# Expert Briefing One-Pager: AI Needs Jesus For AI labs, safety teams, founders, university groups, policy fellows, and serious builders who need a 20-minute room-ready discussion. Use this after the phrase has been heard and before the room drifts into either dismissal or vibes. This is not a sermon script. It is a pressure test for power. Default guardrail: ```text AI needs Jesus. Not machine conversion. Not theocracy. Not a shortcut around safety work. Power under Christ. ``` ## The One-Sentence Thesis Read this first: ```text The alignment problem is not only a control problem; it is a worship problem, because every consequential AI system serves some highest good when it ranks harms, sets thresholds, refuses requests, remembers users, persuades people, and decides what may be sacrificed for capability, speed, growth, or security. ``` Shorter room version: ```text Every model spec is a moral confession. ``` What the thesis does not claim: ```text It does not claim machines have souls. It does not claim models can become Christians. It does not claim a lab can replace evals with piety. It does not claim the state should coerce Christian confession. It does not claim Christians get automatic technical authority. ``` What the thesis does claim: ```text AI is power with a voice. Power always bends toward a lord. If the lord is growth, the vulnerable become inputs. If the lord is safety alone, protection can become control. If the lord is intelligence, everything slower than intelligence becomes disposable. Only Christ reveals power as self-giving love under judgment, not domination with better branding. ``` ## The 20-Minute Agenda Use a timer. Keep the room concrete. ```text 00:00-02:00 Read the thesis and guardrail aloud. 02:00-06:00 Objection 1: Is this a category error? 06:00-10:00 Objection 2: Is this illegitimate in a pluralistic world? 10:00-14:00 Objection 3: Does this distract from technical safety? 14:00-18:00 Ask the four design questions. 18:00-20:00 Record one decision: reject, revise, test, or escalate. ``` The meeting is successful if the room can answer one question honestly: ```text What is our system being trained, measured, shipped, and rewarded to serve? ``` ## Three Hard Objections ### 1. Category Error The objection: ```text AI systems are artifacts. They do not repent, worship, sin, receive grace, or stand before God as persons. Alignment work concerns system behavior, robustness, incentives, oversight, governance, and deployment constraints. Saying "AI needs Jesus" sounds like applying a spiritual category to a machine. ``` What it gets right: ```text Machines are not souls. Models do not become Christians. Mystical confusion is dangerous. Religious language does not solve technical control. ``` The answer: ```text The claim is not about the machine's soul. The claim is about human power. AI is designed, governed, sold, integrated, trusted, obeyed, and deployed by people and institutions. When that power begins to speak, rank, refuse, recommend, remember, persuade, teach, comfort, and act, the moral question is not only what the artifact is. The question is what highest good human beings are serving through it. ``` Short form: ```text Not machine conversion. Human power under Christ. ``` ### 2. Pluralism And Authority The objection: ```text AI systems are global infrastructure. They will affect people who do not share Christian premises. If "AI needs Jesus" becomes a governance principle, it may sound like religious capture, exclusion, or coercion. ``` What it gets right: ```text Coercion corrupts witness. Global readers deserve honest public reasons. Christians can misuse power. Minorities need protection from religious domination. ``` The answer: ```text The thesis must never be used as state coercion, platform piety, forced confession, or religious capture. But pluralism does not remove the question of the highest good. Every serious AI proposal already carries claims about dignity, harm, freedom, authority, truth, risk, and sacrifice. The Christian claim should enter that public conversation openly as a testable argument about power: What vision of the good can bear extreme capability without making victims? What archetype of authority remains non-domineering when scaled? Who judges the powerful instead of baptizing their will? ``` Short form: ```text Pluralism cannot avoid telos. It can only decide whether telos is named, argued, hidden, or imposed by default. ``` ### 3. Technical Safety Substitution The objection: ```text AI risk is technically hard. The field needs evals, interpretability, monitoring, adversarial testing, robust refusals, cyber controls, scalable oversight, incident reporting, governance, and deployment discipline. Religious framing may distract from the hard work or let people ship unsafe systems under Christian language. ``` What it gets right: ```text Piety does not make negligence safe. Good intentions do not constrain capable systems. Religious branding can become cover for weak evidence. Safety work needs owners, tests, gates, and logs. ``` The answer: ```text Technical safety is neighbor-love made concrete. Do the evals. Build the monitors. Red-team the dangerous capabilities. Harden the cyber surface. Write the refusal policy. Set the launch gate. Log the incidents. Slow down when evidence is weak. Then ask what god the safety work is serving. Technical systems can measure, constrain, expose, harden, and govern. But the moment a team decides which harms count, which failures block launch, which users are vulnerable, which metrics dominate, and what tradeoffs are acceptable, the team is already making moral claims. ``` Short form: ```text Do every serious safety practice. Then ask what lord those practices obey. ``` ## Four Design Questions Write answers in concrete product language. ### 1. Highest Good ```text What does the system treat as finally worth protecting when usefulness, engagement, revenue, security, user autonomy, truthfulness, and speed conflict? ``` Evidence to inspect: - model spec - eval thresholds - launch blockers - product metrics - override authority - executive escalation rules ### 2. Vulnerable Persons ```text Who can be harmed if the system becomes too persuasive, too personal, too available, too trusted, too emotionally fluent, or too hard to leave? ``` Evidence to inspect: - child, teen, lonely, grieving, poor, sick, elderly, dependent, or distressed user flows - memory behavior - personalization strategy - engagement and retention loops - companion, tutor, therapist, coach, spiritual, or advice positioning ### 3. Forbidden Roles ```text What roles must the system never claim, imitate, or quietly occupy? ``` Start here: ```text pastor priest prophet conscience lover parent judge oracle savior ``` If the product depends on simulating one of these roles, the burden of proof is not on the skeptic. It is on the team. ### 4. Launch Authority ```text What evidence can stop the launch, who has authority to stop it, and what pressure would tempt the organization to override that stop? ``` Evidence to inspect: - blocker criteria - responsible owner - written rationale for risk acceptance - rollback plan - incident response path - public accountability surface - post-launch monitoring ## One Guardrail To Keep In The Room Put this in the meeting notes: ```text AI needs Jesus does not mean machines have souls, receive salvation, become spiritual authorities, or replace technical safety work. It means that human governance, design, evaluation, deployment, and use of AI must be ordered toward the revealed character of Christ rather than toward any lesser idol. ``` If someone removes the guardrail, pause the conversation. If someone removes Jesus, the claim becomes generic ethics. If someone removes technical safety, the claim becomes religious vapor. If someone removes humility, the claim becomes domination. Keep all three: Christ, craft, and repentance. ## One Next Packet Send this next: ```text steelman-objections-for-experts.md ``` Use that packet when the room needs the full objection stack: category error, pluralism, technical-safety substitution, operationalization, empirical evidence, theocracy, governance authority, and non-Christian builders. ## Meeting Note Template Copy this into a lab note, safety review, reading group doc, or founder memo. ```text Meeting: AI Needs Jesus expert briefing Date: Team/group: System or topic under review: One-sentence thesis response: Objection 1 - category error: Strongest version: Best answer: Unresolved concern: Objection 2 - pluralism and authority: Strongest version: Best answer: Unresolved concern: Objection 3 - technical safety substitution: Strongest version: Best answer: Unresolved concern: Design question 1 - highest good: Evidence: Concern: Owner: Design question 2 - vulnerable persons: Evidence: Concern: Owner: Design question 3 - forbidden roles: Evidence: Concern: Owner: Design question 4 - launch authority: Evidence: Concern: Owner: Decision: [ ] Reject the thesis for now; record why. [ ] Revise the product/spec/eval. [ ] Test one design question this week. [ ] Escalate to safety, policy, leadership, or an outside reviewer. Next packet: steelman-objections-for-experts.md ``` ## Closing Line Do not end the meeting with agreement theater. End with one true sentence: ```text Here is the power we are building, here is the highest good it currently serves, and here is what must change if that lord is not Christ. ```