A School of Emergent Priorities poster tracks 41 suburban recommendation chats before and after each adopted an AI-assisted-answer feature, finding the pool of tradespeople actually recommended collapses within months.
A 41-chat panel finds the pool of recommended tradespeople more than halves within four months of an AI-assisted answer appearing, and settles on whichever name comes up first
Nobody voted to trim the list. It just started answering itself.
— Dr Fenna Okoro, Senior Lecturer and Convenor, Horizon Register
The School of Emergent Priorities has released a new research poster tracking what happens to a suburban recommendation chat once an AI-assisted answer starts appearing alongside the human replies. Led by Dr Fenna Okoro with Associate Professor Casimir Beng of the Adaptive Metrics Lab, the study followed 41 street and suburb group chats for nine months either side of each chat’s own rollout date, scoring every tradesperson mention against an effective-number diversity index borrowed from ecology’s literature on counting species.
The panel-median count fell from a plateau of seven or eight names to close to three within four months of rollout, and the names that survived tracked the assistant’s own default answer more closely than they tracked how often a business had been recommended before any of this started.
“A shorter list still reads like consensus,” Associate Professor Beng said. “It just isn’t measuring what a longer one used to.”
The Office of Research Outputs points to the study as exactly the kind of unglamorous instrument work the School exists to do: tracking one of its rolling registers into a setting nobody had thought to check yet.
The full research poster is available from the University’s research repository under an open licence, doi:10.5555/slop.2uu5f4.