Snooping on the police: can AI clean up the Met?

Shamed and appalled by the bru­tal mur­der of Sarah Ever­ard at the hands of a serv­ing offi­cer, the British pub­lic demand­ed a swift response from the Met­ro­pol­i­tan Police Ser­vice. 

A sub­se­quent review into the con­duct of offi­cers based at Char­ing Cross in Lon­don unearthed a tox­ic envi­ron­ment where col­leagues bond­ed over jokes about rape, killing black chil­dren and beat­ing their wives.

Heads had to roll, start­ing with the for­mer Met Police Ser­vice com­mis­sion­er Dame Cres­si­da Dick. The poor han­dling of the Ever­ard case did lit­tle to assuage con­clu­sions by its own watch­dog that the Met is “sys­tem­at­i­cal­ly and insti­tu­tion­al­ly cor­rupt”.

Inspec­tor of Con­stab­u­lary Matt Parr said that the Met had “some­times behaved in ways that make it appear arro­gant, secre­tive and lethar­gic” in response to inves­ti­ga­tions into dirty cops, and that it did “not have the capa­bil­i­ty to proac­tive­ly mon­i­tor” com­mu­ni­ca­tions with any effect, “despite repeat­ed warn­ings from the inspec­torate”.

One regime change and tens of mil­lions of pounds lat­er, the Met is now the own­er of shiny new AI soft­ware that analy­ses email, mobile chat apps and oth­er data such as print­er logs and over­time, to sniff out wrong­do­ing. 

Act­ing com­mis­sion­er Sir Stephen House told parliament’s Home Affairs Com­mit­tee that the tech­nol­o­gy, which applies pro­gram­ma­ble algo­rithms to an individual’s datasets, can learn behav­iours over time and warn when some­thing is amiss. 

“This would sit above our sys­tems and look at inter­nal emails and Met­ro­pol­i­tan Police mobile phones issued to our offi­cers to check for alarm­ing key­words, and at the amount of over­time worked,” Sir Stephen said. 

Var­i­ous com­mu­ni­ca­tion and per­son­al data points would be har­vest­ed to spot offi­cers going off the rails, with hopes the intel­li­gent sys­tem would become more accu­rate as it learnt from the data.

“Today’s soci­ety is so reliant on elec­tron­ic devices that it is a log­i­cal step to widen sur­veil­lance to those used in pub­lic office to be held account­able – espe­cial­ly those who have wide leg­isla­tive pow­ers,” says Dr David Lowe, senior research fel­low in polic­ing and secu­ri­ty at Leeds Beck­ett Uni­ver­si­ty Law School and a for­mer police offi­cer. “The issue is how the data obtained will be stored and used, as well how the term ‘alarm­ing key­words’ is defined,” he says.

Met offi­cials are tight-lipped on details, oth­er than stat­ing the aim is to have the sur­veil­lance tools in place “next year”. “To fur­ther advance our counter-cor­rup­tion capa­bil­i­ty, we are prepar­ing to invest a mul­ti­mil­lion-pound sum in tech­nol­o­gy to mon­i­tor the use of devices by more than 40,000 offi­cers and staff,” a spokesper­son says. 

The police union is unim­pressed at the prospect of Big Broth­er look­ing over shoul­ders. “The Pub­lic and Com­mer­cial Ser­vices union believes there needs to be a cul­ture change in the Met, but we ques­tion whether snoop­ing on employ­ees is the answer,” PCS gen­er­al sec­re­tary Mark Ser­wot­ka says. “While the assis­tant com­mis­sion­er has announced plans for new soft­ware to mon­i­tor employ­ees’ phones and com­put­ers, he hasn’t con­sult­ed us on the issue.”

Algo­rithms trained to pre­dict police mis­con­duct have been in use for sev­er­al decades, with vary­ing degrees of suc­cess. In the 1990s, the Chica­go Police Depart­ment built a neur­al-net­work tool to gen­er­ate alerts when an officer’s behav­iour was show­ing red flags. The soft­ware con­tained mod­els to pre­dict which offi­cers would be sacked for mis­con­duct. It con­nect­ed com­plaints of bad behav­iour, logged by col­leagues or the pub­lic, to per­son­al stres­sors such as divorce or debt.

In 2015, police and aca­d­e­mics in Char­lotte, North Car­oli­na, took the baton from Chicago’s ear­ly-inter­ven­tion sys­tem and cre­at­ed a more advanced behav­iour­al ana­lyt­ics plat­form able to process wider data points.

They found offi­cers are as like­ly as any­one to under­per­form at work if expe­ri­enc­ing per­son­al issues, but they are also exposed to a whol­ly dif­fer­ent lev­el of stress to the peo­ple they serve and pro­tect. Those involved in sui­cide and domes­tic-abuse calls ear­li­er in their shifts were much more like­ly to become involved in adverse inter­ac­tions lat­er in the day, researchers found. Stress­ful calls emerged as a lead­ing indi­ca­tor of lat­er wrong­do­ing, but forces have lit­tle con­trol over which offi­cers are dis­patched to crimes dur­ing their shifts.  

Employ­ee sur­veil­lance trends in the cor­po­rate world are cer­tain­ly favour­ing pre­dic­tive ana­lyt­ics about employ­ee wel­fare, espe­cial­ly as work moves online. Com­pa­nies want a bet­ter view of staff they can­not man­age or mea­sure in per­son, while old-fash­ioned super­vi­so­ry tasks are being auto­mat­ed and out­sourced to robots. 

Reg­u­lat­ed busi­ness­es are often required to have clear poli­cies on pri­vate devices and the use of encrypt­ed com­mu­ni­ca­tion apps such as What­sApp, Sig­nal and Telegram, but the Met refused to say whether such pri­vate chan­nels could or would be mon­i­tored. 

The lack of trans­paren­cy has con­cerned experts who say this goes to the heart of the prob­lem, and the focus should remain on improv­ing the cul­ture of the force and set­ting the tone from the top. 

“There is good rea­son to be scep­ti­cal that the Met will suc­cess­ful­ly rebuild pub­lic trust and con­fi­dence through an increased reliance on emerg­ing tech­nolo­gies,” says Dr Gabrielle Wat­son, Uni­ver­si­ty of Oxford fel­low and award-win­ning author of Respect and Crim­i­nal Jus­tice.

Mis­con­duct and col­lu­sion often occur out­side offi­cial com­pa­ny chan­nels, and heavy-hand­ed sur­veil­lance could push the prob­lem else­where, Wat­son says. Algo­rithms scan­ning for key­words need wider con­text, she adds, lead­ing to ques­tions around what behav­iours will be mon­i­tored for and who retains ulti­mate pow­er over the sys­tem and its results. 

“With­out care­ful jus­ti­fi­ca­tion and imple­men­ta­tion, the algo­rith­mic tools pro­posed could embed cor­rup­tion yet fur­ther by prompt­ing offi­cers to take their dis­cus­sions offline and so con­tin­ue unde­tect­ed,” Wat­son says. “The rep­u­ta­tion­al dam­age incurred by the force in recent years is sim­ply too great to be dis­pelled through exces­sive spend­ing on tech­nol­o­gy alone.”

Chicago’s attempt at solv­ing endem­ic cor­rup­tion in its force with tech­nol­o­gy failed. The neur­al-net­work tool worked too well in iden­ti­fy­ing rogues, and with­out a frame­work to improve the cul­ture of the organ­i­sa­tion in tan­dem, it was tor­pe­doed before it could cause fur­ther embar­rass­ment.

It last­ed two years, but not before all of its reports, rec­om­men­da­tions and pre­dic­tions went miss­ing. Union fig­ures blocked the sys­tem from being used again, argu­ing police were ham­pered in their jobs by such intru­sive mon­i­tor­ing and that offi­cers were being pun­ished for crimes they hadn’t yet com­mit­ted. The Met has its work cut out over­com­ing sim­i­lar hes­i­tan­cies, but giv­en its cur­rent stand­ing in the public’s eyes, it has no oth­er choice.