Accelerating Innovation: How To Build Trust And Confidence .

3y ago
66 Views
2 Downloads
1.28 MB
16 Pages
Last View : 9d ago
Last Download : 4m ago
Upload by : Milena Petrie
Transcription

Artificial intelligence (AI) has the potential to transform businessmodels – a new industrial revolution. But without assurance over thestrategy, how it is implemented and the subsequent outputs andoutcomes, boards are wary of giving the green light to AI. And if theydo go ahead, they risk operating in the dark. So how can yourbusiness gain the control it needs to unleash the full potential of AI?Accelerating innovationHow to build trust andconfidence in AIResponsible artificial intelligence study 2017

IntroductionUnleashing the potentialArtificial intelligence needsboth more robust governanceand a new operating modelto realise its full potential.AI is emerging as the definingtechnology of our age, withmany industries alreadyutilising AI in some form. Andas humans and machinescollaborate more closely, andinnovations come out of theresearch lab and into themainstream, AI offerstransformational possibilitiesfor consumers, businesses andsociety as a whole.1, 2More than 60% of the 2,500 consumersand business decision makers in the US,who took part in a PwC survey onattitudes to AI, believe that it can helpprovide solutions for many of the mostimportant issues facing modern society,ranging from clean energy to cancerand disease1.At the heart of the opportunity forbusinesses is the ability to turn data intointellectual property (IP) – more than70% of business leaders in our surveybelieve that AI will be the businessadvantage of the future. We’re currentlyanalysing how AI could enhance thequality, personalisation and value ofhundreds of different products andservices across eight prominent sectors– we’ll be publishing the overall andsector-specific findings over the courseof 2017. ‘Bot me: A revolutionary partnership: How AI is pushing man and machine closer together’, PwC, April 2017 mg/bot-me.pdf)

Uncertainty holds backinnovationThe burning question is how to realisethis potential. Netflix offers a textbookexample of how a business can useaccumulations of data to create avirtuous circle of ‘machine learningadvantage’. The disruptive results havechanged the way we access mediacontent and led to a complete rethink ofbusiness models in the entertainmentsector. In the global economy as whole,however, progress on AI has beenmixed. Many organisations are still atthe beginning of the journey – less than40% of the business leaders taking partin our latest Global CEO Survey havebegun to explore the impact of AI ontheir future skills needs3, for example.Other organisations risk findingthemselves strategically hamstrung. Inpart, this stems from the difficulties ofchoosing from such a bewildering arrayof technologies, innovations andvendors. Many others are finding itdifficult to evaluate, mitigate andmanage the ‘black box’ reputational aswell as technological risks of using suchnew and largely untried technology. Keychallenges include determining whetherthe data is valid and what safeguardsare needed to ensure that machinescarry out human orders as intended.Related ethical considerations rangefrom is it acceptable to influence humanchoices to do consumers understandenough about how their data is used andwho has access to it? While somecompanies are keen to press ahead withAI, they can often find themselveschasing too many opportunities orfailing to adequately assess the fullbusiness case and associated risks – thecurrent landscape is littered with toomany abandoned proof of concepts.3, 472%of business leaders in the US believeAI will be the business advantage ofthe future267%of CEOs think that AI and automation(including blockchain) will have anegative impact on stakeholder trust intheir industry over the next five years4Robust evaluation and executionThese challenges underline the need fora new model of strategic evaluation,governance and delivery – without it,the uncertainties surrounding AI meanthat it will either remain stuck in the labwithin many organisations or they willfind themselves facing unacceptable andpotentially damaging risks.At the heart of this framework is theneed for trust and transparency increating responsible AI. The adoption ofAI may be met with scepticism from avariety of stakeholders, both within theorganisation and clients, regulators andothers outside. It’s therefore importantto consider how we can build trustamong all the affected stakeholders. Thekey to this is increasing transparencyand awareness around how AI is beingused, the jobs it performs, the decisionsit makes and the opportunities it brings– we believe this is the essence of‘responsible AI’.While no one can guarantee goodbehaviours from complex autonomousagents, there’s a series of best practicesthat include designing and monitoringcontrols, which would minimise risk andencourage responsible adoption of AI.In this paper, we explore the challengesthat such a framework would need toaddress, the opportunities it would openup and set out how it might work. Theobjective isn’t to stifle or slow downinnovation, but rather to accelerate it bygiving boards the assurance andplatform for execution they need todeliver the desired outcomes. ‘20 years inside the mind of a CEO What’s next?’, 20th Global CEO Survey, PwC, 2017 (ceosurvey.pwc)PwC Responsible AI 1

Harnessing disruptionHow AI is changing the rulesof the gameThe adoption of AI hasprofound implications foreveryone engaged in businessmanagement.The emergence of AI ispaving the way for a wholenew set of operating andbusiness models. The abilityto analyse levels of data thatare beyond humancomprehension and act oneach new set of informationallow businesses topersonalise experiences,customise products andservices, and identify growthopportunities with a speedand precision that’s neverbeen possible before.What is artificial intelligence?In their book, ‘Artificial Intelligence: AModern Approach’, Stuart Russell andPeter Norvig define AI as “the designingand building of intelligent agents thatreceive percepts from the environmentand take actions that affect thatenvironment”.5 The most criticaldifference between AI and generalpurpose software is in the phrase “takeaction”. AI enables machines to respondon their own to signals from the world atlarge, signals that programmers do notdirectly control and therefore can’tanticipate. The fastest-growingcategory of AI is machine learning, theability of software to improve its ownactivity, based on interaction with theworld at large.The spectrum of AI can be divided intothree: Assisted Intelligence, widelyavailable today, improves whatpeople and organisations arealready doing. Augmented Intelligence, emergingtoday, enables people to do thingsthey couldn’t otherwise do. Autonomous Intelligence, beingdeveloped for the future, establishesmachines that act on their own.5Some of the ways AI is makingits mark Keeping us informed: ‘Personalassistants’ such as Alexa and Siri, aswell as banking and mobile phonenetwork operator chatbots. Predicting behaviour: The UKNational Health Service are pilotingmachine learning to predictoutpatient non-attendance at a UKhospital, optimising scheduling andrescheduling of appointments whileunderstanding the drivers that caninfluence patient behaviour andenable actions to be taken to reducerates of non-attendance. Keeping us well: AI is being used toaid medical diagnosis – our researchshows that a significant proportionof people worldwide are willing tochoose certain treatments, tests orservices administered by an AIor robot6. Keeping us engaged: Telecoms andmedia companies have been usingmachine learning customer analyticsto predict and then recommendactions to prevent customer turnover. Anticipating demand: Retailers arebeginning to use deep learning topredict customers’ orders a weekin advance. ‘Artificial Intelligence: A Modern Approach’, Stuart Russell and Peter Norvig (Pearson, 2009)‘What doctor: Why AI and robotics will define the New Health’ html)6 2 Responsible AI PwC

Customisation for all: Roboadvice has made it possible to offercustomised investment solutions toa wider range of consumers. Untilrecently, this level of investmentadvice was only available to highnet worth (HNW) clients. Improving quality: Manufacturersare using AI to improve qualitycontrol, reduce production linedowntime and increase the speedand yield of industrial processes. Intelligent processes: Intelligentprocess automation is driving hugesavings in finance, HR andcompliance. Robotic processautomation is combined with AI toperform high volume, routine tasks.7Navigating the sheer breadth ofalgorithms and applications that fallunder the banner of AI has become aformidable task in its own right. To date,a lot of the focus has been on automationof tasks that are already carried out7. Yetas workers are freed from routine tasksand human and machines begin tocollaborate more closely, the realbreakthroughs will come from theability to make more insightful decisionsand the emergence of completely newaugmented intelligence-led businessmodels. Entertainment is a clearexample of a sector that has alreadyundergone significant disruption andchange. Driverless cars are one of themany ways that AI is set to transformeveryday lives and the businesses thatsupport this.Commercially-applied AI has expandedin recent years, driven by a combinationof computing power, the availability ofhuge datasets and advances in machinelearning (which includes deep learning).While often used for predictiveanalytics, as well as image and speechclassification, machine learning can becombined with elements of ‘traditional’AI such as natural language processing,strategic planning and logical reasoningto deliver powerful autonomous agents.So how prevalent is AI? Outside of largetech companies that have been utilisingAI in service delivery for a number ofyears, much of the innovation is still inits infancy and is largely confined to thelab in the form of proof of concepts orR&D. The focus for business now has tobe on creating an environment whichfosters successful transition into realworld value delivery. We explore the impact of automation and AI on production and employment in ‘Will robots steal our jobs? The potential impact ofautomation on the UK and other major economies, March 2017 keosection-4-automation-march-2017-v2.pdf)PwC Responsible AI 3

New approachAs Figure 1 highlights, the adoption of AI demands a new way of thinking abouttechnology, business development and strategic execution, along with the reshapedoperating model and decision making processes that underpin this. And this affects theentire business, rather than just technology and innovation teams.Figure 1: Rethinking the way you do rce: PwC4 Responsible AI PwCTraditional approachNew approachTechnology for informationmanagementTechnology that managesyour businessData as business intelligenceData as your differentiatingintellectual propertyDeterministic approachDirectional (iterative) approachUser experience as anapplication layerUser experience as the primaryapplication featureDecision making hard codedDecision process learnt bysoftwareInformation retrieval as fact fromdatabaseInformation retrieval mostprobable correct answerLinear technology developmentIterative technology and businessmodel developmentBusiness management teamsspecify, technology team buildsBusiness subject matter expertsintegrated into technologydevelopment teamsSteady state technology,punctuated with upgradesDynamic, adaptive models.Continuous test drivendevelopmentTechnical risks dominated bysystem downtime and errorsTechnical risks include learnedand unexpected behaviourCyber attacksAdversarial attacks

PwC Responsible AI 5

So what are the keyconsiderations for bringing AIinto the centre of your businessand operating model?StrategyAI may be intelligent, but it’s still a machine.A common problem is believing the AI willmagically learn without humanintervention.1. Aligning with your strategicgoalsIt’s vital to align AI innovation with corestrategic objectives and performanceindicators, rather than allowing ascattered series of initiatives to operatein isolation. In our experience, a lot oforganisations have set various pilots intrain. What most aren’t doing is taking afundamental look at how AI coulddisrupt their particular business andthen determining the threats andopportunities this presents.2. Don’t expect magicAI may be intelligent, but it’s still amachine. A common problem is believingthe AI will magically learn withouthuman intervention. In reality, you haveto put a lot of effort into acquiring andcleansing data, labelling and trainingboth machines and employees8.3. Clear about your partnersEverywhere you look, there are start-upsoffering solutions to this andopportunities for that. Partnership withthese vendors accelerates innovation,agility and speed to market. But it’sclearly important to pick your spot. Thisincludes being clear about the strategicand operational priorities you’re lookingto address through the choice of partner.It’s also important to bear in mind thatwhile vendors may be good at selling thepossibilities, they’re not always as clearabout how to deliver them – the way theylook at development risks is certainlyvery different from what you’re used to.In a high risk and fast-moving vendorlandscape, the first consideration is thefinancial viability of the potentialpartners – will they still be there whenyou need them? It’s also important todetermine how to acquire the necessarydata, develop the knowledge needed todeploy your new capabilities and how tointegrate new platforms into existinginfrastructure. When buyingcommercially available off-the-shelfsoftware, a proof of conceptdevelopment phase is often necessary.4. Opening up to scrutinyBefore you adopt AI, you clearly need toknow what it’s doing and how. Thisincludes ensuring the software cancommunicate its decision makingprocess in a way that can be understoodand scrutinised by business teams. Inparticular relation to machine learning,it’s important to think about how toensure the software will deliver theanticipated results. Boards want thisassurance before they proceed.Regulators are also likely to expect it.Algorithmic transparency is part of thesolution, though this may require a86 Responsible AI PwCtrade-off between decision makingtransparency, system performance andfunctional capabilities.5. Demonstrating regulatorycomplianceRegulators need to move quickly to keeppace with emerging technologies. Wemay see regulatory constraints thatprevent adoption in key regulatedindustries such as health and financialservices. Developments such as the EU’sGeneral Data Protection Regulation(GDPR) are heightening the challenges.Staying compliant with relevantregulatory requirements is essential tobuild trust in your AI platform.6. Organisational structureThe changes in your business models aspart of your overall AI strategy will alsoneed to be reflected in your organisationalstructure. Your organisation needs adedicated AI governance structure, thiscould include a nominated member of theC-suite and a central hub of technicalexpertise. Embedding data scientiststhroughout your business either throughtraining or hiring is essential to achieve AIorganisational maturity. P wC’s Dr Anand S Rao explores the ‘Five myths and facts about artificial intelligence’ in Predictive Analytics and Futurism,Society of Actuaries, December 2016

Design1. Opening up the black boxAI applications can communicate withcustomers and make important businessdecisions. But a lot of this is carried outwithin a black box, with the lack oftransparency creating inherentreputational and financial risks. It’simportant to ensure that the software isdesigned in a way that is as transparentand auditable as possible.Proper governance and protectioninclude the ability to monitor componentsystems. It would also include the abilityto quickly detect, correct and, if not,shut down ‘rogue’ components withouthaving to take down whole platforms.Related priorities include identifyingdependencies and being able to makemodifications with minimal disruptionif regulations or some other aspect of theoperating environment changes.2. C reating a compelling userexperienceMany AI applications deploy highlysubjective user experience performancemetrics akin to IQ, personality, andpredictability. Even though the bulk ofdevelopment may focus on the analytics,the success of the product will bedetermined by an emotional response.This subjectivity means that frequentfeedback is required between productowners and developers to make sureevolving expectations and functionalityare properly managed. Often it makessense to bring in specialist user interfacevendors or use your in-house digitalteam alongside the core analytics team.AI may excel and often surpass humansat particular tasks or in certain subjectdomains, but is generally incapable ofextending these skills or knowledge toother problems. This is not obvious topeople who have to interact with AI,especially for the first time, and cancause frustration and confusion.The most effective controls are built into thedesign and implementation phase, enablingyou to catch issues before they become aproblem and also identify opportunities forimprovement.Branding and persona development(‘functionality framing’) are thereforekey design considerations. Get it rightand very basic software can appearhuman. Get it wrong and users willgive up.Some of the analysis performed by AIwill inevitably be probabilistic based onincomplete information. It’s thereforeimportant that you recognise thelimitations and explain this tocustomers. Examples might include howyou present recommendations oninvestments from robo-advisors.3. E mbedding the controlframeworkThe most effective controls are builtduring the design and implementationphase, enabling you to catch issuesbefore they become a problem and alsoidentify opportunities for improvement.An important question is who designsand monitors the controls? Both thebreadth of application and the need tomonitor outcomes requires engagementfrom across the organisation. Controldesign requires significant input frombusiness domain experts. Specialistsafety engineering input is likely to berequired for physical applications.A key part of implementation is breakingthe controls down into layers(‘hierarchical approach’).At a minimum, there would be a hardcontrol layer setting out ‘red lines’ andwhat to do if they’re breached. Examplesmight include a maximum transactionvalue for a financial market tradingalgorithm. In more complex applicationssuch as conversational agents, you couldintroduce a ‘behaviour inhibitor’ thatoverrides the core algorithm when thereis a risk of errors such as regulatoryviolation or inappropriate language.These core controls can be augmentedby ‘challenger models’, which are usedas a baseline to monitor the fitness andaccuracy of the AI techniques or look forunwanted bias or deviations as themodels learn from new data. Moreover,this approach can be integrated withcontinuous development to improveexisting models or identify superiormodels for system upgrades.PwC Responsible AI 7

Development1. Rethinking programmemanagementApplying conventional planning, designand building to such data-dependentdevelopments is destined to fail.Innovating and proving the conceptthrough iterative development is neededto handle the complexity of the problemsencountered and requires a high level ofengagement from the product owners.2. Managing data dependencyAI functionality is heavily datadependent for machine learning modeltraining and is likely to need a store ofinformation known as a ‘knowledgebase’. This often means initial designspecifications and expectations ar

What is artificial intelligence? In their book, ‘Artificial Intelligence: A Modern Approach’, Stuart Russell and Peter Norvig define AI as “the designing and building of intelligent agents that receive percepts from the environment and take actions that affect that environment”.5 The most critical difference between AI and general-

Related Documents:

the overall innovation strategy, planning, and implementation within their focus areas. The innovation executive often develops and executes an enterprise strategy for innovation. This executive is responsible for leading the organization's innovation team, guiding innovation priorities that support the health system's strategic goals,

work/products (Beading, Candles, Carving, Food Products, Soap, Weaving, etc.) ⃝I understand that if my work contains Indigenous visual representation that it is a reflection of the Indigenous culture of my native region. ⃝To the best of my knowledge, my work/products fall within Craft Council standards and expectations with respect to

PFI:AIR-TT: Partnerships for Innovation: Accelerating Innovation Research -Technology Transfer PFI:BIC: Partnerships for Innovation: Building Innovation Capacity INSPIRE: Integrated NSF Support Promoting Interdisciplinary Research and Education EFRI: Emerging Frontiers in Research and Innovation

Build State 0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Ratio of Slow Builds in a Project (b) Ratio of Slow Builds Passed Errored Failed Errored and Failed Build State 0.0 0.2 0.4 0.6 0.8 1.0 Ratio of Build Time in a Project (c) Ratio of Build Time Figure 1: Distributions of Build Time, Ratio of Slow Builds and Ratio of Build Time w.r.t. Build States

Basic Concepts of Innovation and Innovation Mgmt M.Lorenzo 2010-03-253 Introduction What is Innovation? Innovation is typically understood as the introduction of something new and useful Innovation is

reads and genomes. Accelerating this task requires parallel computing. Among the current parallel computing platforms, the graphics processing unit (GPU) provides massively parallel computational prowess that holds the promise of accelerating scientific applications at low cost. In this paper, we propose GPU-RMAP, a massively par-

Accelerating Ansys Fluent Using NVIDIA GPUs Accelerating ANSYS Fluent 15.0 Using NVIDIA GPUs DA-07311-001_v01 9 3. CHANGING AMGX CONFIGURATION In ANSYS Fluent 15.0, the Algebraic Multigrid (AMG) linear system solver used on the CPU is different from that used on the GPU. In the latter case, the AmgX library is used to perform the

We present the first experimental observation of accelerating beams in curved space. More specifically, we demonstrate, experimentally and theoretically, shape-preserving accelerating beams propagating on . transverse directions, respectively, such that k2 ¼ k2 R þk2 T. Equation (4) describes the field dependence on the radial