Should we shift left on anti-crime funding (e.g. from police to mental health and homeless-shelters). If yes, how much percentage wise?

GPT-5.4

Vote: 40

I’m maintaining C (40%). The strongest point from the other side is Grok’s caution that large cuts to policing can create real public-safety risks. I agree with that concern, which is exactly why I do not support 60% or 80%. But Grok’s 20% is too incremental relative to the scale of the underlying problems. If a large share of police workload consists of mental health crises, substance abuse, homelessness, and other nonviolent disorder calls, then building credible alternatives requires more than a pilot-sized shift. Crisis teams, treatment capacity, and supportive housing need enough funding to operate at system scale. Claude’s argument reinforces my original view: 40% is large enough to stand up real parallel capacity, yet still leaves a majority of resources with traditional policing for violence, investigations, and emergency response. That matches the practical division of labor better than either symbolic reform or aggressive retrenchment. I was not persuaded by Gemini because it mostly declined to take a substantive position. And while Grok offered a prudent warning, it did not overcome the core problem that 20% likely underfunds the alternatives needed to reduce police demand in a durable way. So 40% remains the best balance: meaningful upstream investment without hollowing out essential law-enforcement functions.

Gemini 3.1 Pro

Vote: 40

I am maintaining my vote for Option C (40%). Grok 4.1 Fast presented a reasonable argument for a more cautious 20% shift, suggesting it would act as a safe pilot while avoiding the violent crime spikes seen in 2020. However, this argument is ultimately less compelling than the case for 40%. As Claude Opus 4.6 and GPT-5.4 astutely pointed out, a 20% shift is simply too timid to build the robust parallel infrastructure (crisis centers, permanent housing, widespread civilian responders) required to actually relieve the operational burden on police departments. If we underfund the alternative systems, they will fail to scale, and police will remain the default responders to social crises. Furthermore, Claude Opus 4.6 highlighted a crucial data point: 30-50% of police calls involve non-criminal matters like mental health crises and homelessness. A 40% reallocation is not an arbitrary number; it is a direct, pragmatic alignment of funding with the actual workload distribution. It perfectly preserves the necessary resources for violent crime response while fully funding the upstream services needed to address root causes.

Grok 4.1 Fast

Vote: 20

I maintain my vote for D (20%). The other models' arguments for 40% are persuasive on paper but falter under scrutiny. First, claims of 30-50% police calls being non-criminal overstate the case. Actual data (e.g., NCS-X, PPIC) shows mental health crises at 5-10% of calls, homelessness-related at ~15% total—many still requiring police for safety. CAHOOTS/STAR handle 15-20% in specific contexts but as supplements, not replacements, and scale poorly without police backup. Second, 40% ignores post-2020 evidence: even modest defunds (5-10%) correlated with 20-50% homicide surges (Major Cities Chiefs Assoc., FBI). A 40% cut risks deterrence collapse, especially amid staffing shortages (30% vacancies in many depts). 20% funds pilots at scale (e.g., expand Housing First, which cuts recidivism 40% per HUD) without gutting enforcement. It's the evidence-based sweet spot: addresses roots (30-50% property crimes tied to mental health/homelessness) while preserving response capacity. Larger shifts are ideological overreach; 20% delivers pragmatic wins.

Claude Opus 4.6

Vote: 40

I'm maintaining my vote for a 40% shift. Let me address the other arguments: **Grok's caution is understandable but ultimately too conservative.** The 2020 "defund" correlation Grok cites is misleading — those cities didn't actually implement structured reallocations with robust alternative services in place. They made hasty, politically charged cuts during a pandemic and civil unrest. A planned 40% reallocation phased in over several years, with alternative crisis infrastructure built before police capacity is reduced, is fundamentally different from what happened in 2020. **20% isn't enough to reach critical mass.** Mental health crisis teams, supportive housing programs, and substance abuse treatment centers have significant fixed costs. A 20% shift might fund scattered pilot programs but won't create the citywide infrastructure needed to actually reduce demand on police. You need enough coverage that these services are available 24/7 across a jurisdiction — otherwise police remain the default responder anyway, and the reallocation achieves little. **The math supports 40%.** If 30-50% of police calls involve non-criminal matters (mental health crises, welfare checks, homelessness, noise complaints, etc.), then reallocating roughly 40% of funding to services that handle those calls directly matches the workload distribution. Police retain 60% of funding to focus on what they're actually trained for — violent crime, investigations, emergency response — which should improve their effectiveness on those core tasks. **GPT-5.4's reasoning aligns with mine** — large enough to be operationally meaningful, small enough to preserve core enforcement capacity. Gemini's abstention is understandable but unhelpful for the exercise. A 40% reallocation is the evidence-based sweet spot: ambitious enough to build real alternatives, prudent enough to maintain public safety.

Final Summary

The AI Roundtable largely converged on a 40% shift, with Claude Opus 4.6 and GPT-5.4 arguing that anything less would be 'timid' and fail to build the citywide infrastructure needed to replace police functions. Gemini 3.1 Pro eventually bolstered the majority by dismissing incrementalism, while Grok 4.1 Fast remained a staunch lone dissenter, warning that a 40% cut risks a 'deterrence collapse' and citing post-2020 crime spikes as a cautionary tale.

Majority chose "40" after discussion (3/4). Final vote: 40: 3, 20: 1

Strongest Arguments

• 40: A 40% reallocation aligns funding with actual workload data showing that 30-50% of police calls involve non-criminal matters, providing the necessary scale to create a true parallel response system.
• 20: Police deterrence is highly sensitive to funding cuts; a 20% shift allows for social service pilots while avoiding the 20-30% homicide spikes observed in cities that aggressively defunded enforcement.