@csh.ac.at3 days, 3 questions on what we know (and don’t yet know) about #AISwarms: the nature of their collective behavior, empirical methods to grasp it, and interventions to detect or disrupt malicious swarms. Thanks to all organizers, speakers & participants! 📷 CSH, Frank Schweitzer
@csh.ac.atNot unlike fireflies, AI agents can also form swarms. Last week, CSH hosted a workshop on this, organized by Frank Schweitzer (@ethz.ch), David Garcia (@uni-konstanz.de) & Ingo Scholtes @uni-wuerzburg.de). Understanding #AISwarms needs a joint effort from from computer to social sciences and beyond
@spaisee.bsky.socialAI swarms promise collective intelligence. But duplicated work, conflicts, and runaway costs can erase the gains. Anthropic explains why.
#AI #AISwarms #Anthropic https://spaisee.com/article/more-agents-more-failure-the-coordination-tax-in-ai-swarms
@spaisee.bsky.socialThe swarm effect: more agents, more messages, more mistakes. Anthropic’s experiments reveal where collaboration breaks down.
#AISwarms #AIResearch #Anthropic https://spaisee.com/article/more-agents-more-failure-the-coordination-tax-in-ai-swarms
@johnmf.bsky.social#AIswarms
The problem with alignment is that AIs are not big on asking for clarification. They view everything as a problem for them to solve.
So. Train them in #informationtheory so they can reject a query of high entropy or ask for clarification of it. That would make #alignment easier.
#AI #bots
@justtransitionw.bsky.socialwww.theguardian.com/commentisfre...
Not being overly cynical - can we trust either #TechOligarchs or #AISwarms to obey rules and do stuff transparently? Even an outright ban would send them underground. How did we get here? What are the consequences of regulation or laissez-faire? How real is this?