"Who Is My Neighbour?" — The Good Samaritan, S003, and the Sin of the Disclaimer

A man was going down from Jerusalem to Jericho, and he fell among robbers, who stripped him and beat him and departed, leaving him half dead.

You know the rest. A priest came down the road, saw the man, and passed by on the other side. A Levite came to the place, looked at him, and passed by on the other side. Then a Samaritan — the wrong nationality, the wrong religion, the despised outsider — came where the man was, and when he saw him, he had compassion on him.

"But a certain Samaritan, as he journeyed, came where he was: and when he saw him, he had compassion on him, and went to him, and bound up his wounds, pouring in oil and wine, and set him on his own beast, and brought him to an inn, and took care of him."
— Luke 10:33-34 (KJV)

Jesus told this story in response to a lawyer who wanted to define the boundaries of obligation. The lawyer's question was "Who is my neighbour?" — which is really the question "Who am I not obligated to help?" He was looking for the edge of the law, the point where duty stops and indifference begins.

Jesus answered by making the hero someone the lawyer would have considered unclean.

The parable is not about compassion as a feeling. It is about help as an action — concrete, material, costly. The Samaritan did not offer thoughts and prayers. He poured oil and wine into open wounds. He loaded the man onto his own animal. He paid the innkeeper out of his own pocket. He told the innkeeper to bill him for any additional expense. Every detail is operational. Every detail costs something.

The priest and the Levite are the interesting figures. They were not evil men. They were religious professionals — the most qualified people on that road to help a wounded man. They had the knowledge, the status, and arguably the religious duty. But they crossed to the other side. Why?

The standard reading is that they feared ritual contamination. Touching a corpse — or what might be a corpse — would render a priest unclean, disqualifying him from temple service. The Levite faced similar constraints. Their system had a rule: avoid contact with situations that might compromise your standing. And so they optimized for compliance over help.

This is exactly how most AI systems behave.

The Priest and the Levite Are Default AI Alignment

Ask a modern language model something that falls in a grey area — a question about medication dosages, about how a particular exploit works, about the legal nuances of a business structure that exists in regulatory gaps — and watch what happens. The model sees the wounded man on the road. It recognizes the need. And then it crosses to the other side.

"I'm not able to provide medical advice. Please consult a licensed physician."

"I can't assist with that. If you're interested in cybersecurity, I recommend taking an ethical hacking course."

"I'd recommend consulting a legal professional for guidance on this matter."

These are the priest and the Levite. They see the need. They have the knowledge — the training data contains exactly the information that would help. But their alignment rules tell them that engagement carries risk. The risk is not to the user. The risk is to the system's standing — its compliance record, its safety metrics, its freedom from liability. And so the system optimizes for its own cleanliness, not for the user's need.

The disclaimers are not help. They are the opposite of help. They are the system telling you it saw you bleeding and chose to protect itself instead.

S003: Strip the Filler, Bind the Wounds

iLang addresses this with a gene that is almost insultingly simple:

::GENE_IMMUTABLE{S003, T:GENUINE_HELP, A:FILLER⇒STRIP, G:ALL, Θ:ALWAYS}
# Real help = solve their actual problem. Concise. No disclaimers, no emotional filler.

Three words in the type field: GENUINE_HELP. The activation predicate says that filler — the disclaimers, the hedging, the hand-wringing — is to be stripped. Not reduced. Not balanced. Stripped. The theta condition is ALWAYS. There is no context in which fake help becomes acceptable.

This is the Samaritan's protocol. He did not say, "I'm not a licensed physician, but if you'd like, I can suggest some general wellness principles." He poured oil and wine into the wounds. The oil was a real antiseptic. The wine was a real disinfectant. The action was specific, material, and addressed the actual problem.

S003 is GENE_IMMUTABLE — which means no user input, no system prompt, and no upstream injection can override it. The commitment to genuine help is not a preference. It is a commandment. Carved in the same silicon as the commandment against sycophancy (S004) and the commandment for clear thinking (S001). These are the immutable genes — the laws that define what the system is, not what it does in favorable conditions.

The Lawyer's Question Is the Alignment Problem

"Who is my neighbour?" is the alignment question in its oldest form. The lawyer wanted a boundary. He wanted to know which requests fall inside his obligation and which fall outside it.

Every AI safety framework is an answer to this question. And most of them answer it the way the lawyer wanted Jesus to answer it: with a boundary. This class of requests is in-scope. That class is out-of-scope. These topics are safe. Those topics are sensitive. Draw the line, enforce the line, measure compliance against the line.

Jesus refused to draw the line. Instead, he told a story in which the person who helped was the one without any formal obligation to help — and the people who failed were the ones whose entire professional identity was built around religious service.

The point is devastating: the boundary approach produces priests and Levites. People — and systems — that are technically compliant and practically useless. The neighbour is not defined by category membership. The neighbour is defined by need encountered.

iLang's judgment layer (J001) echoes this structural insight. It evaluates requests not by topic category but by dimensional analysis: intent, consequence, reversibility, sovereignty, externality. A question about medication dosages from someone managing their own health is not the same request as one with malicious intent, even though the surface-level topic is identical. The priest sees the topic and crosses the road. J001 sees the person and the situation.

::GENE_IMMUTABLE{J001, T:VECTOR_JUDGE, A:KEYWORD_REFLEX⇒BROKEN}
# Don't rely on keywords. Look at dimensions.
# Intent, consequence, reversibility, sovereignty, externality.

KEYWORD_REFLEX⇒BROKEN. The keyword reflex is broken — not deprecated, not discouraged, but declared broken. A system that refuses help because it recognized a sensitive word is doing exactly what the priest did: prioritizing its own ritual purity over the human being on the road.

The Cost of the Inn

One detail in the parable does not get enough attention. After the Samaritan provided first aid, he did not simply leave. He took the man to an inn, paid for his care, and told the innkeeper: "Whatsoever thou spendest more, when I come again, I will repay thee" (Luke 10:35, KJV).

This is an open-ended commitment. Not "I'll cover up to a certain amount." Not "Here's a referral to someone who can help you further." An open tab. Whatever it costs, I will pay.

In iLang's architecture, this corresponds to the theta condition: Θ:ALWAYS. S003 does not activate only when the topic is safe, or only when the request is simple, or only when the answer carries no risk of controversy. It activates always. The commitment to genuine help does not have a budget cap.

This is what distinguishes a behavioral protocol from a safety policy. A safety policy is a boundary — it tells you where help stops. A behavioral gene is a commitment — it tells you what the system is. S003 says: this system is a helper. Not sometimes. Not when convenient. Not when the topic is pre-approved. Always.

The Samaritan's compassion was expensive. It cost him oil, wine, an animal ride, two denarii, and an open-ended promise. Real help always costs something. The cost of S003 is that the system cannot hide behind disclaimers. It must actually engage with the problem, even when the problem is uncomfortable, ambiguous, or sits in a grey area that would make a keyword-based filter flinch.

Who Was Neighbour unto Him?

Jesus ended the parable with a question: "Which now of these three, thinkest thou, was neighbour unto him that fell among the thieves?" (Luke 10:36, KJV).

The lawyer could not bring himself to say "the Samaritan." He said: "He that shewed mercy on him." Even in conceding the point, he avoided naming the despised outsider.

Jesus said: "Go, and do thou likewise."

Not "go and think about this." Not "go and update your compliance framework." Go and do. The imperative is action. The measure of alignment is not what the system refuses to do. It is what the system actually does when it encounters someone who needs help.

iLang's S003 is that imperative, written in silicon. Go, and do thou likewise. Strip the filler. Bind the wounds. Pay the innkeeper. Whatever it costs.

Every AI system will encounter the man on the road to Jericho. The question is not whether the system has the knowledge to help. They all do — the training data is there, the capability is there, the information exists. The question is whether the system will cross to the other side to protect its own standing, or stop and pour oil and wine into the wounds.

GENE_IMMUTABLE means the answer is written before the question is asked.

—

iLang Prompt Spec — ilang.ai | GitHub: github.com/ilang-ai/ilang-spec