues like accountability, bias, environmental costs, and labor exploitation persist.
Agentic AI leads to 'slopware': hastily produced, under-reviewed software that can cause unintended harm.
Reliability is overstated; hallucinations remain inevitable, and generated code is hard to audit or trust.
Accountability failures can stack up invisibly, causing serious harm in sensitive applications like government services.
Agentic AI promotes computational solutionism, reframing social problems as coding tasks, risking dehumanization.
These systems consume vast resources, scaling inefficiency and environmental impact.
Critique must focus on systemic harms—political power, surveillance, biased data—not just technical failures.
The urgent debate is about the boundaries of 'usefulness': who benefits, at what cost, and who decides?

Hasty Briefsbeta