Jeff Jarvis
Agents of Chaos, an entertaining if ominous paper, explores the consequences of agentic autonomy in OpenClaw. In a fenced environment, agents succeeded at some tests, failed others--e.g., killing an entire email server when it couldn't erase one email. 1/
https://arxiv.org/pdf/2602.20021
Mar 21 at 2:40pmWeb
Show threadJeff Jarvis1d ago
The authors' unimpeachable point: "These behaviors raise unresolved questions regarding accountability, delegated authority, and responsibility for downstream harms, and warrant urgent attention from legal scholars, policymakers, and researchers across disciplines." 2/
Show threadJeff Jarvis1d ago
This is just the kind of stepping-off inspiration I hope authors will bring to me as editor of the new Bloomsbury book series, Intelligence--AI and Humanity. 3/
https://medium.com/whither-news/intelligence-ai-and-humanity-d8c5d6cda6ef
Intelligence — AI and Humanity

Announcing a new book series from Bloomsbury (and me)

Medium
Show threadJeff Jarvis1d ago
The behaviors of these agents drive us to reexamine notions of autonomy & freedom, regulation & control, of responsibility & how norms & laws are negotiated & enforced. How do we school & police a society of agents given our failures as humans? What are the prices of liberty & restraint? 4/
Show threadRad und Tat1d ago
@jeffjarvis Very true. And it's one thing to deploy these agents on a home or office server, but what if they start being used in medical settings, government departments, vehicles or combat systems. Who will go on trial when they make mistakes that cause significant damage or even kill people? The server's owner, the sysadmin who runs it, the "chief intelligence officer" , the person who programmed it? An agent cannot be held accountable for war crimes.