AI, Au­to­ma­tion and the Fu­ture of Work

There is work for ev­ery­one to­day and there will be work for ev­ery­one to­mor­row, but that work will re­quire new skills and a high de­gree of adapt­abil­ity.

Rotman Management Magazine - - CONTENTS - By James Manyika and Kevin Sneader

There is work for ev­ery­one to­day and there will be work for ev­ery­one to­mor­row. But that work will re­quire new skills and a high de­gree of adapt­abil­ity.

AU­TO­MA­TION AND AR­TI­FI­CIAL IN­TEL­LI­GENCE are trans­form­ing busi­nesses and will con­trib­ute sig­nif­i­cantly to eco­nomic growth via con­tri­bu­tions to pro­duc­tiv­ity. Th­ese tech­nolo­gies will trans­form the very na­ture of ‘work’ and the work­place it­self. Machines will be able to carry out more of the tasks done by hu­mans, com­ple­ment the work that hu­mans do, and even per­form some tasks that go be­yond what hu­mans can do. As a re­sult, some oc­cu­pa­tions will de­cline, oth­ers will grow, and many more will change.

While we be­lieve there will be enough work to go around (bar­ring ex­treme sce­nar­ios), so­ci­ety will need to grap­ple with sig­nif­i­cant work­force tran­si­tions and dis­lo­ca­tion. Work­ers will need to ac­quire new skills and adapt to the in­creas­ingly ca­pa­ble machines along­side them in the work­place. They may have to move from de­clin­ing oc­cu­pa­tions to grow­ing and, in some cases, brand new ones.

In this ar­ti­cle we will ex­am­ine both the prom­ise and the chal­lenge of au­to­ma­tion and AI in the work­place and out­line some of the crit­i­cal is­sues that pol­i­cy­mak­ers, com­pa­nies and in­di­vid­u­als need to con­sider.

Op­por­tu­ni­ties Ahead

Au­to­ma­tion and AI are not new, but re­cent tech­no­log­i­cal progress is push­ing the fron­tier of what machines can do. Our re­search sug­gests that so­ci­ety needs th­ese im­prove­ments to pro­vide value for busi­nesses, con­trib­ute to eco­nomic growth, and make once unimag­in­able progress on some of our most dif­fi­cult so­ci­etal chal­lenges. Fol­low­ing are some of the key op­por­tu­ni­ties that lie ahead.

RAPID TECH­NO­LOG­I­CAL PROGRESS. Be­yond tra­di­tional in­dus­trial au­to­ma­tion and ad­vanced ro­bots, new gen­er­a­tions of more ca­pa­ble au­ton­o­mous sys­tems are ap­pear­ing in en­vi­ron­ments rang­ing from au­ton­o­mous ve­hi­cles to au­to­mated check-outs in grocery stores. Much of this progress has been driven by im­prove­ments in sys­tems and com­po­nents, in­clud­ing me­chan­ics, sen­sors and soft­ware. AI has made es­pe­cially big strides in re­cent years, as ma­chine learn­ing al­go­rithms have be­come more so­phis­ti­cated and made use of huge in­creases in com­put­ing power and of the ex­po­nen­tial growth in data avail­able to train

al­go­rithms. Spec­tac­u­lar break­throughs are mak­ing head­lines, many in­volv­ing be­yond-hu­man ca­pa­bil­i­ties in com­puter vi­sion, nat­u­ral lan­guage pro­cess­ing and com­plex games such as Go.

PO­TEN­TIAL TO CON­TRIB­UTE TO ECO­NOMIC GROWTH Th­ese tech­nolo­gies are al­ready gen­er­at­ing value in var­i­ous prod­ucts and ser­vices, and com­pa­nies across sec­tors use them in an ar­ray of pro­cesses to per­son­al­ize prod­uct rec­om­men­da­tions, find anom­alies in pro­duc­tion, iden­tify fraud­u­lent trans­ac­tions and more. The lat­est gen­er­a­tion of AI ad­vances, in­clud­ing tech­niques that ad­dress classification, es­ti­ma­tion and clus­ter­ing prob­lems, prom­ises sig­nif­i­cantly more value still. An anal­y­sis we con­ducted of sev­eral hun­dred AI use cases found that the most ad­vanced deep learn­ing tech­niques de­ploy­ing ar­ti­fi­cial neu­ral net­works could ac­count for as much as US$ 3.5 tril­lion to US$ 5.8 tril­lion in an­nual value, or 40 per cent of the value cre­ated by all an­a­lyt­ics tech­niques.

At a time when ag­ing and fall­ing birth rates are act­ing as a drag on growth, the de­ploy­ment of AI and au­to­ma­tion tech­nolo­gies can do much to lift the global econ­omy and in­crease global pros­per­ity. Labour pro­duc­tiv­ity growth — a key driver of eco­nomic growth — has slowed in many economies, but AI and au­to­ma­tion have the po­ten­tial to re­verse that de­cline: Pro­duc­tiv­ity growth could po­ten­tially reach two per cent an­nu­ally over the next decade, with 60 per cent of this in­crease from dig­i­tal op­por­tu­ni­ties.

PO­TEN­TIAL TO HELP TACKLE SO­CI­ETAL ‘MOONSHOT’ CHAL­LENGES. AI is also be­ing used in ar­eas rang­ing from ma­te­rial science to med­i­cal re­search and cli­mate science. Ap­pli­ca­tion of the tech­nolo­gies in th­ese and other dis­ci­plines could help tackle so­ci­etal ‘moonshot’ chal­lenges. For ex­am­ple, re­searchers at Geisinger have de­vel­oped an al­go­rithm that could re­duce di­ag­nos­tic times for in­tracra­nial hem­or­rhag­ing by up to 96 per cent. Re­searchers at Ge­orge Wash­ing­ton Univer­sity, mean­while, are us­ing ma­chine learn­ing to more ac­cu­rately weight the cli­mate mod­els used by the In­ter­gov­ern­men­tal Panel on Cli­mate Change. CHAL­LENGES RE­MAIN BE­FORE TH­ESE TECH­NOLO­GIES CAN LIVE UP TO

AI and au­to­ma­tion still face chal­lenges. The THEIR PO­TEN­TIAL. lim­i­ta­tions are partly tech­ni­cal, such as the need for mas­sive train­ing data and dif­fi­cul­ties ‘gen­er­al­iz­ing’ al­go­rithms across use cases. Re­cent in­no­va­tions are just start­ing to ad­dress th­ese is­sues. Other chal­lenges are in the use of AI tech­niques. For ex­am­ple, ex­plain­ing de­ci­sions made by ma­chine learn­ing al­go­rithms is tech­ni­cally chal­leng­ing, which par­tic­u­larly mat­ters for use cases in­volv­ing fi­nan­cial lend­ing or le­gal ap­pli­ca­tions. Po­ten­tial bias in the train­ing data and al­go­rithms, as well as data pri­vacy, ma­li­cious use and se­cu­rity are all is­sues that must be ad­dressed.

Europe is lead­ing with the new Gen­eral Data Pro­tec­tion Reg­u­la­tion, which cod­i­fies more rights for users over data col­lec­tion and us­age. A dif­fer­ent sort of chal­lenge con­cerns the abil­ity of or­ga­ni­za­tions to adopt th­ese tech­nolo­gies, where peo­ple, data avail­abil­ity, technology and process readi­ness of­ten make it dif­fi­cult. Adop­tion is al­ready un­even across sec­tors and coun­tries. The fi­nance, au­to­mo­tive and telecom­mu­ni­ca­tions sec­tors lead AI adop­tion. Among coun­tries, U.S. in­vest­ment in AI ranked first at $15 bil­lion to $23 bil­lion in 2016, fol­lowed by Asia’s in­vest­ments of $8 bil­lion to $12 bil­lion, with Europe lag­ging at $3 bil­lion to $4 bil­lion.

How AI and Au­to­ma­tion Will Af­fect Work

Even as AI and au­to­ma­tion bring ben­e­fits to busi­ness and so­ci­ety, we need to pre­pare for some ma­jor dis­rup­tions to work.


COULD BE AU­TO­MATED. Our anal­y­sis of more than 2,000 work ac­tiv­i­ties across more than 800 oc­cu­pa­tions shows that cer­tain cat­e­gories of ac­tiv­i­ties are more eas­ily au­tomat­able than oth­ers. They in­clude phys­i­cal ac­tiv­i­ties in highly pre­dictable and struc­tured en­vi­ron­ments, as well as data col­lec­tion and data pro­cess­ing. Th­ese ac­count for roughly half of the ac­tiv­i­ties that peo­ple do across all sec­tors. The least sus­cep­ti­ble cat­e­gories in­clude man­ag­ing oth­ers, pro­vid­ing ex­per­tise, and in­ter­fac­ing with stake­hold­ers.

Only five per cent of oc­cu­pa­tions could be fully au­to­mated by cur­rently demon­strated tech­nolo­gies.

Nearly all oc­cu­pa­tions will be af­fected by au­to­ma­tion, but only about five per cent of oc­cu­pa­tions could be fully au­to­mated by cur­rently demon­strated tech­nolo­gies. Many more oc­cu­pa­tions have por­tions of their con­stituent ac­tiv­i­ties that are au­tomat­able: We find that about 30 per cent of the ac­tiv­i­ties in 60 per cent of all oc­cu­pa­tions could be au­to­mated. This means that most work­ers — from welders to mort­gage bro­kers to CEOS — will work along­side rapidly evolv­ing machines — and the na­ture of th­ese oc­cu­pa­tions will likely change as a re­sult.

JOBS WILL BE LOST. We have found that around 15 per cent of the global work­force, or about 400 mil­lion work­ers, could be dis­placed by au­to­ma­tion by 2030. This re­flects our mid-point sce­nario in pro­ject­ing the pace and scope of adop­tion. Un­der the fastest sce­nario we have modelled, that fig­ure rises to 30 per cent, or 800 mil­lion work­ers; while in our slow­est-adop­tion sce­nario, only about 10 mil­lion peo­ple would be dis­placed — close to zero per cent of the global work­force.

This wide range un­der­scores the mul­ti­ple fac­tors that will im­pact the pace and scope of AI and au­to­ma­tion adop­tion. Tech­ni­cal fea­si­bil­ity of au­to­ma­tion is only the first in­flu­enc­ing fac­tor. Oth­ers in­clude the cost of de­ploy­ment; labour-mar­ket dy­nam­ics, in­clud­ing labour sup­ply quan­tity, qual­ity, and the as­so­ci­ated wages; the ben­e­fits be­yond labour sub­sti­tu­tion that con­trib­ute to busi­ness cases for adop­tion; and, fi­nally, so­cial norms and ac­cep­tance.

JOBS WILL BE GAINED. Even as work­ers are dis­placed, there will be growth in de­mand for work and, con­se­quently, jobs. We de­vel­oped sce­nar­ios for labour de­mand to 2030 from sev­eral cat­a­lysts of de­mand for work, in­clud­ing ris­ing in­comes, in­creased spend­ing on health­care, and con­tin­u­ing or stepped-up in­vest­ment in in­fra­struc­ture, en­ergy and technology de­vel­op­ment and de­ploy­ment. Th­ese sce­nar­ios showed a range of ad­di­tional labour de­mand of be­tween 21 and 33 per cent of the global work­force (555 mil­lion and 890 mil­lion jobs) to 2030, more than off­set­ting the num­bers of jobs lost. Some of the largest gains will be in emerg­ing economies such as In­dia, where the work­ing-age pop­u­la­tion is al­ready grow­ing rapidly.

Ad­di­tional eco­nomic growth, in­clud­ing from busi­ness dy­namism and ris­ing pro­duc­tiv­ity growth, will also con­tinue to cre­ate jobs. If his­tory is a guide, many other new oc­cu­pa­tions that we can­not cur­rently imag­ine will also emerge and may ac­count for as much as 10 per cent of jobs cre­ated by 2030. More­over, technology it­self has his­tor­i­cally been a net job cre­ator. For ex­am­ple, the in­tro­duc­tion of the per­sonal com­puter in the 1970s and 1980s cre­ated mil­lions of jobs, not just for semi-con­duc­tor mak­ers, but also for soft­ware and app de­vel­op­ers of all types, cus­tomer ser­vice rep­re­sen­ta­tives and in­for­ma­tion an­a­lysts.

JOBS WILL CHANGE. More jobs than those lost or gained will be changed as machines com­ple­ment hu­man labour in the work­place. Par­tial au­to­ma­tion will be­come more preva­lent as machines com­ple­ment hu­man labour. For ex­am­ple, AI al­go­rithms that can read di­ag­nos­tic scans with a high de­gree of ac­cu­racy will help doc­tors di­ag­nose pa­tient cases and iden­tify suit­able treat­ment. In other fields, jobs with repet­i­tive tasks could shift to­wards a model of man­ag­ing and trou­bleshoot­ing au­to­mated sys­tems. At Ama­zon, em­ploy­ees who once lifted and stacked ob­jects have be­come robot op­er­a­tors, mon­i­tor­ing the au­to­mated arms and re­solv­ing is­sues such as an in­ter­rup­tion in the flow of ob­jects.

Work­force Tran­si­tions and Chal­lenges

While we ex­pect there will be enough work to en­sure full em­ploy­ment in 2030 based on most of our sce­nar­ios, the tran­si­tions that will ac­com­pany au­to­ma­tion and AI adop­tion will be sig­nif­i­cant. The mix of oc­cu­pa­tions will change, as will skill and ed­u­ca­tional re­quire­ments. Work will need to be re­designed to en­sure that hu­mans work along­side machines most ef­fec­tively.

WORK­ERS WILL NEED DIF­FER­ENT SKILLS TO THRIVE IN THE WORK­PLACE OF THE FU­TURE. Au­to­ma­tion will ac­cel­er­ate the shift in re­quired work­force skills we have seen over the past 15 years. De­mand for ad­vanced tech­no­log­i­cal skills such as pro­gram­ming will grow rapidly. So­cial, emo­tional and higher cog­ni­tive skills, such as cre­ativ­ity, crit­i­cal think­ing and com­plex in­for­ma­tion pro­cess­ing will also see grow­ing de­mand. Ba­sic dig­i­tal skills de­mand has

Au­to­ma­tion could ex­ac­er­bate wage po­lar­iza­tion, in­come in­equal­ity and the lack of in­come ad­vance­ment that has char­ac­ter­ized the past decade.

al­ready been in­creas­ing, and that trend will ac­cel­er­ate. De­mand for phys­i­cal and man­ual skills will de­cline, but will re­main the sin­gle largest cat­e­gory of work­force skills in 2030 in many coun­tries. This will put ad­di­tional pres­sure on the al­ready ex­ist­ing work­force skills chal­lenge, as well as the need for new cre­den­tial­ing sys­tems. While some in­no­va­tive so­lu­tions are emerg­ing, so­lu­tions that can match the scale of the chal­lenge will be re­quired.

Many work­ers will need to change oc­cu­pa­tions. Our re­search sug­gests that, in a mid-point sce­nario, around three per cent of the global work­force will need to change oc­cu­pa­tional cat­e­gory by 2030, though our sce­nar­ios range from zero to 14 per cent. Some of th­ese shifts will hap­pen within com­pa­nies and sec­tors, but many will oc­cur across sec­tors and even geog- ra­phies. Oc­cu­pa­tions made up of phys­i­cal ac­tiv­i­ties in highly struc­tured en­vi­ron­ments or in data pro­cess­ing or col­lec­tion will see de­clines. Grow­ing oc­cu­pa­tions will in­clude those with dif­fi­cult-to -au­to­mate ac­tiv­i­ties such as man­agers and those in un­pre­dictable phys­i­cal en­vi­ron­ments such as plum­bers. Other oc­cu­pa­tions that will see in­creas­ing de­mand for work in­clude teach­ers, nurs­ing aides, and tech and other pro­fes­sion­als.


ALONG­SIDE MACHINES. As in­tel­li­gent machines and soft­ware are in­te­grated more deeply into the work­place, work­flows and workspaces will con­tinue to evolve to en­able hu­mans and machines to work to­gether. As self-check­out machines are in­tro­duced in stores, for ex­am­ple, cashiers can be­come check­out as­sis­tance

helpers, who can help an­swer ques­tions or trou­bleshoot the machines. More sys­tem-level so­lu­tions will prompt re­think­ing of the en­tire work­flow and workspace. Ware­house de­sign may change sig­nif­i­cantly as some por­tions are de­signed to ac­com­mo­date pri­mar­ily ro­bots and oth­ers to fa­cil­i­tate safe hu­man-ma­chine in­ter­ac­tion.


The oc­cu­pa­tional mix shifts will likely put pres­sure on wages. Many of the cur­rent mid­dle-wage jobs in ad­vanced economies are dom­i­nated by highly au­tomat­able ac­tiv­i­ties, such as in man­u­fac­tur­ing or in ac­count­ing, which are likely to de­cline. High-wage jobs will grow sig­nif­i­cantly, es­pe­cially for high-skill med­i­cal and tech or other pro­fes­sion­als, but a large por­tion of jobs ex­pected to be cre­ated, in­clud­ing teach­ers and nurs­ing aides, typ­i­cally have lower wage struc­tures. The risk is that au­to­ma­tion could ex­ac­er­bate wage po­lar­iza­tion, in­come in­equal­ity, and the lack of in­come ad­vance­ment that has char­ac­ter­ized the past decade across ad­vanced economies, stok­ing so­cial and po­lit­i­cal ten­sions.

Ten Things to Solve For

In the search for ap­pro­pri­ate mea­sures and poli­cies to ad­dress th­ese chal­lenges, we should not seek to roll back or slow dif­fu­sion of the tech­nolo­gies. Rather, the fo­cus should be on ways to en­sure that the com­ing work­force tran­si­tions are as smooth as pos­si­ble. This is likely to re­quire more ac­tion­able and scal­able so­lu­tions in sev­eral key ar­eas:

• En­sur­ing ro­bust eco­nomic and pro­duc­tiv­ity growth. Strong growth is not the magic an­swer for all the chal­lenges posed by au­to­ma­tion, but it is a pre-req­ui­site for job growth and in­creas­ing pros­per­ity. Pro­duc­tiv­ity growth is a key con­trib­u­tor to eco­nomic growth. There­fore, un­lock­ing in­vest­ment and de­mand, as well as em­brac­ing au­to­ma­tion for its pro­duc­tiv­ity con­tri­bu­tions, is crit­i­cal.

• Fos­ter­ing busi­ness dy­namism. En­trepreneur­ship and more rapid new busi­ness for­ma­tion will not only boost pro­duc­tiv­ity, but also drive job cre­ation. A vi­brant en­vi­ron­ment for small busi­nesses as well as a com­pet­i­tive en­vi­ron­ment for large busi­ness fos­ters busi­ness dy­namism and, with it, job growth. Ac­cel­er­at­ing the rate of new busi­ness for­ma­tion and the growth and com­pet­i­tive­ness of busi­nesses, large and small, will re­quire sim­pler and evolved reg­u­la­tions, tax and other in­cen­tives.

• Evolv­ing ed­u­ca­tion sys­tems and learn­ing for a changed work­place. Pol­i­cy­mak­ers work­ing with ed­u­ca­tion providers (tra­di­tional and non-tra­di­tional) and em­ploy­ers them­selves could do more to im­prove ba­sic STEM skills through the school sys­tems and im­proved on-the-job train­ing. A new em­pha­sis is needed on cre­ativ­ity, crit­i­cal and sys­tems think­ing, and adap­tive and life­long learn­ing. There will need to be so­lu­tions at scale.

• In­vest­ing in hu­man cap­i­tal. Re­vers­ing the trend of low, and in some coun­tries, de­clin­ing pub­lic in­vest­ment in worker train­ing is crit­i­cal. Through tax ben­e­fits and other in­cen­tives, pol­i­cy­mak­ers can en­cour­age com­pa­nies to in­vest in hu­man cap­i­tal, in­clud­ing job cre­ation, learn­ing and ca­pa­bil­ity build­ing, and wage growth, sim­i­lar to in­cen­tives for the pri­vate sec­tor to in­vest in other types of cap­i­tal, in­clud­ing R&D.

• Im­prov­ing labour mar­ket dy­namism. In­for­ma­tion sig­nals that en­able match­ing of work­ers to work and cre­den­tial­ing could work bet­ter in most economies. Dig­i­tal plat­forms can also help match peo­ple with jobs and re­store vi­brancy to the labour mar­ket. When more peo­ple change jobs, even within a com­pany, ev­i­dence sug­gests that wages rise. As more va­ri­eties of work and in­come-earn­ing op­por­tu­ni­ties emerge, in­clud­ing the gig econ­omy, we will need to solve for is­sues such as porta­bil­ity of ben­e­fits, worker classification and wage vari­abil­ity.

• Re­design­ing work. Work­flow de­sign and workspace de­sign will need to adapt to a new era in which peo­ple work more closely with machines. This is both an op­por­tu­nity and a chal­lenge, in terms of cre­at­ing a safe and pro­duc­tive en­vi­ron­ment. Or­ga­ni­za­tions are chang­ing too, as work be­comes more col­lab­o­ra­tive and com­pa­nies seek to be­come in­creas­ingly ag­ile and non-hi­er­ar­chi­cal.

• Re­think­ing in­comes. If au­to­ma­tion (full or par­tial) does re­sult in a sig­nif­i­cant re­duc­tion in em­ploy­ment and/or

greater pres­sure on wages, some ideas such as conditional trans­fers, sup­port for mo­bil­ity, universal ba­sic in­come and adapted so­cial safety nets could be con­sid­ered and tested. The key will be to find so­lu­tions that are eco­nom­i­cally vi­able and in­cor­po­rate the mul­ti­ple roles that work plays for work­ers, in­clud­ing pro­vid­ing not only in­come, but also mean­ing, pur­pose, and dig­nity.

• Re­think­ing tran­si­tion sup­port and safety nets for work­ers af­fected. As work evolves at higher rates of change be­tween sec­tors, lo­ca­tions, ac­tiv­i­ties and skill re­quire­ments, many work­ers will need as­sis­tance ad­just­ing. Many best prac­tice ap­proaches to tran­si­tion safety nets are avail­able, and should be adopted and adapted, while new ap­proaches should be con­sid­ered and tested.

• In­vest­ing in driv­ers of de­mand for work. Gov­ern­ments will need to con­sider step­ping up in­vest­ments that are ben­e­fi­cial in their own right and will also con­trib­ute to de­mand for work (e.g. in­fra­struc­ture, cli­mate change adap­ta­tion). Th­ese types of jobs, from con­struc­tion to rewiring build­ings and in­stalling so­lar pan­els, are of­ten mid­dle-wage jobs — those most af­fected by au­to­ma­tion.

• Em­brac­ing AI and au­to­ma­tion safely. Even as we cap­ture the pro­duc­tiv­ity ben­e­fits of th­ese rapidly evolv­ing tech­nolo­gies, we need to ac­tively guard against the risks and mit­i­gate any dan­gers. The use of data must al­ways take into ac­count con­cerns, in­clud­ing data se­cu­rity, pri­vacy, ma­li­cious use and po­ten­tial is­sues of bias — is­sues that pol­i­cy­mak­ers, tech and other firms and in­di­vid­u­als will need to find ef­fec­tive ways to ad­dress.

In clos­ing

There is work for ev­ery­one to­day and there will be work for ev­ery­one to­mor­row, even in a fu­ture with au­to­ma­tion. But that work will be dif­fer­ent, re­quir­ing new skills and a far greater adapt­abil­ity of the work­force than we have seen. Train­ing and re­train­ing both mid-ca­reer work­ers and new gen­er­a­tions for the com­ing chal­lenges will be an im­per­a­tive. Gov­ern­ment, pri­vate sec­tor lead­ers and in­no­va­tors all need to work to­gether to bet­ter co­or­di­nate pub­lic and pri­vate ini­tia­tives, in­clud­ing cre­at­ing the right in­cen­tives to in­vest more in hu­man cap­i­tal.

The fu­ture with au­to­ma­tion and AI will be chal­leng­ing, but it will also be a much richer one if we har­ness the tech­nolo­gies with aplomb — and mit­i­gate the neg­a­tive ef­fects.

The mix of oc­cu­pa­tions will change, as will skill and ed­u­ca­tional re­quire­ments.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.