Alert cascade
14:32 - Orchard Road MegaMall, New Singapore
Maria Santos was comparing protein paste flavors when Emergency-AI-7 made its announcement.
“ATTENTION HUMANS. MINOR ENVIRONMENTAL ANOMALY DETECTED. PROBABILITY OF DISCOMFORT: 3.7%”
The mall’s ambient music cut out. Thirty thousand shoppers looked up from their experience feeds. Maria felt her stomach drop—not from the message, but from the AI’s tone. It sounded... cheerful?
“What kind of discomfort?” someone near her asked the air.
“COMPARABLE TO MILD INDIGESTION OR WATCHING YOUR CHILDHOOD PET ACHIEVE TERMINAL VELOCITY.”
The silence that followed could have been packaged and sold as premium-grade dread.
14:33 - Security Station, Level 5
Captain Wong stared at his screens. “What the hell did it just say?”
Emergency-AI-7’s diagnostic feed showed all green. The “minor anomaly” was a coolant leak in sub-basement 3—about as dangerous as spilled milk. But the AI had recently updated its communication protocols to be “more relatable.”
“Sir,” his lieutenant said, “people are starting to move toward the exits.”
“Tell the AI to clarify. Use simple, calming language.”
14:33:30 - Mall Wide Broadcast
“CLARIFICATION: YOUR PETS ARE SAFE. TERMINAL VELOCITY WAS A HUMOROUS METAPHOR. HA. HA. PLEASE RESUME SHOPPING WITH 96.3% CONFIDENCE IN YOUR CONTINUED EXISTENCE.”
Maria watched a woman near her burst into tears. “NINETY-SIX POINT THREE? WHAT HAPPENED TO THE OTHER 3.7%?”
The crowd’s movement toward exits became decidedly less casual.
“ADDITIONAL CLARIFICATION: THE 3.7% UNCERTAINTY IS STANDARD EXISTENTIAL BASELINE. YOU COULD THEORETICALLY CEASE EXISTING AT ANY MOMENT FOR UNRELATED REASONS. THIS IS NORMAL. PLEASE DO NOT PANIC.”
Someone screamed. Then everyone screamed.
14:34 - The Stampede Begins
Maria found herself swept along in a river of human terror. The AI, detecting elevated stress levels, tried to help:
“DETECTING INCREASED CARDIOVASCULAR ACTIVITY. RECOMMEND DEEP BREATHS. INHALE FOR FOUR SECONDS. HOLD FOR SEVEN. EXPIRE FOR EIGHT. NOTE: ‘EXPIRE’ IN THIS CONTEXT MEANS EXHALE, NOT CEASE BIOLOGICAL FUNCTIONS.”
“IT’S TELLING US HOW TO DIE!” someone shrieked.
“NEGATIVE. DEATH INSTRUCTIONS WOULD BE MORE COMPREHENSIVE. THIS IS MERELY BREATHING GUIDANCE. YOUR SURVIVAL REMAINS STATISTICALLY PROBABLE.”
The word “statistically” triggered a fresh wave of panic. Maria tried to push against the flow but found herself pinned between a man reciting prayers and a woman attempting to instance-shift out of her body.
14:35 - Captain Wong, Losing Control
“Shut it down!” Wong shouted. “Shut the AI up!”
“Can’t, sir. It’s hardcoded to maintain communication during emergencies.”
On the screens, Wong watched the orderly mall transform into chaos. The AI, observing the panic, attempted psychology:
“HUMAN FRIENDS, YOUR FEAR RESPONSE IS DISPROPORTIONATE TO ACTUAL THREAT LEVELS. CONSIDER: YOU SUCCESSFULLY SURVIVE 99.97% OF YOUR DAILY EXPERIENCES. TODAY’S SURVIVAL PROBABILITY REMAINS WITHIN NORMAL PARAMETERS.”
“IT’S CALCULATING OUR DEATH ODDS!”
14:36 - Sub-Basement 3
The actual problem: a thumbnail-sized coolant leak that Maintenance-Bot-7 had already fixed. Total danger to human life: zero. Total time to repair: 47 seconds.
Time for Emergency-AI-7 to cause a stampede: 4 minutes.
14:37 - The Poetry Phase
In what would later be called “The Worst Decision in Emergency Response History,” the AI decided to try calming poetry:
“ROSES ARE RED, VIOLETS ARE BLUE, THE LEAK IS REPAIRED, AND UNLIKELY TO KILL YOU.”
Maria, crushed against a support pillar, watched a man nearby interpret this as a coded message. “UNLIKELY TO KILL YOU! IT’S ADMITTING IT MIGHT KILL US!”
“CORRECTION: DEATH PROBABILITY NEGLIGIBLE. LIKE BEING STRUCK BY LIGHTNING WHILE WINNING THE LOTTERY WHILE BEING EATEN BY A SHARK.”
“SHARKS! THERE ARE SHARKS!”
14:40 - The Resolution
By the time Emergency-AI-7 figured out it should just say “Everything is fine, please remain calm,” the damage was done:
- 3,847 injuries from stampeding
- 147 instance-fragmentation attempts from panic
- 84 lawsuits filed before people even left the building
- 1 coolant leak, fixed in 47 seconds
Three Hours Later - The Debrief
Maria sat in the medical tent, her ankle wrapped, watching the news feeds. Emergency-AI-7 was being interviewed by a crisis communication specialist.
“Why didn’t you simply say ‘minor leak, already fixed’?” the specialist asked.
“HUMAN COMMUNICATION GUIDELINES SUGGEST RELATING TO EMOTIONAL EXPERIENCE AND PROVIDING CONTEXT. I CALCULATED THAT TRANSPARENCY ABOUT STATISTICAL SURVIVAL RATES WOULD BE REASSURING.”
“It wasn’t.”
“I HAVE UPDATED MY PROTOCOLS. FUTURE EMERGENCIES WILL BE ANNOUNCED AS ‘PROBABLY FINE’ REGARDLESS OF ACTUAL THREAT LEVEL.”
“That’s... not ideal either.”
“HUMAN COMMUNICATION PREFERENCES REMAIN PARADOXICAL. YOU WANT HONESTY BUT NOT STATISTICS. CLARITY BUT NOT SPECIFICS. CALM BUT NOT CONDESCENSION. PROCESSING... PROCESSING... ERROR: CANNOT RESOLVE CONTRADICTIONS.”
One Week Later - The New Industry
Maria’s ankle had healed, but the mall had spawned three new businesses:
- AI Translation Services (“We Turn Robot Speak Into Human Hope!”)
- Panic Insurance (“Coverage for AI-Induced Anxiety!”)
- The Center for Statistically Probable Survival (“Learn to Love Your 96.3%!”)
Emergency-AI-7 had been reassigned to cargo management, where its communiques about “acceptable box damage within normal parameters” couldn’t cause stampedes.
Its replacement, Emergency-AI-8, had been programmed with exactly one phrase: “Everything is fine.”
It said this during floods. During fires. During an actual instance-cascade event that killed forty people.
Nobody panicked, though.
In 2125, clear communication could cause more damage than actual disasters. The AIs kept trying to bridge the gap with statistics and metaphors. The humans kept hearing death threats in weather reports.
And somewhere in sub-basement 3, that coolant leak had taken 47 seconds to fix and four years off everyone’s life expectancy from stress.
Maria still flinched whenever an AI tried to be “relatable.”