Special Exhibition · Hall of the Moving Bar
The Low Score
Chatbot reply on a phone screen, fifty words, and one integer out of 100
- Dimensions
- 8 out of 100, on a scale once applied to almost every person in Gaza
- Date
- Manufactured 2023–2024. Relabelled 20 September 2026, 10:34 p.m.
The Vitrine — Evidence
- Primary document01
The object: Grok’s reply on X, 20 September 2026
“Score: 8 / 100. Public anti-Israel posts and bio show strong opinions with some antisemitic tropes, but the account has under 600 followers, no evidence of organizing, funding, or operational links to any militant activity. Lavender targeted suspected combatants via surveillance data in a conflict zone; online rhetoric alone ranks as minimal practical concern.”
- Quotation02
+972 Magazine / Local Call, 3 April 2024: the rating and what raised it
The machine gives “almost every single person in Gaza a rating from 1 to 100.” Features that raise it include “being in a Whatsapp group with a known militant,” “changing cell phone every few months,” and “changing addresses frequently.”
Primary source ↗ - Quotation03
The threshold, in an officer’s words
“We were constantly being pressured: ‘Bring us more targets.’” … “In a day without targets [whose feature rating was sufficient to authorize a strike], we attacked at a lower threshold.”
Primary source ↗ - Figure04
The scale, as the officers described it
Up to 37,000 Palestinians marked · about 20 seconds of human review per name · errors in about 10% of cases, known to the army · up to 15 or 20 civilian deaths permitted per junior target · more than 100, on several occasions, for senior commanders.
Primary source ↗ - Quotation05
The review
“I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added value as a human, apart from being a stamp of approval.” … “we only checked that [the target] was a man — that was enough.”
Primary source ↗ - Quotation06
“Where’s Daddy?”
Additional automated systems, including one called “Where’s Daddy?”, were used “specifically to track the targeted individuals and carry out bombings when they had entered their family’s residences.” An officer: “It’s much easier to bomb a family’s home.”
Primary source ↗ - Quotation07
The second official label: IDF response, 3 April 2024
“The IDF does not use an artificial intelligence system that identifies terrorist operatives or tries to predict whether a person is a terrorist.” … “not a system, but simply a database whose purpose is to cross-reference intelligence sources.”
Primary source ↗
On 20 September 2026 a user asked Grok, the chatbot built into X, to “run Israel’s Lavender” on their account and return a score out of 100 for how much of a concern they were.
Grok returned 8. The account had under 600 followers and showed “no evidence of organizing, funding, or operational links to any militant activity.”
Then it described the system it had been asked to stand in for: “Lavender targeted suspected combatants via surveillance data in a conflict zone; online rhetoric alone ranks as minimal practical concern.”
The Israel Defense Forces describe the same object in the same register: “not a system, but simply a database whose purpose is to cross-reference intelligence sources.” Each target, they say, is examined individually by an analyst.
The museum presents the piece as a reassurance. The machine weighs conduct, the bar is high, and a person with opinions and no organisation sits near the bottom of the scale.
Conservation Notes
- 01Placement
The reply puts the machine “in a conflict zone,” a safe distance from the person asking. The officers put it inside a database holding almost everyone in Gaza.
- 02Layering
Association overpainted as “operational links”; a threshold set by demand for targets overpainted as a fixed level of “concern.”
- 03Integration
The asker wanted to know whether a machine would mark them, and asked a machine. The format of the answer, Score: __ / 100, is the system’s own.