Pitfalls to avoid when using AI for the first time

Arti­fi­cial intel­li­gence has long been her­ald­ed as the tech­nol­o­gy of the future – although it is still the future if an inter­na­tion­al sur­vey of 700 busi­ness lead­ers by Juniper Net­works in April is any­thing to go by. Only 6% of respon­dents report­ed hav­ing adopt­ed AI, but 95% said that their firms would ben­e­fit from doing so. 

A sim­i­lar­ly sized sur­vey of IT deci­sion-mak­ers by Insid­er in the same month found that a third of respon­dents were plan­ning invest­ments in AI. 

While many busi­ness­es are clear­ly keen to start using the tech­nol­o­gy, experts warn that they need to intro­duce it judi­cious­ly. Firms may well have more pit­falls to avoid than ben­e­fits to reap, so it’s vital to learn from pre­vi­ous AI inte­gra­tions else­where. 

BT has been using var­i­ous gen­er­a­tions of the tech­nol­o­gy for some time, accord­ing to Paul O’Brien, direc­tor of AI, ser­vice, secu­ri­ty and oper­a­tions research. Today, the com­pa­ny utilis­es tech rang­ing from neur­al net­works and deep learn­ing to evo­lu­tion­ary com­put­ing and heuris­tics in the effort to stream­line its oper­a­tions. 

“AI improves the way the com­pa­ny man­ages its net­works and ser­vices,” O’Brien says. “The tech­nol­o­gy auto­mates rou­tine tasks and aug­ments people’s capa­bil­i­ties with smart insights and sup­port.” 

AI plays a role in many of BT’s activ­i­ties, from plan­ning where next to install its fibre broad­band net­work to han­dling the 27,000-strong team of field engi­neers and their vehi­cles. It helps man­agers to pre­dict line fault vol­umes, organ­ise ros­ters and sched­ule work. It’s even being used by tech­ni­cians to con­struct a dig­i­tal twin (a vir­tu­al sim­u­la­tion) of the nation­al phone net­work

No silver bullet

Despite his company’s suc­cess­ful appli­ca­tions of the tech­nol­o­gy, O’Brien warns prospec­tive AI adopters to inte­grate it into their exist­ing sys­tems with great care. 

“There is too much hype around AI, which rais­es expec­ta­tions and leads to mis­un­der­stand­ings of what it can do,” he says, adding that its per­for­mance can depend heav­i­ly on the qual­i­ty of the data fed into it.

Dr Catri­ona Wolfend­en, part­ner and inno­va­tion man­ag­er at law firm Weight­mans, agrees. “Many peo­ple fall into the trap of think­ing that AI is some kind of mag­ic wand that’s sim­ply going to fix all ills. It’s not at all,” she says. “You need to ensure that you’re using AI on the right kind of thing and you must get the under­ly­ing prin­ci­ples and data col­lec­tion right.” 

The lim­i­ta­tions of AI and the built-in bias­es that have dogged cer­tain sys­tems have attract­ed plen­ty of neg­a­tive head­lines. This risk fac­tor has encour­aged firms such as EY to employ AI ethics experts in senior posi­tions as a mit­i­ga­tion mea­sure. Ensur­ing that the tech­nol­o­gy is used eth­i­cal­ly is vital. But, while 87% of respon­dents to the Juniper Net­works sur­vey agreed that their firms need­ed to put prop­er gov­er­nance poli­cies in place to min­imise any harm result­ing from the use of AI, this task ranked as one of their low­est pri­or­i­ties in the adop­tion process. 

“Com­pa­nies need domain exper­tise on board, so that they can under­stand how to both exploit AI and under­stand its lim­i­ta­tions,” O’Brien advis­es.

Start AI on the mundane work

Wolfend­en has been care­ful to inte­grate AI into Weight­mans’ work grad­u­al­ly. “We’ve tak­en a very con­scious approach that it’s about the aug­men­ta­tion of a lawyer’s skill,” she says. “It’s there to enhance the professional’s exper­tise, not to replace them.” 

AI was first applied in a chat­bot for inter­nal use. “We start­ed real­ly small to sur­vey the mar­ket and pick a use case,” Wolfend­en recalls. 

Once that appli­ca­tion had proved its worth, she applied the tech to anoth­er straight­for­ward task: pulling data from one set of files and insert­ing it into anoth­er. 

“It’s real­ly easy at the start to get car­ried away and think you’re going to do every­thing with AI,” she warns. “Have the idea and scale it back – prob­a­bly twen­ty­fold – to begin in the right place.” 

The firm was also pru­dent in how it pre­sent­ed the tech­nol­o­gy to its clients, know­ing that they might wor­ry that their high­ly paid solic­i­tor was being replaced by a machine. 

“A lot of it is down to care­ful mes­sag­ing, both inter­nal­ly and exter­nal­ly,” Wolfend­en says. “We pro­vide fact sheets that explain why we would use AI, stress­ing that humans are still involved.”

Cary Coop­er, pro­fes­sor of organ­i­sa­tion­al psy­chol­o­gy and health at Man­ches­ter Busi­ness School, advis­es firms intro­duc­ing AI to “engage the work­ers with the process, rather than impose it. Get­ting them to come up with the solu­tion so that it works for both them and the busi­ness is the most effec­tive strat­e­gy.” 

Sim­ply foist­ing AI on your staff is a sure-fire way to trig­ger resent­ment, Coop­er warns. “That will cre­ate uncer­tain­ty and inse­cu­ri­ty, moti­vat­ing them to find ways to make it less effec­tive,” he says. “On the oth­er hand, if employ­ees have own­er­ship in the intro­duc­tion of AI, they won’t try to under­mine it. They will make it work and it’s like­ly to prove more pro­duc­tive.”