12. AC?L EYLEM PLANI 12.1 Amac ... plan?, ar?tma tesisi i?letme personelinin herhangi bir acil durum kar??s?nda almas? gereken
Report CopyRight/DMCA Form For : Controlling Ai The Imperative For Transparency And Kpmg
Our authors, Martin Sokalski Professor Dr Sander Klous Swami Chandrasekaran. Principal Advisory Emerging Partner Data Analytics Lead Managing Director. Technology Risk Services KPMG in the Netherlands KPMG Innovation Enterprise Solutions. KPMG in the U S KPMG in the U S, Martin Sokalski is a Global Leader for Sander Klous is a Data Analytics Swami Chandrasekaran is a leader in. KPMG s Emerging Technology Risk Leader for KPMG in the Netherlands and KPMG s Innovation Enterprise Solutions. practice He helps organizations around the a professor of Big Data ecosystems for group and helps lead the architecture. globe embrace the art of the possible business and society at the University of technology and creation of Intelligent. enabled by emerging technologies Amsterdam He has a PhD in High Energy Automation including AI and emerging. like artificial intelligence by facilitating Physics HEP and has worked for over a technology offerings He has led incubation. ideation innovation and responsible decade on a number of projects for CERN design and creation of several complex AI. adoption Over the years he has helped the world s largest physics institute in products and solutions across a wide range. many organizations across multiple Geneva His best selling book We are Big of challenges in areas such as Tax Audit. sectors assess design and implement Data was runner up for the management Industrial Automation Aviation Safety. new digital operating and governance book of the year award in 2015 His new Contact Centers Insurance Claims Field. models to help them achieve desired book Building Trust in a Smart Society Service Multimedia Enrichment Social. business outcomes while embedding key is a top selling management book in Care Digital Marketing and Mergers and. governance trust and value imperatives the Netherlands Sander has significant Acquisitions Swami also has significant. Martin regularly speaks at conferences experience in large scale distributed experience in business process automation. and contributes to thought leadership on computing real time systems and data and systems and data integration Swami is. artificial intelligence digital transformation processing technologies His current focus an IBM Distinguished Engineer Emeritus. and emerging technologies Martin believes is on the broad use of reliable analyses He recently published Learning to Build. that adoption of AI at scale is currently ethical algorithms and trusted analytics in Apps Using Watson AI and with 20. inhibited by lack of trust and transparency a way that is valued by clients and society patents filed and 17 issued most of them. explainability and unintended bias and he at large in the field of AI he was appointed an IBM. aims to work with industry leaders to solve Master Inventor. for that challenge, 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms affiliated with KPMG. International Cooperative KPMG International a Swiss entity All rights reserved. 1 Introduction,5 Key developments,11 Governance,and ethics of AI. 15 Key to governing AI a framework,that helps enable transparency. 17 AI in Control a framework to,govern algorithms, 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms affiliated with KPMG International Cooperative. KPMG International a Swiss entity All rights reserved. Controlling AI,Introduction,World changing,technologies over human. history all involve a,common element Control,Steam and light and a. long list of inventions and,technologies emerged,because we were able to. guide natural forces into,transformative power,Aviation would not exist. without the mastery we,have attained over flight,Artificial Intelligence AI. has the potential to be just,as world changing, 2019 KPMG LLP a Delaware limited liability partnership and the. U S member firm of the KPMG network of independent member. firms affiliated with KPMG International Cooperative KPMG. International a Swiss entity All rights reserved,But we don t know the full extent. of what AI can do for the world, And like other transformative technologies the power and promise. of AI can only be fully unlocked by our understanding and control. of its build and actions This is why companies need to establish. an overall management policy for AI with a focus on responsibly. unleashing the power of these technologies, AI unveils a world hidden in complexity The insights from. algorithms that learn and continuously evolve are changing our. businesses and our lives Many scientists see a future where some. of the deepest mysteries and intractable problems facing humanity. can be solved We are already seeing the benefits emerge from. algorithms that discover subatomic particles and help capture. the first photograph of a black hole to the enterprise level where. sophisticated data and analytics driven by AI are making mission. critical decisions that affect the bottom line and the brand and the. health and safety of consumers,AI on the ground, Picture a line of business owner LOB for consumer loans at a large. financial institution A situation involving bias and discrimination has. surfaced along with a headline or two in the news During a board. meeting one member after another asks this leader and the chief. digital officer CDO to explain the decisions and rationale behind. the denial of loans to applicants of a certain age group or race At. play is an AI algorithm that produced the results or augmented a. decision by loan officers in the field The problem for these two. leaders No one can explain exactly why the algorithm did what. Moments similar to this are playing out in areas across business. and the public sector recruiting transportation marketing. healthcare college admissions housing and the management. of smart cities Any organization that builds or adopts advanced. continuous learning technologies is tapping into a power for insight. and decision making that far exceeds the capabilities of the human. mind This is a massive opportunity, But algorithms can be destructive when they produce inaccurate or. biased results an inherent concern amplified by the black box facing. any leader who wants to be confident about their use That is why. in the midst of enormous excitement around AI there is hesitancy. in handing over decisions to machines without being confident in. how decisions are made and whether they re fair and accurate This. is a trust gap, 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm. of the KPMG network of independent member firms affiliated with KPMG International. Cooperative KPMG International a Swiss entity All rights reserved. Controlling AI, Gaining confidence According to the KPMG 2019 U S CEO Outlook. organizations are at different levels of their AI,The enormous benefits of AI will fully. emerge only when algorithms become,deployment journeys. explainable and hence understandable in,simple language to anyone The trust gap. exists because there is no transparency,of AI instead there is an inherent fear of. have begun limited,the unknown surrounding this technology. Gaining trust also involves understanding implementation for. the lineage of the AI models and protecting specific processes. them and data that forms them from,different types of adversarial attacks. and unauthorized use Critical business,decisions made by AI affect the brand and. consumer trust in the brand and they can,piloting AI. have an enormous impact on the well being,or safety of consumers and citizens No one. wants to say because the machine said,so No one wants to get AI wrong. Closing the trust gap,Fair and explainable AI is more than a. big ask in the C suite and the boardroom,today it s a demand KPMG s 2019 CEO. Outlook1 for example found that 66,of leaders surveyed overlooked insights. provided by computer driven data analysis,because they were contrary to their. experience or intuition,For most organizations AI is still in the lab. so to speak deployed on a functional level,and not yet an integral part of the decision. have already fully, making in the business although that is implemented AI. rapidly changing for some of their,KPMG 2019 U S CEO Outlook Agile or. Irrelevant Redefining Resilience June 2019, 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms affiliated with KPMG. International Cooperative KPMG International a Swiss entity All rights reserved. What s the,For AI to move ahead toward,the common good for leaders. to assume responsibility and,accountability over the results. it s essential to establish,a framework powered by. methods and tools to facilitate,responsible adoption and scale. The true art of the,possible for Artificial,This report is for. leaders involved in Intelligence will become,unlocked as soon as. the world of Artificial,Intelligence and Machine,Learning algorithms. The business and,there is more trust,compliance imperative to. understand and be confident,and transparency,in AI technologies has. reached critical mass This can be achieved,This paper explains the. urgency and describes,by incorporating,methods and tools that can. help leaders govern their,foundational AI program,AI programs. imperatives like,integrity explainability,fairness and resilience. Martin Sokalski,Principal Advisory Emerging,Technology Risk Services. KPMG in the U S, 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm. of the KPMG network of independent member firms affiliated with KPMG International. Cooperative KPMG International a Swiss entity All rights reserved. Controlling AI,developments,Based on interviews,with executives driving. AI strategy at large,companies we heard,a consistent message. Many companies are just,beginning to invest in,AI control frameworks. compared to other AI,deployment priorities2,KPMG 2019 Enterprise AI Adoption Study. 2019 KPMG LLP a Delaware limited liability partnership and the. U S member firm of the KPMG network of independent member. firms affiliated with KPMG International Cooperative KPMG. International a Swiss entity All rights reserved,Gaining trust around AI is a top goal of leaders. 45 of surveyed executives say that trusting AI systems was either. challenging or very challenging3, New policy initiatives and regulations around data. and AI signal the end of self regulation and the rise. of a new oversight model4, Most leaders aren t clear on what an AI governance. approach should be, Some 70 say they don t know how to govern algorithms5. Companies are struggling to decide who is,accountable for AI programs and results. During our interviews we heard that most companies are still. trying to determine who has authority over AI deployment Some. companies have established a central authority in an AI council or. Center of Excellence others have assigned responsibility to different. leaders like the Chief Technology Officer or Chief Information Officer. A framework that includes technology enabled,methods can help address the inherent risks and. ethical issues in AI, The objective is to help business users gain control over their AI. programs by enabling four trust anchors integrity explainability. fairness and resilience, Forrester Research Q2 2018 Global AI Online Survey. AI Internet Policy Proposals Signal Shift Away From Self Regulation Wall. Street Journal WSJ Pro Artificial Intelligence April 9 2019. Source KPMG Why AI Must Be Included in Audits 2018. 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent. member firms affiliated with KPMG International Cooperative KPMG International a Swiss entity All rights reserved. Controlling AI,The need to,know trust,The cost of getting AI. wrong extends beyond,the financials lost,revenue fines from. compliance failures to,reputational brand and,ethical concerns. We just pictured a CDO and LOB leader trying to explain. the outcome of a single model before the board The. layers of accountability extend from the C suite and the. line of business owner for the entire credit card division. of a bank someone who owns everything related to, this business including the AI models all the way to. the customer level with a loan officer who may face. accountability and who in many ways represents, 2019 KPMG LLP a Delaware limited liability partnership and the. U S member firm of the KPMG network of independent member. firms affiliated with KPMG International Cooperative KPMG. International a Swiss entity All rights reserved, Key business decisions at It isn t that hard to determine whether the. conclusions they reach are acceptable and sound, scale have a determining and scalable supervision is relatively easy. effect on success as These rules can get very complex especially. an example when the number of attributes also known as. features or variables in the data or the number of. Should the division approve a credit card for records increases. a customer,Machine learning and deep learning and other. Among the decisions for each customer the types of AI are creatures of a different kind They. annual percentage rate the spending limit and a are trained to learn from data commonly referred. long list of other factors Machine learning models to as ground truth instead of being explicitly. are typically making these decisions for millions of programmed which means they can understand. customers In a very real sense given the scale learn uncover the nuances and the patterns in the. the business is in the hands of a handful of smart data they can handle a very large set of attributes. data scientists and the machines they build and and are often significantly more complex in how. train using ground truth created from historical they do what they do. Think of training a prediction model from a set of. a million past loan applications which in turn uses. Autonomous algorithms 100 attributes Think of detecting a tumor from a. then vs now million MRI images Think of classifying emails. Once trained and evaluated these models can,Most algorithms today are relatively simple and. be provided with new or unseen data from which,deterministic They produce the same output from. they can make predictions They are probabilistic in. a predetermined set of states and a fixed number,nature and respond with a degree of confidence. of rules The approaches for evaluating them for, validity and integrity are largely established and While all of these aspects are good it can be. adopted In fact in our estimation over 80 of the unclear what the models are doing what they. leading practices needed to maintain their accuracy learn particularly when employing opaque deep. and effectiveness are known learning techniques such as neural nets how they. will behave or whether they will develop unfair,Think of expert systems in manufacturing Think. bias over time as they continue to evolve That s, of actuarial science that uses deterministic rules. why understanding which attributes in the training. or decision tables in insurance Think of robotic,data influence the model s predictions has become. process automation in financial services,very important. 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms. affiliated with KPMG International Cooperative KPMG International a Swiss entity All rights reserved. Controlling AI,Algorithmic Risk A number of techniques. including those based,Among the risks are,adversarial attacks that. With complex continuous,learning algorithms, Trust in the Machine on renormalization hit the very foundation humans need to know. group theory have been of these algorithms by more than just the data. Let s take a closer look at a, proposed 6 As models poisoning the models or or attributes and their. potential problem for the CDO across AI tasks including tampering with training respective weights to fully. and line of business owner computer vision speech data sets potentially realize the implications of. for the loan division of a big recognition and natural compromising privacy the AI getting it wrong or. financial firm language processing the user experience going rogue they need. become more sophisticated intellectual property to understand aspects. If an error hides within an and autonomous they and any number of other such as the context and. algorithm or the data feeding take on a higher level of key business aspects intended purpose under. risk and responsibility Consider the impact on which the model was. or training the algorithm it, When left untrained for lives or an environment developed who trained. can influence the integrity long periods things can of an adversarial attack them provenance of the. and fairness of the decision go awry runtime bias in medical devices or data and any changes. made by the machine This creep concept drift and industrial control systems made to it and how the. could include adversarial data issues such as adversarial Tampering with data models were and are. attacks can compromise could disrupt consumer served and protected And. or data masking as ground, what these models learn experiences by providing they need to understand. truth The business leaders Imagine compromised MRI inappropriate suggestions what questions to ask and. are on the hook for preserving scans or traffic lights being in retail or financial what key indicators to look. the brand reputation for manipulated in a smart city services Such attacks for around an algorithm s. the firm even as the AI might ultimately erode the integrity explainability. Continuous learning competitive advantage fairness and resilience. models increasingly make algorithms also introduce that the algorithms were. decisions that might not be a new set of cybersecurity intended to create. understood or in line with considerations Early, corporate policies corporate adopters are still grappling. values guidelines and with the magnitude of risks, presented by these issues The No 1 challenge for AI. the public s expectations,Multiply these issues by the. on the business adopters is quality data The, number of algorithms the CTO of a government agency. loan division is utilizing This specifically stated in our Global. is when trust weakens or, actually evaporates AI survey that if they can t trust. data they can t use AI7, A n Exact Mapping Between the Variational Renormalization Group and Deep Learning Pankaj Mehta David J Schwab 2014. Forrester Research Q2 2018 Global AI Online Survey. 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms affiliated with KPMG. International Cooperative KPMG International a Swiss entity All rights reserved. The anchors of trust, When you break down all the actions and capabilities needed to secure trust in your algorithms and models and hence your brand KPMG. believes that four dimensions emerge,Algorithm integrity Resilience. Think of a home inspection that checks the bones of Here is where we re talking about the robustness. a house as a metaphor for determining the structural and resilience of the models or algorithms that are. flaws and integrity of an algorithm What leaders need to know deployed or served The served models are typically exposed. is this the provenance and lineage of training data controls over as APIs or embedded within applications and they need to be. model training build model evaluation metrics and maintenance portable and operate across diverse and complex ecosystems. from start to finish and the verification that no changes. compromise the original goal or intent of the algorithm Also Resilient AI should cover all the aspects of secure adoption and. key would be continuous monitoring of the model performance holistically address risks through securely designed architecture. metrics including concept drift detection and the detection of anomalies using AI concepts like generative. adversarial networks that pit algorithms against each other to. produce better and more nuanced outcomes The goal is to. Explainability help ensure all the components are adequately protected and. Understanding the reasons a model made a monitored Why External circumstances can lead to errors when. prediction and being able to interpret the reasons algorithms are unable to correct or compensate for data that is. is essential in trusting the system especially if one has to inaccurate or anomalous Protecting the usage and feedback data. take an action based on those probabilistic results This is a that could be used to continuously train the models is also critical. subjective capability in AI Being able to explain why and how Basic actions include continuously monitoring models endpoints. a model produced an output insight decision depends on the and controlling access to the models. definition of success established and the overall governance of. the algorithm from the assemblage of ground truth that is clean. sufficient and appropriate to the continuous assessment of. results Several approaches exist including LIME an explanation A central question needs to be. technique that focuses on local or isolated aspects of resolved Who among the humans. decisioning 8 and Defense Advanced Research Projects Agency is accountable for the results of AI. DARPA Explainable AI XAI program which aims to create a Accountability is a crucial governance issue that must. suite of machine learning techniques leading to more explainable be established across all AI initiatives down to each. models with an explanation interface individual model We found significant variation among. KPMG s 2019 Enterprise AI Adoption Study in assigning. authority and accountability Some organizations have. Fairness ethics and accountability created a centralized authority such as an AI council. AI and algorithms won t be trusted if they re not fair others have assigned it to functions such as the office. For them to be fair they need to be designed and built of the chief technology or chief information officer But. as free from bias as possible and they need to maintain fairness few organizations have solid accountability practices in. as they evolve Attributes used to train algorithms need to be place a leadership gap that can weaken trust internally. relevant appropriate for the goal and must be allowed for use and among external stakeholders A big reason for this. In some instances however personal information is relevant to missing link Most organizations lack tools and expertise to. the model as in healthcare when gender or race can be a critical gain a full understanding and introduce transparency into. part of studies or treatment Careful oversight and governance is their algorithms. needed to make sure proxy data doesn t train a model A postal. code for example can be a proxy for ethnicity or income and. inadvertently produce biased results and downstream risks just. one being regulatory violations Techniques must be applied to. understand bias that inherently exist in the data and mitigate. them using approaches such as rebalancing reweighting or. adversarial debiasing, Tools for continuous monitoring as well as governance are. essential to help ensure models that are continuously trained. with usage and feedback data don t cause bias to creep in 8. Marco Tulio Ribeiro Sameer Singh Carlos Guestrin Why Should I Trust. during runtime You Explaining the Predictions of Any Classifier 2016. 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of independent member firms affiliated with KPMG International Cooperative. KPMG International a Swiss entity All rights reserved. Controlling AI,Governance,and ethics of AI,Governance and ethics. become the how of,responsible adoption of AI,by addressing the risk that. complex algorithms could,take a wrong turn, Look at the rules and regulations that govern the aviation. industry and the internal best practices that dominate. procedures at each individual airline from the C suite to. the cockpit Look to the trust placed in the experience. by everyone involved the crew the passengers and, businesses that transport valuable assets by air This is. what industry must aim for with AI, 2019 KPMG LLP a Delaware limited liability partnership and the. U S member firm of the KPMG network of independent member. firms affiliated with KPMG International Cooperative KPMG. International a Swiss entity All rights reserved, A tipping point has arrived in terms of Governance. the need for effective governance and,A long list of questions emerges when one. responsible adoption and scale of AI In,digs deep into the workings of AI and. many cases organizations are developing,many of them are human issues Why. internal policies and governance functions,and how were certain use cases chosen. to oversee any matters relating to AI in an,as candidates for AI Why did the team. effort to engender trust and transparency,choose the features it chose and exclude. across the enterprise and external,what it excluded How do we measure and. stakeholder groups including consumers,demonstrate success or explain failures. Why did the algorithm do what it did and,In the U K and in the E U with its evolving. who is responsible for the outcome,General Data Protection Regulation the. Because the algorithms said so will not,tide is now firmly moving toward the. work for leaders and the general public as,establishment of oversight And the timing. these systems become ever more powerful,is a good thing as the seeds of AI are firmly. and pervasive,in the ground and growing The scale is. not there yet but these technologies are,The need Seeing the big picture and. set to expand within the enterprise and,setting the right tone at the beginning. across industry sectors and assume greater,If you don t have a governance or an. autonomy and responsibilities Now is the,operating model construct for AI it will be. time to set a framework for governance,difficult to achieve the desired business. and ethics around the anchors of trust,outcomes or have confidence in your. Controlling AI will help enable a responsible,AI s integrity explainability fairness. expansion of power,or resiliency Governing AI is also the. right thing to do in terms of trust and, Governance and ethics visibility That means looking at enterprise. frameworks and governance through a,Assessing and securing the trust. new lens around people process and,anchors of AI can come from a new. technology across the entire lifecycle,set of leading practices and methods. from a model s early stages through,aimed at maintaining control over AI and. strategy delivery monitoring training and,machine learning algorithms An effective. capabilities and continuing measurement,governance strategy lays a foundation. of trust and transparency by putting in,place the mechanisms and tools that. will continuously measure AI Leaders,will be able to make informed decisions. and their organizations will build a,culture of accountability that is stronger. and consciously representative of an,organization s ethical compass. 2019 KPMG LLP a Delaware limited liability partnership and the U S member firm of the KPMG network of. independent member firms affiliated with KPMG International Cooperative KPMG International a Swiss entity All.
31-Mar-2020 3 Views 10 Pages
12. AC?L EYLEM PLANI 12.1 Amac ... plan?, ar?tma tesisi i?letme personelinin herhangi bir acil durum kar??s?nda almas? gereken
31-Mar-2020 4 Views 19 Pages
Coming#Home#to#the#Internet Ijoined#Washington#University#in#1990#in#their# CIMCenterproviding trainingforengineers. CIMLabwasconnectedtotheInternet " I"developed"an ...
31-Mar-2020 4 Views 71 Pages
How To Build A Website Or A Blog From Scratch The guide will walk you through setting up a your own website, on your own web hosting, without paying a designer or developer tons of money to get the job done!
31-Mar-2020 3 Views 11 Pages
FounDations oF Human ResouRCe Development 3 all outdoors. Lee (2001) refuses to define HRD as she argues that to proffer definitions of HRD is to misrepresent it as a thing of being rather than a process of becoming.She
31-Mar-2020 2 Views 19 Pages
7 Towards a Model of Learning and Development Practice Steven Hodge, Griffith University, Australia Erica Smith, Federation University, Australia Llandis Barratt-Pugh, Edith Cowan University, Australia It is widely acknowledged that learning and development (L&D) is key to well-being, innovation and
31-Mar-2020 2 Views 42 Pages
The Church of the Christ Pantocrator, southwest view, Visoki De?ani Monastery
31-Mar-2020 3 Views 48 Pages
coming soon! the new center for agrarian learning (cal) at mchenry county college will launch its first series of classes and workshops in spring 2020. A new path for the future food economy. Food is central to our well-being, our environment, our shared sense of community.
31-Mar-2020 2 Views 5 Pages
Photovoltaic Power Control Using MPPT and Boost Converter A.Attou, A.Massoum and M.Saidi S P N-pv I I d sc R s R sh I ... M. Arrouf, Optimization of Inverter, ...
31-Mar-2020 2 Views 21 Pages
prepare for the next, and advance your marriage into spring and summer. The seasons of marriage come and go. Each one holds the po-tential for emotional health and happiness, and each one has its challenges. The key is to develop the necessary skills to enhance your marriage in all four seasons.
31-Mar-2020 3 Views 6 Pages
(2007), Arrouf and Bouguechal (2003) and Arrouf and Ghabrour (2007). The stator voltages (vsa, vsb, vsc) are expressed in terms of the upper switches as follows: 1 2 3