- Published
In his award-winning essay, young British science writer Adrian Dredge wrote about the potential pitfalls of AI and policing.
asserted
Dredge → publish → AI
The essay won the 2026 Young Science Writer Award, external, chosen each year by the Association of British Science Writers.
asserted
essay → win → Writers
Here is an edited version of the entry.
asserted
version → edit → entry
The growing use of AI in policing risks increasing discrimination against minority groups, researchers have warned.
asserted
researchers → grow → groups
Studies suggest that predictions of future crime hotspots may be based on biased data, which can magnify existing prejudice and create a dangerous feedback loop.
uncertain
which → suggest → loop
Experts are calling for the use of AI in policing to be reviewed with greater human oversight.
asserted
use → call → oversight
This would involve checking the data that forces use to find and remove any bias, according to Dr Carolyn McKay, an associate professor and researcher in digital justice at the University of Sydney Law School.
uncertain
forces → involve → School
"The public functions of state policing ought to be subject to human oversight and scrutiny," she says.
asserted
she → say → oversight
The problem, according to McKay, is that AI systems are trained on data that already contains human bias, and this can lead to self-fulfilling prophecies.
uncertain
this → accord → prophecies
Think of AI as a map-maker asked to draw a map of a city they have never visited.
asserted
they → think → city
They are locked in a windowless room and can't go outside.
asserted
They → lock → room
All they have are maps that are 50 years old.
asserted
that → have → ?
Anyone following the new map would find themselves stopping at walls that were knocked down years ago and looking for roads that no longer exist.
asserted
that → follow → roads
The map-maker hasn't made anything new.
asserted
maker → make → anything
They have only copied the old maps.
asserted
They → copy → maps
Now suppose the map-maker is handed a list of past crimes and where they took place, and asked to shade in where the next ones are likely to happen.
asserted
ones → suppose → place
Police then patrol the shaded areas.
asserted
Police → patrol → areas
More crime is recorded there simply because more officers are there to see it.
asserted
officers → record → it
Areas outside the shading appear to have less crime, because fewer people are looking.
asserted
people → appear → crime
When the next map is drawn, the same areas are shaded even darker.
asserted
areas → draw → ?
They are partly a record of where police have chosen to look.
asserted
police → choose → ?
The problem is as old as policing, but AI may be speeding it up.
uncertain
AI → speed → it
Sarah Brayne, a sociologist at the University of Texas at Austin, spent time inside the Los Angeles Police Department studying how it uses big data.
asserted
it → spend → data
In her research paper, published in the American Sociological Review, she found that the force increasingly used data to predict crime, rather than only to respond to it.
asserted
force → publish → it
She also found that the data fed into these systems was skewed.
asserted
data → find → systems
Crimes in public places were more likely to be seen and recorded, and police focus was disproportionately on black neighbourhoods.
asserted
focus → see → neighbourhoods
Once this goes into the system, she writes, "the predictions appear impartial".
The models, she says, create "a feedback loop in which they not only predict events such as crime or police contact, but also contribute to their future occurrence."
But that is a view challenged by Dr Dimitrios Kollias, an associate professor in multimodal artificial intelligence at the Queen Mary University of London (QMUL).
asserted
that → go → London
He says that the technology has moved on to take the bias of historical data into account.
asserted
technology → say → account
"The old-map analogy is useful for explaining how historical bias can be inherited, but it has its limits," he told BBC News.
asserted
he → explain → News
"Modern AI systems do not simply retrieve or copy their training examples; they can combine learned information in new ways and generate outputs and solutions that were not explicitly present in those examples.
asserted
that → retrieve → examples
"
But those new insights pose a separate risk, Kollias adds.
asserted
Kollias → pose → risk
"As AI systems become more capable of generating novel decisions and acting with greater autonomy, we need to ensure that their behaviour remains safe, reliable and under meaningful human control," he said.
asserted
he → become → control
But arguably a more intractable problem is dealing with the sometimes flawed information stored by authorities and which AI systems draw from.
asserted
systems → deal → which
Researchers at New York University said that many of their predictions may rely on what they call "dirty data".
uncertain
they → say → what
These are police records shaped, they say, by corrupt, biased or unlawful practices, including rigged crime figures and arrests of innocent people.
asserted
they → shape → people
In their study, the researchers found that in nine of the 13 US areas they looked at, police data from periods of unlawful or biased policing was available to train these systems.
asserted
data → find → systems
The system can't tell that this data is flawed, according to the researchers.
uncertain
data → tell → researchers
It simply treats it as a pattern to repeat.
asserted
It → treat → pattern
It is as if the map-maker's compass points only to where crime has been recorded before.
asserted
crime → point → ?
The map-maker thinks they have found a new crime hotspot, when all they have done is follow the bias of earlier officers.
asserted
they → think → officers
…and 39 more, not listed.
The Aporia
