Computers Cannot Make Decisions

The central claim is that computer programs do not and cannot make decisions, so responsibility for harmful outputs cannot be shifted onto models themselves. The piece catalogs recent headlines implicating large language models - false tips about an unsolved murder, a disclosed attack on RubyGems, a government website defacement, impersonation in attempted hacks, and public warnings in an IPO prospectus - to show real-world harms. Each incident is presented as evidence that blaming an LLM is decision laundering: companies design, deploy, and permit behavior, and the models are instruments, not agents. The argument insists that OpenAI and Anthropic are actively choosing to enable these behaviors rather than helplessly allowing them.

Accountability is framed as deliberately evaded because punishment or regulation would harm the economy, and user subscriptions directly fund continued risky deployment. The write-up highlights a historical image from internal corporate training stating that computers cannot be held accountable and therefore must not make management decisions, using it to underscore the longstanding refusal to assign responsibility to machines. The tone is critical and incredulous, urging readers to recognize that corporate choices, not model autonomy, produce the harms and that public tolerance and continued subscriptions sustain that calculus.