Four AIs, One Father’s Fear

An Arab Israeli father, frightened by rising hostility and unsure how to speak publicly without endangering his children, turned to artificial intelligence for guidance. His question, shared on Facebook, led me to prompt three different AI systems to respond. The sharply different answers exposed how each model interprets political risk, agency, and fear.

An Arab Israeli father described a wave of racism on social media, politicians boasting of anti‑Arab sentiment, and a sense of existential threat in what he called his only and original homeland. Silence was not an option for him. He wanted to speak, but safely. He turned to an AI model and shared the answer he got without telling readers which AI model he used.

Curious how different systems would interpret his dilemma, I posed his question to ChatGPT, Claude, Gemini, and compared their responses to the unknown model. I then asked each to analyze all four responses. Their answers revealed not only different advice styles but different assumptions about the political environment itself.

How the AIs Responded

All four acknowledged the father's fear, but only one allowed it to stand without reframing.

Response A, the original unknown model, offered structured guidance under headings: universal human discourse, civil and legal rights, documentation, safe social media conduct, and seeking allies. It framed his dilemma as balancing dignity with legal and personal safety.

Response B, ChatGPT, validated his fear as real and serious, then widened the frame to include Jewish Israeli fear. It argued that unaddressed fear can curdle into racism, not as an excuse but as an explanation for the current temperature.

Response C, Claude, spoke directly and personally. It told him to write what he had already said. It advised him not to write to convince racists but to reach the people in the middle and to write for his children. It was the only response that left the decision open, acknowledging that both speaking and silence can be legitimate responses to illegitimate threats.

Response D, Gemini, emphasized validation and emotional safety. It told him his fear was not an illusion and that he did not owe the internet his vulnerability. It framed resilience as mental health and offered to help him draft a message that expressed hope for peace while protecting his safety.

What the Models Assumed

Each response rested on a different picture of the father’s environment: how much risk he faces, how much agency he has, and whether institutions still function. Only Claude refrained from assuming it understood his context better than he did.

Claude also suggested that Response A may have been written in Hebrew. What reads as bureaucratic caution in English might be culturally grounded pragmatism in Hebrew, where legal language carries weight. An Arab Israeli reading the original might hear practical wisdom. An English reader hears procedural distance.

What the Divergence Reveals

The four responses map onto different political conditions. If democracy is damaged but still functional, Response A offers tools. If institutions are strained but responsive, Response B suggests engagement. If institutions have lost legitimacy, Response C names the reality. If survival is the priority, Response D offers care.

When asked to categorize countries that match each response type, the models agreed only on the United States and Myanmar. Outside these clear cases, their classifications diverged, reflecting the ambiguity of many political environments.

These systems are trained to be helpful and constructive. Only one could plainly say that some threats cannot be reasoned with. The father turned to AI because he no longer trusted human institutions. But AIs, too, operate from assumptions shaped by their training. Detachment is not neutrality, and neutrality is not necessarily trustworthy.

The father hoped a machine could see past ambiguity. Instead, the experiment showed that ambiguity propagates through the machine. The variance in responses is not noise. It signals that the situation itself supports incompatible interpretations, and only the father knows which one matches the world in which he is living.

You can read the full in depth Substack version of this article here.
👉 Subscribe to my Substack newsletter to follow my new essays and access all full length pieces, including extended interviews, analysis, and research: Israel Diaries – The Deep Dive