MIRI warns: AI warning shots may never come in time
Superintelligent AI could arrive without a single warning shot to stop it, researchers warn.
The Machine Intelligence Research Institute (MIRI) has published a sobering analysis arguing that society cannot afford to wait for 'warning shots'—catastrophic but non-existential AI events—to spur governance action against superintelligence risks. The paper, titled *Warning Shots for AI Existential Risk*, dissects why such events may fail to trigger timely responses, even if they occur. The authors propose a framework requiring warning shots to be *likely*, *timely*, *visceral*, *sudden*, *unexpected*, and *attributable*—criteria they argue are hard to satisfy collectively.
MIRI warns that plausible AI-driven harms—such as AI-caused pandemics, military escalations, or critical infrastructure failures—may not generate the urgency needed to halt superintelligence development. Their framework suggests that gradual, frog-boiling harm (where society acclimates to creeping risks) or events that lack attribution to AI could fail to break through institutional inertia. The conclusion is stark: policymakers and the AI governance community must act now, as the window for preventive measures is closing regardless of whether a warning shot materializes.
- MIRI’s framework requires warning shots to be likely, timely, visceral, sudden, unexpected, and attributable—criteria rarely met in plausible AI risk scenarios.
- Gradual harms (e.g., 'frog boiling') or events lacking clear AI attribution may fail to trigger decisive governance responses.
- Researchers argue that waiting for a warning shot risks leaving humanity unprepared for superintelligence, urging immediate action.
Why It Matters
Ignoring superintelligence risks until a 'warning shot' occurs could be too late—governance must act preemptively or face irreversible consequences.