Coming of Age of Artificial Intelligence
Transcription
Coming of Age of Artificial Intelligence
Coming of Age of Artificial Intelligence Disruption, Transformation Ahead Cyrille Bataller Robotics in Business Operations: Understanding the Landscape Robotics Spectrum Robotic Process Automation Unified Desktops/Mashups Basic Automation – Minibots Automation of transactions and work flow activities by handling input, processing, and output of data across systems. Artificial Intelligence Cognitive Robotics/ Virtual Agents Systems that gain knowledge from data as “experience” and generalize what is learned in upcoming situations. Sense, comprehend, act – and learn. Natural language dialogue. Multiple screens are consolidated into a single view for the operator thus saving the time to toggle between screens. Suitable for contact center type applications. Applying technology to manipulate existing application software to complete a process. Primarily based on XL, AutoHotKey, or Visual Basic. Copyright © 2015 Accenture All rights reserved. 2 AI: IT Systems that can Sense, Comprehend, Act – and Learn Maybe the Most Disruptive Emerging Technology on the Horizon Artificial Intelligence enables machines to interact naturally with their environment, people and data. These systems create more intuitive interactions and extend the capabilities of what either human or machine can do on their own. From coding to training AI enables intelligent systems that learn from a body of knowledge without analyzing and coding all business rules manually. Advances in machine learning, coupled with big data and cheap, ubiquitous cloud computing will unleash remarkable new potential for organizations across industries. Copyright © 2015 Accenture All rights reserved. 3 AI: IT Systems that can Sense, Comprehend, Act – and Learn A Collection of Technologies Underpinned by Machine Learning Artificial Intelligence encompasses multiple technologies that enable computers to • perceive the world (e.g., computer vision, audio processing or sensor processing) • analyze and understand the information collected (e.g., natural language processing or knowledge representation) • make informed decisions (e.g., inference engines, machine learning/deep learning or expert systems) Copyright © 2015 Accenture All rights reserved. Source: Gartner's Hype Cycle for Human-Computer Interactions 4 Artificial Intelligence History Various Attempts and “Winters” since Computing was Invented Initial research focused on Artificial Intelligence (AI), "the science aiming to create intelligent machines that are as capable as humans.” Google driverless car 1.7M miles accident-free Audi’s driverless race car Ask Jenn Ask Sgt Star Border control “robots” Source: Accenture Secondary Research Distinction between “strong” and “weak” AI. Also, recently, the focus has shifted from AI to Intelligence Augmentation (IA), intelligent systems that can support humans in their activities. Copyright © 2015 Accenture All rights reserved. 5 Artificial Intelligence Toolbox – Various Application Domains Solutions Text Analytics Research Assistants Image Analytics Multimedia Search Cognitive Robotics Video Analytics Identity Analytics Speech Analytics Data Visualization Domain-specific Calculations Capabilities Semantic Web Automatic Trend Detection Anomaly Detection Ontology Learning Computer Vision Sense Technologies Audio Processing Sensor Processing Comprehend Act Copyright © 2015 Accenture All rights reserved. Inference Engines Machine Learning Expert Systems - Text Analytics: process large volumes of unstructured text and classify/group etc - Research Assistant: natural language search engine to retrieve subset of relevant documents from large unstructured dataset - Deep Learning: apply deep learning to detect certain characteristics – e.g. cancer tumor in MRI scan, or find tall trees from a drone Multimedia Search: retrieve specific images, Virtual Agents Expert Systems videos or sounds based on content rather than user metadata - Cognitive Robotics: automate manual processes using UIs to access multiple systems e.g. testing, claims processing - Virtual Agent: automate helpdesks to solve user issues Recommendation Self-Adjusting IT - Expert System: learn a product manual and Systems Systems answer questions or assist a user with step by step instructions - Video Analytics: apply computer vision to CCTV cameras - Identity Analytics: recognize people based on what they have/are/know/do incl. biometrics Speech Analytics: speech to text/speech Augmented Affective recognition, text to speech Reality Computing - Data Visualization: ask questions in natural language on large structured dataset - Domain-specific Calculations: perform calculations for a specific domain in natural language - Recommendation Systems: provide “people Natural Language Processing like you” type recommendations - Self-Adjusting IT Systems: ability for IT system to adjust based on historical usage Knowledge Representation patterns 6 Text Analytics Case Study: PharmacoVigilance Monitoring Adverse Drug Reactions Creation of Adverse Drug Events (ADE) is required for assessing safety of a drug against usage, severity etc. NLP can help generate Adverse Event signals from various therapeutic reports, establish associations and report suspected new cases Assisted PV Processing Individual Case Safety Reports and Narratives database Regulatory warning and self reporting Social and Public Data Copyright © 2015 Accenture All rights reserved. Text Screening and extraction 1 Relevant text portions are extracted Semantic modeling and signal generation Entity Recognition 2 Entities like Patient, Symptom, Drug, Usage etc. are recognized 3 Relationships between Patients, Disease, Symptom, Drug, Dosage etc are established Case Suggestion 4 Based on Statistical association measures on aggregate data a case is suggested to the writer 8 Accenture Text Analyzer for PharmacoVigilance Copyright © 2015 Accenture All rights reserved. 9 Accenture Text Analyzer for PharmacoVigilance Inference on Expectedness and Causality Copyright © 2015 Accenture All rights reserved. 10 Cognitive Robotics Benefits of Robotic Automation Reduces cost, as we automate transaction processing Helps to provide higher productivity benefits Higher efficiency in process & reduction of non value added activities Reduces transactional errors Drives higher accuracy Mistake proof processes Enhances compliance & controls Drives improvements in business outcome through improvement in time, quality & cost of transaction Accelerating business outcomes without increasing program complexity or headcount Increases customer satisfaction Copyright © 2015 Accenture All rights reserved. 12 Case Studies Value Delivered by Robotic Automation Client Tools Deployed Productivity Impact Business Outcomes Large Retail client (Order Entry Indexing) Mashups 20% FTE reduction • Real Time TAT achieved • with 100% accuracy 25% FTE reduction • • • • 25% process automated Real Time TAT achieved BOI - Enabling sales team to focus on top line Potential new opportunities for @client Large retail client (Contract Management @Client) Mashups Large manufacturing (Invoice Processing for Single Line invoices PO & Non-PO Automated) Mashups Large Distribution (Order Entry process) OCR 21 FTEs • 52% reduction in transaction handling time (10.5mins to 5mins) • Enables adherence to peak demand TAT SLA • Elimination of defects due to manual entry Large Consumer Goods (Exit process automation) RPA 50% FTE reduction • Transactional handling time reduced from 90 mins to 45 mins • Improved DPA (Data Privacy Act) compliance Large retail client (PTP, RTR and Store Accounting) RPA 28 FTEs • 30% Productivity Benefit • Accuracy improved to 100% Copyright © 2015 Accenture All rights reserved. 30% FTE reduction • 30% productivity benefit • TAT Reduced to 1 Day (earlier 3 days) • 100% Accuracy 13 Image Analytics Case Study: Auto Insurance Claims Processing Key Technology: Deep Learning Deep Learning: different layers of abstraction to replicate human cognition Computer science: The learning machines Nature News & Comment, 2014 Copyright © 2015 Accenture All rights reserved. 15 Case Study: Auto Insurance Claims Processing Automated Classification of Car Damage Level Classify if a car is undamaged (new), damaged or totaled Problem Statement • An Insurance Company wanted to automate claims processing using advanced machine learning technology, namely Deep Learning. • When customers sent a picture of their damaged car, the Company would like to have the ability of automatically detect the level of damage and use it to, for example, order spare parts and possibly detect fraud, if any. • Accenture developed a Convolutional Neural Network algorithm (which belongs to the family of Deep Learning techniques) using a data set of toy images. •90% accuracy achieved. Value Delivered • By automatically detecting level of damage, an Insurance Company saves on sending a human to assess the damage • Apply the same technique for more use cases and other lines of businesses like Home Insurance with enhanced complexity and accuracy For Auto Insurance, spare parts could be ordered automatically For Home Insurance, evaluate building resistance, check if customers are telling the truth about additions to houses, identify multiple damages Similarities/differences in damage patterns could be used to detect fraud Copyright © 2015 Accenture All rights reserved. 16 Video Analytics Video Analytics Safety, Security, Operation and Marketing Insights Application of Computer Vision to automate observation of video surveillance cameras, generating real time alerts and detailed structured meta data for trend and anomaly detection Singapore Safe City Copyright © 2015 Accenture All rights reserved. French Telco counting/conversion rate Oil Major Fuel Terminals French Police Events 18 Example in Traffic Management 98.7% 94.5% count accuracy count accuracy 91.8% count accuracy Copyright © 2015 Accenture All rights reserved. 19 Examples in an Airport Copyright © 2015 Accenture All rights reserved. 20 Examples in Retail Copyright © 2015 Accenture All rights reserved. 21 Identity Analytics Case Study: UK BA, BAA – London Airports Self Clearance for EU ePassport Holders Copyright © 2015 Accenture All rights reserved. 23 Case Study: Amsterdam Schiphol Self Clearance for EU ePassport Holders Copyright © 2015 Accenture All rights reserved. 24 Virtual Agents Virtual Agents Goal: a Virtual Helpdesk Assistant that Interacts, Solves and Learns like a Human Agent Communicates with the customer using natural language speech, online chat and email… Forwards a customer query to another department when required… Understands the customer's query or issue and associated context... Provides fact or knowledge-based answers to customer queries when appropriate... Copyright © 2015 Accenture All rights reserved. Identifies and utilizes pertinent data and information specified by the customer... Accesses corporate backoffice systems to obtain specific information or triggers transactions when required... Asks for and utilizes additional customer documentation when necessary Performs offline tasks and follow ups on a customer query when required Learns new skills, information an capabilities through documentation, training observation and active discovery... Follows a specific process or script to resolve a customer query or issue and can deviate from it when appropriate… Provides a resolution to a customer query in a variety of ways where necessary… Escalates to a supervisor or another agent when appropriate... Disambiguates a customer query by asking clarifying questions when required... Documents the customer query and outcomes in the ticket tracker system... 26 Currently Being Piloted at an Oil Services Company • Phase I established the end-to-end flow where most customer enquiries would be escalated to human agents and “observed”. • Phase II increases the Customer Experience scope, to enable a virtual agent to answer 25-35% of customer inquiries, including: Urgent Payment Requests (18%) • Invoice Statements (12%) Missing Invoices – Not Due (8%) Offline Processes and follow ups Redirects (6%) MySupplier Portal Access and Updates (3%) Remittance Query (7%) Phase II also introduces Cognitive Robotics to access WQM (BMC Remedy) to trigger events. Copyright © 2015 Accenture All rights reserved. 27 Where To? Ripped from the Headlines … 30 Years Ago Copyright © 2015 Accenture All rights reserved. 29 Accenture Technology Vision 2015 Digital Business Era – Stretch Your Boundaries Copyright © 2015 Accenture All rights reserved. www.accenture.com/technologyvision 30 Workforce Re-Imagined More Effective, More Interesting Copyright © 2015 Accenture All rights reserved. 31 Workforce Re-Imagined Building Trust in Intelligent Systems Copyright © 2015 Accenture All rights reserved. 32 Turning Artificial Intelligence into Business Value • Cognitive technologies promise to automate or augment a wide range of work activities that today are largely done by humans, including manual workers and knowledge workers. • Delivering business value from artificial intelligence requires understanding the nature of the work being done along two dimensions: Data Complexity: degree to which complex unstructured changing data needs to be taken into account o Work Complexity: degree to which individuals need to apply their judgment and interpret a variety of information o Copyright © 2015 Accenture All rights reserved. 33 Turning Artificial Intelligence into Business Value – Healthcare Copyright © 2015 Accenture All rights reserved. 34 Turning Artificial Intelligence into Business Value – Banking Copyright © 2015 Accenture All rights reserved. 35 Creating an Artificial Intelligence Capability Video Analytics Natural Language Processing Technology-Rich Business-Outcome Focused Biometrics Sensor Processing Mini Bots Deep Learning Expert Systems Knowledge Representation Computer Vision People First Mindset Robotic Process Automation Emotion Recognition Inference Engines Machine Learning Neural Networks Gesture Recognition Ontologies Copyright © 2015 Accenture All rights reserved. 36 Artificial Intelligence Exponential Business Potential Doing Things Differently Doing Different Things Copyright © 2015 Accenture All rights reserved. 37 For More Information Cyrille Bataller Managing Director Artificial Intelligence www.accenture.com/aitechnology #AI @CyrilleBataller Mobile: +33 6 85 54 24 14 cyrille.bataller@accenture.com Copyright © 2015 Accenture All rights reserved. 38