Yes, AI is when a com­puter now thinks for it­self

The Star Early Edition - - OPINION & ANALYSIS - Andile Ma­suku

SO, THERE’S this South African mate of mine who hap­pens to be a gifted poly­math. He is a me­dia-shy physics PhD who spent much of his early pro­fes­sional ca­reer work­ing in fi­nance – pro­gram­ming ar­ti­fi­cial in­tel­li­gence (AI) driven so­lu­tions to take over com­plex port­fo­lio man­age­ment tasks that were tra­di­tion­ally han­dled by hu­man be­ings. In­deed, to­day a lot of the ac­tiv­i­ties that hap­pen on the world’s lead­ing se­cu­ri­ties ex­changes are now run by ma­chines.

How­ever, my learned buddy is of­ten quick to re­mind me how mis­lead­ing the no­tion of things be­ing “run by ma­chines” can be when we for­get that ul­ti­mately, the ma­chines them­selves are “run by hu­mans”.

This is an im­por­tant ob­ser­va­tion, given the me­dia’s ten­dency to use hu­man de­scrip­tors to chron­i­cle ad­vances in AI.

At­ten­tion grab­bing me­dia head­lines like a re­cent Busi­ness In­sider one which reads: Google’s Deep­Mind AI just taught it­self to walk, cap­i­talise on so­ci­ety’s Hol­ly­wood-in­duced fas­ci­na­tion with the idea of ma­chines tak­ing over the world.

Af­ter all, if bots can teach them­selves to walk, they may well teach them­selves to be us at some point, right? There’s no doubt that such so­cial me­dia-op­ti­mised, SEO-friendly state­ments are de­signed to cap­ture the imag­i­na­tion of the vast ma­jor­ity of us who, all too of­ten, do not read past the head­lines to learn more about what is of­ten a fairly un­spec­tac­u­lar AI evo­lu­tion.

I’ve come to ap­pre­ci­ate that huge leaps in logic are only pos­si­ble when we sur­ren­der our­selves to fu­tur­is­tic sen­sa­tion­al­i­sa­tion and con­ve­niently for­get that hu­mans are ul­ti­mately re­spon­si­ble not only for seed­ing the “in­tel­li­gence” in AI, but also for de­ter­min­ing the pa­ram­e­ters, or in­deed the lack thereof, within which au­to­mated soft­ware ought to tire­lessly toil to­wards op­ti­mi­sa­tion.

Given all that, this re­cent TechCrunch head­line which reads Any­one can teach this MIT ro­bot how to teach other ro­bots more ac­cu­rately ac­counts for the in­flu­ence of hu­mans in con­tribut­ing to­wards the pro­lif­er­a­tion of AI – par­tic­u­larly within the sub-field of ma­chine learn­ing.

Ryan Falken­berg is co-founder and co-chief ex­ec­u­tive of a South African start-up called Clevva – an AI plat­form that “al­lows non-coders to build and main­tain nav­i­ga­tion apps” which or­gan­i­sa­tions can then de­ploy to help em­ploy­ees per­form real-time anal­y­sis that re­sults in sound de­ci­sion-mak­ing. Falken­berg is a man who un­der­stands the value of the hu­man fac­tor in cre­at­ing and de­ploy­ing AI so­lu­tions, and he is well-placed to help us wrap our minds around the prac­ti­cal im­pli­ca­tions of ad­vances in AI.

Sim­plest terms

In a re­cent e-mail-in­ter­view, I asked Falken­berg to break down the no­tion of AI into the sim­plest pos­si­ble terms – an ex­pla­na­tion a 3-year-old might understand. He told me that AI is when a com­puter thinks for it­self, and does not sim­ply fol­low a pro­gramme of pre-coded in­struc­tions. The com­puter does this by analysing all the data it can ac­cess to work out the high­est prob­a­bil­ity of cer­tain out­comes based on spe­cific re­quests.

It then tries to learn from the out­comes and im­prove the prob­a­bil­ity of pro­vid­ing cor­rect an­swers or per­form­ing de­sired ac­tions the next time. Un­like hu­mans, how­ever, AI can learn via many com­put­ers and draw on huge batches of data to make de­ci­sions. This al­lows it to get better at tasks far quicker, and con­sider more vari­ables than hu­mans are typ­i­cally able to do.

When asked to il­lus­trate the dif­fer­ence be­tween AI and ma­chine learn­ing, Falken­berg stated that AI is a broad um­brella that en­com­passes more than just ma­chine learn­ing, and is of­ten loosely split into two ar­eas, namely:

(1) Ar­ti­fi­cial Nar­row In­tel­li­gence – which is AI fo­cussed on a very spe­cific ar­eas or fields.

(2) Ar­ti­fi­cial Gen­eral In­tel­li­gence – like IBM’s mega ques­tion-an­swer­ing ma­chine, Wat­son. Ap­par­ently, Nasa’s Mars Ex­plo­ration Rover is an ex­cel­lent ex­am­ple of ma­chine learn­ing in ac­tion. The Rover was de­signed to op­er­ate with­out hu­man in­ter­ven­tion. It was de­ployed to gather data from the sur­face of Mars and use that data to make de­ci­sions.

The con­cept of “learn­ing” in the phrase ma­chine learn­ing stems from the fact that the more data is made avail­able to the ma­chine, the better the de­ci­sions it will make over time. The past re­sults of de­ci­sions made then in­form fu­ture de­ci­sions, al­low­ing the ma­chine to adapt – much like a baby would by ex­plor­ing the world around him/her.

The more the ma­chine “tries” things, the more it “learns”, and there­fore the better it gets at “de­ci­sion-mak­ing”. This form of ma­chine learn­ing, also dubbed cog­ni­tive com­put­ing, can be seen in things like driver­less cars and pretty much wher­ever hu­man de­ci­sion-mak­ing is sup­planted to im­prove ef­fi­cien­cies.

In to­day’s in­creas­ingly digi­tised world, AI is all around us. It has become ubiq­ui­tous no small thanks to its wide­spread de­ploy­ment by firms like Ap­ple and Google, who use it to power vir­tual as­sis­tants like Siri and Google Now, for ex­am­ple. It is also used by Face­book to help iden­tify and tag people, places, and things, in gam­ing to make char­ac­ters more real and give them “per­son­al­i­ties”.

In busi­ness, AI is used to pre­dict con­sumer ac­tions, de­tect fraud and pre-empt crim­i­nal ac­tiv­i­ties us­ing pre­dic­tive mod­el­ling. For ex­am­ple, that’s how on­line re­tail­ers creep­ily “know” what you will buy be­fore you do. They are able to send you tai­lored pro­mo­tions and coupons based on cal­cu­lated pre­dic­tions they’ve made about you.

De­ci­sion nav­i­ga­tor

When chal­lenged to sug­gest what one of the more per­ti­nent trends within AI might be within the African con­text, Falken­berg cited a lesser-known kind of AI that cap­tures known in­tel­li­gence and ex­per­tise, and then by us­ing three or four-di­men­sional logic, as­sists hu­mans in their de­ci­sion-mak­ing.

This form of AI is called a de­ci­sion nav­i­ga­tor. De­ci­sion nav­i­ga­tors work like GPS’s to aug­ment hu­mans rather than re­place them.

In his re­sponse, Falken­berg un­doubt­edly took the op­por­tu­nity to give a not-so­sub­tle nod to Clevva’s area of spe­cial­ity. None­the­less, I do rate the im­por­tance of that spe­cific field of AI be­cause of the real threat to African liveli­hoods be­ing posed by the de­ploy­ment of au­to­mated soft­ware.

It is a re­al­ity I have high­lighted in this col­umn sev­eral times be­fore. Quite no­tably, de­ci­sion nav­i­ga­tors help com­pa­nies use their ex­ist­ing work­forces more ef­fi­ciently to do higher-value work, rather than re­plac­ing them al­to­gether.

De­ci­sion nav­i­ga­tors are be­ing de­ployed by banks, in­sur­ers, petro-chems, and even tel­cos to guide staff in their de­ci­sion-mak­ing, within their prod­uct, pol­icy and pro­ce­dural en­vi­ron­ments.

Be­cause de­ci­sion nav­i­ga­tors leave an au­dit trail, they fre­quently form part of regulation tech­nol­ogy regimes de­signed to pro­mote cor­po­rate gov­er­nance.

De­ci­sion nav­i­ga­tors also help com­pa­nies on board and train new staff in sub­stan­tially less time than was pre­vi­ously re­quired.

In­stead of hav­ing to teach people everything they need to know by rote learn­ing, or­gan­i­sa­tions can con­cen­trate on get­ting em­ploy­ees up to speed with the bare es­sen­tials and aug­ment that with de­ci­sion nav­i­ga­tors that will help people nav­i­gate their daily tasks – how­ever ba­sic or com­plex.

Fi­nally, I asked Falken­berg what in­no­va­tions within AI he is most en­thu­si­as­tic about right now. He told me that he’s rather ex­cited by the in­creas­ing abil­ity of AI plat­forms to ac­cu­rately sense and in­ter­pret the en­vi­ron­ment, al­low­ing for more ac­cu­rate de­ci­sion-mak­ing.

He reck­ons that as we get better at nat­u­ral lan­guage trans­la­tions and the ac­cu­rate in­ter­pre­ta­tion of in­tent, the ad­vances in sen­sors that can pro­vide in­puts across all five senses will ac­cel­er­ate AI to­wards full au­toma­tion in many ar­eas that im­pact daily liv­ing, in­clud­ing proac­tive mon­i­tor­ing of pretty much everything.

De­ci­sion nav­i­ga­tors are be­ing de­ployed by banks, in­sur­ers, petro-chems, and even tel­cos to guide staff in their de­ci­sion-mak­ing.

Andile Ma­suku is a broad­caster and en­tre­pre­neur based in Jo­han­nes­burg. He is the ex­ec­u­tive pro­ducer at AfricanTechRoundup.com. Fol­low him on Twit­ter @Ma­sukuAndile and The African Tech Round-up @african­roundup

Newspapers in English

Newspapers from South Africa

© PressReader. All rights reserved.