AN­GELA ZU­TAV­ERN the Math­e­mat­i­cal Cor­po­ra­tion

An ex­pert in ma­chine in­tel­li­gence de­scribes the po­ten­tial of ‘the math­e­mat­i­cal cor­po­ra­tion’.

Rotman Management Magazine - - FROM THE EDITOR - In­ter­view by Karen Chris­tensen

You be­lieve that the world of lead­er­ship has hit an in­flec­tion point. How so?

As use­ful as pop­u­lar men­tal mod­els and heuris­tics are, ma­chine mod­els now out­strip hu­man per­for­mance in about half of the port­fo­lio of cog­ni­tive tasks. Go­ing for­ward, we will be able to beat ma­chine mod­els less and less of­ten — ex­cept in the realms of imag­i­na­tion, cre­ativ­ity, prob­lem-solv­ing and some kinds of rea­son­ing.

En­ter the ‘math­e­mat­i­cal cor­po­ra­tion’. How do you de­fine this new type of or­ga­ni­za­tion?

In the past cou­ple of years, there have been ma­jor ad­vances in ma­chine learn­ing, bring­ing mas­sive new po­ten­tial across in­dus­tries. The math­e­mat­i­cal cor­po­ra­tion is a term that de­scribes how or­ga­ni­za­tions will need to op­er­ate in the fu­ture, to em­brace these ad­vances. Sim­ply put, ma­chine in­tel­li­gence — built on data sci­ence — en­ables us to see pat­terns, anom­alies and as­so­ci­a­tions that were pre­vi­ously uniden­ti­fi­able, and this emerg­ing abil­ity re­quires a new form of lead­er­ship. In a math­e­mat­i­cal cor­po­ra­tion, peo­ple ac­tively col­lab­o­rate with ma­chines. AI ba­si­cally has a ‘seat’ at the board­room ta­ble.

Is this about em­brac­ing Big Data?

Of course, lead­ers of math­e­mat­i­cal cor­po­ra­tions use an­a­lyt­ics and Big Data as well as AI and other ad­vanced tech­nol­ogy. But the ca­pa­bil­i­ties of the math­e­mat­i­cal cor­po­ra­tion ex­tend be­yond min­ing data sets — an en­deav­our that has fo­cused nar­rowly on an­swer­ing known ques­tions by query­ing spe­cific piles of data. The math­e­mat­i­cal cor­po­ra­tion also fo­cuses on un­cov­er­ing new ques­tions by query­ing a uni­verse of data. Search­ing for and an­swer­ing ques­tions out­side the spot­light of con­ven­tional thought can pro­vide knowl­edge about the fu­ture.

If the re­cent past was about an­a­lyt­ics and Big Data, the fu­ture is about the ‘big mind’ of the math­e­mat­i­cal cor­po­ra­tion, which comes from com­bin­ing the math­e­mat­i­cal smarts of ma­chines with peo­ples’ imag­i­na­tive in­tel­lect. This is what will trig­ger the next leaps in or­ga­ni­za­tional per­for­mance.

Does in­tu­ition have any role in the math­e­mat­i­cal cor­po­ra­tion?

In­tu­ition has served lead­ers well — and still does — prin­ci­pally be­cause the hu­man mind ab­sorbs and un­der­stands more de­tail and sub­stance than we con­sciously know. But with each pass­ing day, ma­chines are catch­ing up. So, when should you trust a de­ci­sion to your gut, and when to data? Most of us rec­og­nize that biases, pol­i­tics and wish­ful think­ing dis­tort our per­cep­tion; but as ma­chines start to com­pre­hen­sively re­flect the real world, their biases dis­si­pate. So, we need both.

For­mer Rot­man School Dean Roger Martin has said that data can only be used to un­der­stand the past—not to pre­dict the fu­ture, be­cause the fu­ture hasn’t hap­pened yet. How do you re­act to that state­ment?

I think Roger is prob­a­bly talk­ing about the crit­i­cal skill sets that hu­mans bring to the ta­ble in our data-filled world, which in­clude things like cre­ativ­ity and imag­i­na­tion, strat­egy-set­ting and vi­sion. A ma­chine is not go­ing to do those things for us, but ma­chine in­tel­li­gence can cer­tainly be very help­ful in other ways.

For in­stance, when In­ter­con­ti­nen­tal Ho­tels pitches a new of­fer to its loy­alty club mem­bers, its mar­keters still de­vise the con­tent, but the com­puter chooses which in­di­vid­u­als get which of­fer. We will see more and more such part­ner­ships be­tween hu­man and ma­chine.

In its man­u­fac­tur­ing pro­cesses to make vac­cines, Merck uses a four-stage process. In the past, they looked at data for each of the four pro­cesses sep­a­rately and op­ti­mized within that process. But they con­tin­ued to have prob­lems: Batches were go­ing bad and they didn’t un­der­stand why. So, they brought to­gether the data across all four pro­cesses to op­ti­mize across the sys­tem. That’s where they un­cov­ered some re­ally good in­sights. They had sus­pected that raw ma­te­ri­als were caus­ing the batches to go bad, but it turned out that wasn’t the case at all. It had to do with the el­e­ments of the fer­men­ta­tion process. The ma­chine in­tel­li­gence model pre­dicted when a prob­lem was about to hap­pen, so that they could stop it from hap­pen­ing.

In man­u­fac­tur­ing, com­pa­nies are now able to pre­dict when their ma­chines will break down and do preventative main­te­nance be­fore that hap­pens; and in sports, we’re able to use sen­sor data to pre­dict when an ath­lete is about to get hurt, what kind a pitch a base­ball player is about to throw, and what’s the best way to train and con­di­tion an ath­lete for a cer­tain type of per­for­mance. So, there are al­ready lots of ex­am­ples where data alone has been able to pre­dict the fu­ture.

Are hu­mans even re­quired in a math­e­mat­i­cal cor­po­ra­tion?

Ab­so­lutely! I don’t be­lieve the doom-and-gloom sce­nar­ios where AI takes over the world. Hu­mans are ab­so­lutely crit­i­cal for func­tions that ma­chines just can’t do. Ma­chine in­tel­li­gence is great at things like num­ber crunch­ing, rec­og­niz­ing pat­terns, or­ga­niz­ing in­for­ma­tion and re­mem­ber­ing things; but, as in­di­cated, peo­ple still rule when it comes to prob­lem-solv­ing, rea­son­ing, cre­ativ­ity and imag­i­na­tion. Hu­man lead­ers are ab­so­lutely crit­i­cal for set­ting strate­gies, for ask­ing the big ques­tions, and putting to­gether the right com­bi­na­tions of back­grounds to cre­ate a team that will come up with new break­throughs and so­lu­tions. It’s not about choos­ing one or the other. It’s got to be both: The best or­ga­ni­za­tions will have hu­man and ma­chine in­tel­li­gence work­ing to­gether.

Ma­chine in­tel­li­gence en­ables us to see pat­terns, anom­alies and as­so­ci­a­tions that were pre­vi­ously uniden­ti­fi­fi­able.

What is your favourite ex­am­ple of a math­e­mat­i­cal cor­po­ra­tion?

One of my favourites is the Data Sci­ence Bowl, which set out to iden­tify new ways of pre­dict­ing and iden­ti­fy­ing lung can­cer. We or­ga­nized this with Kag­gle — a plat­form for pre­dic­tive mod­el­ling and an­a­lyt­ics com­pe­ti­tions — putting it out to a com­mu­nity of hun­dreds of thou­sands of data sci­en­tists around the world. They com­peted (as vol­un­teers) to de­velop al­go­rithms to find new ways of iden­ti­fy­ing lung can­cer ear­lier than it can be iden­ti­fied to­day.

Some real break­throughs oc­cured. Sev­eral of the win­ners have gone on to re­ceive fund­ing to con­tinue to re­search these ideas and ‘pro­duc­tize’ the al­go­rithm, so it can be used more widely. I pre­dict that we are go­ing to see some in­cred­i­ble break­throughs in the health arena. What is so cool is that the teams that are win­ning these com­pe­ti­tions don’t nec­es­sar­ily even have a med­i­cal back­ground. They’re able to learn enough from on­line tu­to­ri­als to make break­throughs that re­searchers in the med­i­cal com­mu­nity have been try­ing to make for years. It goes to show that when you bring to­gether peo­ple with com­pletely dif­fer­ent back­grounds, you of­ten get break­through ideas.

De­scribe how Ford Mo­tor Com­pany is tak­ing steps to be­come a math­e­mat­i­cal cor­po­ra­tion.

It might be one of the old­est cor­po­ra­tions in North Amer­ica, but Ford is ded­i­cated to em­brac­ing ma­chine in­tel­li­gence. Their new CEO, Mark Hack­ett, ac­tu­ally came from the au­to­mated ve­hi­cles area of the busi­ness — one of the key re­search ar­eas for ma­chine in­tel­li­gence.

In a math­e­mat­i­cal cor­po­ra­tion, you have to con­stantly ex­per­i­ment, fail, learn and it­er­ate. Ford has about 25 ex­per­i­ments go­ing on at any one time. They want to learn ev­ery­thing there is to learn about trans­porta­tion. In the past, like many other com­pa­nies, Ford re­lied on cus­tomer sur­veys for feed­back. But, of course, a sur­vey is just a sam­pling: You never get a 100 per cent re­sponse. So now, they are us­ing real data about ac­tual con­sumer be­hav­iour. The con­sumer doesn’t have to pro­vide the data: The sen­sors in the com­pany’s cars track and pro­vide it. They’re learn­ing a lot about how peo­ple move around in the world to­day, and it’s not based on what peo­ple say they’re do­ing, but what they are ac­tu­ally do­ing. The fact is, in most cases, sen­sor data is much more ac­cu­rate than what peo­ple say.

What has to change about our ap­proach to dis­cov­ery in to­day’s world?

When we’re think­ing about new prod­ucts and ser­vices, we tend to rea­son de­duc­tively: We have a pre­sup­posed idea of what we plan to find; we might even have drawn a model on a nap­kin about the way some­thing works. Through ex­per­i­men­ta­tion, we then check to see whether we’re right and es­tab­lish our ‘right­ness’ with a large de­gree of cer­tainty.

To­day, we need to rea­son in­duc­tively. We do this by ex­am­in­ing the way the world cur­rently works and in­fer­ring re­la­tion­ships be­tween key el­e­ments in the sys­tem. We rec­og­nize that we can never say for sure that we’ve dis­cov­ered the truth, but we can reach con­clu­sions based on rea­son­able logic. While the con­clu­sion of a de­duc­tive ar­gu­ment is cer­tain, the con­clu­sion of an in­duc­tive ar­gu­ment is prob­a­ble, based upon the ev­i­dence.

We nat­u­rally en­gage in both types of rea­son­ing, but when we’re cre­at­ing and launch­ing a new prod­uct, we of­ten limit our­selves to think­ing de­duc­tively, be­cause we see that as the only way to gain a level of cer­tainty. With ma­chine in­tel­li­gence, we in­creas­ingly have the op­por­tu­nity to rea­son in­duc­tively and dis­cover new knowl­edge that we never could have found on our own.

For ex­am­ple, in mar­ket­ing, dis­cov­ery in data sets can un­veil pat­terns in cus­tomer de­fec­tions and pre­dict means to re­verse them; in lo­gis­tics, dis­cov­ery can ex­am­ine daily or sea­sonal flow of ma­te­ri­als or prod­ucts and guide trans­porta­tion plan­ning; in HR, dis­cov­ery can iden­tify un­happy em­ploy­ees and guide changes in or­ga­ni­za­tional prac­tices; in man­u­fac­tur­ing, it can process war­ranty data for pat­terns of prod­uct fail­ure to sug­gest en­gi­neer­ing im­prove­ments. This type of dis­cov­ery prom­ises to turn busi­nesses — and lives — around.

You also be­lieve we need to learn how to ‘re-frame ques­tions’ for this new era. Why is that so im­por­tant?

If we don’t break away from how we’ve done things in the

Search­ing for and an­swer­ing ques­tions out­side the spot­light of con­ven­tional thought can pro­vide knowl­edge about the fu­ture.

past, we will never be able to make the dis­cov­er­ies that ma­chine in­tel­li­gence makes pos­si­ble. One ex­am­ple that comes to mind is the work that the U.S. Cen­sus Bureau is do­ing. Over the years, it had al­ways con­ducted cen­sus sur­veys in the same way, by re­ly­ing on enu­mer­a­tors — peo­ple go­ing door-to-door, col­lect­ing data. But re­cently, the bureau’s lead­ers com­pletely rethought ev­ery­thing about how to con­duct the up­com­ing 2020 ‘De­cen­nial Cen­sus’. They re­al­ized that they can pull data to avoid in-per­son vis­its. For in­stance, some­times they need to ver­ify that an apart­ment build­ing is still there; to­day, you can do that us­ing satel­lite im­agery.

In ad­di­tion, in the past, enu­mer­a­tors were just left on their own to fig­ure out which routes to take, what time of day to visit, what or­der to go in, etc. Not sur­pris­ingly, they of­ten found that peo­ple weren’t home and had to make re­peat trips. This time around, there will be a mo­bile app that enu­mer­a­tors can use to pre­dict the best time to make a visit, in what or­der they should be made and what route to take. That is a key ex­am­ple of why it’s so im­por­tant to be open to new ap­proaches. And im­por­tantly, lead­ers need to rec­og­nize that in many cases, the new ap­proach they de­velop will be com­pletely dif­fer­ent from the way they have has op­er­ated in the past.

Newspapers in English

Newspapers from Canada

© PressReader. All rights reserved.