Conversational AI is reshaping the human-machine interaction November 20202ContentsWhat is CAI0201 Customer experience04030506CAI developmentImplementation journeyIndustry applicationsD.BotintroductionCopyright © 2020 Deloitte Development LLC. All rights reserved. 3What is Conversational AI ? 4Conversational AI (CAI) combines natural language processing , AI, and machine learning to understand and respond to free-form text or voice in an engaging and personalized manner. HealthcareVoice diagnosisHospital guidanceEducationSpeech teaching Speech assessmentFinance Call centerInteractive voice responseManufacturingIndustrial robotSmart phoneVoice assistant APPSmart homeIntelligent lightingIntelligent kitchenWearable deviceSmart watch Smart bandIntelligent vehicleSpeech navigation Voice control Typical scenariosTransform the ways we live and work Consumer application Business application 5 Generate quick responses to reduce waiting time and processing time Reduce operational costs such as labor cost and time cost Answer questions and process tasks in a uniform format Reduce human errors with more consistent services Use CAI to replace manual operationsenabling employees to focus on more creative tasks Interact with customerssmoothly and efficiently creating better customer experience More diverse interaction methods can be achieved, such as text, voice, gesture, etc. CAI can be applied toscenarios where the traditional interaction methods could be inconvenient, such as self-driving CAI will change ways how human and machine interact, optimize task-based working processes, and can tremendously reduce the time for information retrieval Being a productive personal assistant, CAI will augment human capabilities in the future of workEfficiencyStandardizationExperienceDiversityPersonalizationBenefits of CAI in organizations6CAI development7CAI growth in applicationsCAI plays a major role in the applications of Artificial Intelligence Facial recognitionCall-Center virtual customer agentsChatbotsFraud analysis on transactional dataProcess optimization Market/Consumer segmentationVirtual personal assistantsSmart robotics45%42%39%37%37%37%17%17% Speech-activated applications have been widely adopted in the field of AI, such as call-center virtual customer agents, chatbots, virtual personal assistants, and smart robotics. Source: 1. Market Guide: China AI Startups, Gartner, November 20192. Global AI Development Whitepaper, Deloitte, September 2019 CAI penetration in leading AI industriesLow High High Finance Education Healthcare Government Smart City Self-driving Communications Retail Market sizePenetration(Percentage of China respondents) CAI has a higher penetration rate in finance, education, government and healthcare among applications across industries. Leading use cases for CAI in AI deployment (Penetration: industry application degree; Market size: marketing opportunity)1 28 Extensive applications of CAI have emerged in consumer market. CAI brings value enhancement to various industries. Users hold more positive attitudes toward CAI applications. The development of chips and cloud computing shows the trend of integrating with AI. With the development of edge AI chips, CAI will find its way into mobile devices. Cloud computing enables enterprises and governments to offer more personalized and intelligent services and products. Deep learning has made great breakthroughs in speech recognition, natural language processing and speech synthesis. In the future, it is possible to realize barrier-free human-machine emotional communications.AlgorithmUser demandPolicyComputing powerMachine learning, deep learning and other technologies build solid foundation for CAI CAI is transforming the ways we live and work Development of chips and cloud technology has fueled the basic computing power to CAI1234Policy is a catalyst for the future of AI growth Policy has gone through three phases, from single products to data-driveninnovation platforms, from individual actions to national strategies, and most importantly, from AI technology development to the integration of AI and the real economy. Four drivers of CAI growth9Prior to 2011 when deep learning, big data and cloud computing were not integrated with speech recognition technology, the accuracy of speech recognition was 54.61%.In 2011, Microsoft introduced deep learning into speech recognition, which improved the accuracy of speech recognition to 81.55%. Before 2011, voice applications developed by Microsoft and Google were only based on basic grammar analysis and machine translation,mainly in information retrieval and extraction.After 2011, NLP text analysis was advanced towards deep understanding, making dialogue robots more practical and scene oriented.With the application of pre-training language model in NLP task, NLP technology started to focus on emotional text analysis and text reasoning.Before 2011, first-generation voice assistants were mainly used for navigating on PCs and information retrieval in a monotone voice. After artificial neural network had been used in speech synthetics technology, voice assistants started to imitate human intonation. In 2017, Tacotron 2, a speech synthesis system released by Google, was developed as close ashuman voice and became a benchmark system.In 2017, the accuracyof Microsoft's switchboard reached 94.9%, surpassing human for the first time.Deep learning outbreak periodDeep learning exploration period Traditional machine learning periodSpeech recognition Natural language processing Speech synthesisAccuracy Text analysis Emotion Algorithm102006 2007 2008 2010 2013 2015 2016 2017IaaSIn 2006, Amazon created AWS, providing a variety of cloud-based services on IaaSincluding storage and computation.Before 2007 when AI was in the early stage, CPU chips were sufficient to provide enough computing power.CPUPaaSSalesforce pioneered the enterprise PaaS market when it launched the Force platform in 2008. SaaSAfter 2010, SAP, Oracle and other traditional software companies began to launch cloud services, and enterprise SaaSdeveloped rapidly.GPUAfter 2013, GPU were widely used for AI. ASICIn 2015, Google first released the ASICchip tpu1.0, and the industry started to develop special chips for AI. FPGAcame out after Intel acquired Altera in 2015.FPGAIn 2017，Huawei Qilin 970 became the first mobile phone AI chip, introducing AI into mobile devices.In 2016, AWS officially launched its own AI product line, and cloud computing started to show the trend of integration with AI.The continuous chip evolution integrated with cloud computing Bringing AI to the device: Edge AI chips come into its own. Edge AI chips will find the way into consumer devices and enterprise markets, such as smartphones, smart speakers, wearables and robots, cameras, sensors, and other IoTdevices in general. Services become more intelligent: virtual computing platform greatly improves the data processing and reduces the cost of using data. Enter the era of customization: based on the extensive collection of information, user behavior and needs are accurately pinpointed. Chips CloudAI + Edge Computing AI + Cloud Computing powerThe increasing computing power releases the potential of AI algorithms