It's much easier to hold computers accountable than to hold humans accountable (andymasley.substack.com)

The piece argues that, contrary to a popular meme, computers are actually far easier to hold accountable than humans. Software and robots generate verifiable logs, are visibly branded, and can be monitored, audited, patched or rolled back in ways people cannot. Concrete examples include autonomous vehicles (Waymo’s branded cars are easier to identify and pressure than anonymous human drivers who cause many more animal and human injuries) and large-scale chat systems (companies can be targeted and forced to change model behavior when users are harmed, whereas the anonymous people spreading harmful content often go unpunished). The author reframes “accountability” as not only social punishment but technical control: detecting misbehavior with telemetry, applying fixes, and changing policies are far more effective for machines than relying on messy, inconsistent social sanctions for humans. For the AI/ML community this matters practically and ethically. Practically, it highlights leverage points—logging, monitoring, auditing, deployment controls and corporate incentives—that make system-level oversight effective and policy interventions feasible. Ethically, it warns of trade-offs: the same surveillance and reprogramming powers that enable accountability can become totalitarian if misused, so teams must balance robustness, transparency, and civil liberties. The takeaway: keep high technical accountability (instrumentation, audits, remediation pathways), but pair it with governance and safeguards to avoid authoritarian overreach.