Top six trends in Ar­ti­fi­cial In­tel­li­gence for 2019

Voice search and ‘Mi­nor­ity Re­port’-style ads among the tech to be­come the norm this year

The Irish Times - Business - - BUSINESS INNOVATION - Marie Bo­ran

It’s not all robot de­liv­ery dogs, self-driv­ing cars that let you snooze, or dig­i­tal as­sis­tants that can tell you need to talk about your day. Although these real ex­am­ples of ar­ti­fi­cial in­tel­li­gence de­vel­op­ments are si­mul­ta­ne­ously ter­ri­fy­ing and ex­cit­ing, they are more in­dica­tive of what is to come in three or four years – Volvo’s 360c “bed on wheels” is just a con­cept for now while Huawei is still work­ing on their emo­tional AI soft­ware. Fresh from a CES demo, Con­ti­nen­tal’s robo-dogs are just about hot to trot, so to speak.

But the best way to pre­dict what peo­ple will be talk­ing about in AI in 2019 is to take a look at what hap­pened in 2018 and pay close at­ten­tion to those end-of-year re­ports is­sued by in­dus­try, academia and spe­cial in­ter­est groups for their ex­pert in­sights into the year ahead.

While some ar­eas con­tinue to steadily grow – ma­chine learn­ing and deep learn­ing – oth­ers will creep up on us, sud­denly feel­ing as ubiq­ui­tous as the smart­phone while steadily be­com­ing ca­pa­ble of in­creas­ingly com­plex be­hav­iours: the AI voice as­sis­tant. Mean­while, there are a few sur­prises like the ar­rival of Mi­nor­ity Re­port-style ad­ver­tis­ing and the rise of so­phis­ti­cated com­puter vi­sion.

The AI In­dex 2018 An­nual Re­port pro­vides data-driven in­sights into progress in ar­ti­fi­cial in­tel­li­gence re­search and ap­pli­ca­tion over the past year, giv­ing us a valu­able source from which we can pre­dict trends for the year ahead. Pub­lished by the Hu­man-Cen­tered AI Ini­tia­tive at Stan­ford Univer­sity in De­cem­ber, the re­port gath­ers in­for­ma­tion from ac­tiv­ity across re­search and in­dus­try.

Mean­while, Mi­crosoft’s Ar­ti­fi­cial In­tel­li­gence in Europe: Out­look for 2019 and Be­yond, gives us a glimpse of how AI will af­fect the work­place.

1 The rise and rise of voice search and smart speak­ers

The key­board and touch­screen con­tinue to take a back seat as voice search through dig­i­tal as­sis­tants be­comes com­mon­place: ac­cord­ing to UK cus­tomer en­gage­ment firm Edit, 40 per cent of adults use voice search on a daily ba­sis while Coms­core pre­dicts that by next year, this will rise to half of all in­ter­net searches.

When we are not in­vok­ing Siri, Cor­tana or Google As­sis­tant on our mo­bile de­vices we’re do­ing it else­where: one in six Amer­i­cans now own a smart speaker like Ama­zon’s Alexa-pow­ered Echo. Ac­cord­ing to the re­port from NPR and Edi­son Re­search, this fig­ure has risen by 128 per cent since Jan­uary 2017. At this rate, we can pre­dict an ex­po­nen­tial rise that could lead to one in three by the end of 2019, es­pe­cially con­sid­er­ing Ap­ple’s rel­a­tively re­cent ad­di­tion of its HomePod to the mar­ket.

As with our smart­phones, these de­vices are be­com­ing an in­te­gral part of our lives: a Google/Peer­less In­sights re­port found the peo­ple tend to keep their smart speaker in a com­mon room, ie liv­in­groom, while 72 per cent of users say it has be­come part of their daily rou­tine.

Alexa, what is the hottest AI skill in 2019?

2 Deep learn­ing willbe hottest AI skill in 2019

Here’s what I found: jobs in AI call­ing for knowl­edge of deep learn­ing are grow­ing at a faster rate than any other area ac­cord­ing to data gath­ered from job re­cruit­ment site Mon­ster.com. Deep learn­ing is a form of ma­chine learn­ing that de­vel­ops al­go­rithms known as ar­ti­fi­cial neu­ral net­works which model the struc­ture and func­tion of the hu­man brain. While ma­chine learn­ing is the most com­mon skill listed as a re­quire­ment for these roles, the num­ber of jobs re­quir­ing deep learn­ing has in­creased 35-fold be­tween 2015 and 2017. There is also a fast grow­ing in­ter­est in grad­u­ates versed in com­puter vi­sion.

Deep learn­ing is be­ing ap­plied in many ar­eas of AI in­clud­ing the afore­men­tioned com­puter vi­sion, au­tonomous ve­hi­cles, and au­to­matic text gen­er­a­tion (although “new” works of Shake­speare or po­etry gen­er­ated by these neu­ral net­works are still at a very early stage).

There is also grow­ing de­mand for those versed in gen­eral ma­chine-learn­ing tech­niques, nat­u­ral lan­guage pro­cess­ing (NLP), speech recog­ni­tion, and robotics.

3Mi­nor­ity Re­port-style ad­ver­tis­ing is fi­nally ar­riv­ing

Like it or loathe it, ad­vanced com­puter vi­sion means that AI-driven per­son­alised ad­ver­tis­ing is fi­nally be­com­ing part of the re­tail ex­pe­ri­ence. Ear­lier this week, Mi­crosoft an­nounced a part­ner­ship with US su­per­mar­ket chain Kroger.

Shelves in two pi­lot su­per­mar­kets will ditch pa­per price tags in favour of real-time dig­i­tal ver­sions that also dis­play pro­mo­tions and nutri­tional in­for­ma­tion while work­ing with Kroger’s Scan, Bag, Go app to guide you through the store based on your shop­ping list.

The Mi­nor­ity Re­port part in­volves cam­eras with fa­cial recog­ni­tion: when pass­ing in-store dig­i­tal dis­plays you will be shown ads and pro­mo­tions based on your age and gen­der or, should you opt in, you will be greeted per­son­ally and have prod­ucts suggested based on pre­vi­ous pur­chases. “Good morn­ing, Marie. Would like to pur­chase more Ben & Jerry’s and white wine?” It must have for­got­ten about that time I bought a grape­fruit.

4 2019 will be a turn­ing point for AI in busi­ness

The most im­por­tant role AI will play in the busi­ness world is that of in­creas­ing cus­tomer en­gage­ment, ac­cord­ing to a re­cent re­port is­sued by Mi­crosoft. The re­port shows 74 per cent of all re­spon­dents ex­pect in the near fu­ture that AI will help them win and re­tain more cus­tomers, ris­ing to 100 per cent for more ad­vanced and ma­ture com­pa­nies. Ad­di­tion­ally, 56 per cent of com­pa­nies ex­pect AI to have a large im­pact on “busi­ness ar­eas that are en­tirely un­known to­day” while 65 per cent ex­pect AI to have a sim­i­lar im­pact on their core busi­ness.

With this con­fi­dence in and en­thu­si­asm for AI, 2019 may be a turn­ing point for AI in­te­gra­tion: a mere 4 per cent of com­pa­nies are ac­tively us­ing AI in “many pro­cesses and to en­able ad­vanced tasks” but 61 per cent are al­ready in the process of plan­ning or pi­lot­ing how AI will trans­form their busi­ness model.

Where is AI most and least pop­u­lar from a cor­po­rate stand­point? Mi­crosoft’s re­port shows that, un­sur­pris­ingly, it is most widely adopted within IT or gen­eral tech­nol­ogy-re­lated pro­cesses: 47 per cent of or­gan­i­sa­tions who have adopted AI re­port this is its pri­mary use. The se­cond most pop­u­lar use is re­search and de­vel­op­ment (36 per cent) fol­lowed by cus­tomer ser­vice at 24 per cent, so these are the ar­eas we can ex­pect the high­est level of adop­tion in 2019.

5 Com­puter vi­sion will shape fu­ture of sur­veil­lance

Com­puter vi­sion, as an area of AI re­search, has come on in leaps and bounds in the past few years: al­go­rithms tak­ing the COCO (Com­mon Ob­jects in Con­text) chal­lenge have im­proved their pre­ci­sion by 72 per cent since 2015. The COCO chal­lenge, ac­cord­ing to the AI In­dex 2018 An­nual Re­port, sets com­plex vi­sion tasks that re­quire more so­phis­ti­cated rea­son­ing, like pin­point­ing ob­jects with pixel-level ac­cu­racy.

An­other mea­sure of ad­vance­ment is how quickly com­puter vi­sion al­go­rithms can be trained on a stan­dard­ised dataset of im­ages – used by most re­search in­sti­tutes – known as ImageNet. The av­er­age train­ing time be­came 16 times faster be­tween June 2017 and No­vem­ber 2018.

There are, of course, mul­ti­ple ap­pli­ca­tions for com­puter vi­sion but one area that is con­cern­ing con­sumers and cit­i­zens, is au­to­mated video sur­veil­lance. A Ja­panese ma­chine-learn­ing al­go­rithm called AI Guard­man can de­tect “sus­pi­cious be­hav­iour” among shop­pers that may pre­dict their like­li­hood to shoplift. This is linked to the shop owner’s phone, which re­ceives an alert.

Sim­i­larly, this kind of com­puter vi­sion is be­ing put in drones that hover over large crowds, de­tect­ing po­ten­tially vi­o­lent in­di­vid­u­als. Per­haps you like the idea of pre­vent­ing a riot, per­haps it will leave you quak­ing at the no­tion of throw­ing your fris­bee lest you are ac­ci­den­tally added to a watch­list. Wel­come to 2019.

6 Ethics will in­creas­ingly be ‘baked into’ AI

This past year has seen a rise in pub­lic scep­ti­cism around AI and rightly so. If all of these tech­nolo­gies are ap­plied with­out rig­or­ous eth­i­cal frame­works or with no in­put from the gen­eral pub­lic, we will con­tinue to see ap­pli­ca­tions that skirt pri­vacy, ig­nore en­vi­ron­men­tal and so­ci­etal im­pli­ca­tions and even limit hu­man au­ton­omy.

In De­cem­ber 2018, a draft of AI ethics guid­lines was pub­lished by the Eu­ro­pean Com­mis­sion’s High-Level Ex­pert Group on Ar­ti­fi­cial In­tel­li­gence, of which the In­sight Cen­tre for Data An­a­lyt­ics’ Barry O’Sul­li­van is vice chair.

It has come up with sev­eral rec­om­men­da­tions on for the de­vel­op­ment of trust­wor­thy AI.

Trust­wor­thy AI, it says, has two com­po­nents: “1) it should re­spect fun­da­men­tal rights, ap­pli­ca­ble reg­u­la­tion and core prin­ci­ples and val­ues, en­sur­ing an ‘eth­i­cal pur­pose’ and 2), it should be tech­ni­cally ro­bust and re­li­able since, even with good in­ten­tions, a lack of tech­no­log­i­cal mas­tery can cause un­in­ten­tional harm”.

Volvo’s 360c “bed on wheels” is still a con­cept at this point. Below: Mi­crosoft has part­nered with US su­per­mar­ket chain Kroger to cre­ate the ul­ti­mate “con­nected store ex­pe­ri­ence”

Newspapers in English

Newspapers from Ireland

© PressReader. All rights reserved.