Conversational AI Summit Conversational AI Summit schedule

Sprinkling In the Human Touch

Conversational AI has exponentially improved in its ability to bake in excellent user experiences since its inception. As the industry continues to quickly rise, what's the line of what’s too human? What’s not human enough? How do we navigate the world of Conversational AI when it comes to staying true to our human spirit, while delivering excellent user experience at the same time? We’ll tap into how to think about it, practically apply it, and dive into a real-world case of how GoDaddy strikes this sizzling balance.

Sonia leads the chatbot vision for GoDaddy. Prior to this role, Sonia designed the conversation for the first HIPAA-complaint voice app at health-insurance giant, Anthem, and started her own company in the space to help preserve people's life stories, HereAfter AI, developing, testing, and training "the world's first digital humans" as The Washington Post noted and The Wall Street Journal won an Emmy for its minidocumentary about it. Sonia was formerly a TV personality and columnist for Barron’s and holds an M.A. from Columbia University and a B.A. from UCLA.

Top 3 Most Complex Projects in the Current Environment

Even if AI provides “state of the art” solutions and research is gaining more ground, in real life Conversational AI doesn’t always have the impact expected by many. Florin and Zsolt Szekely will address the current challenges and will present a live demo of the latest and most complex developments. We at Bosch Service Solutions believe that the time for power point presentations, in which we create unrealistic expectations, need to be put behind us, and the focus should be on projects that can have immediate impact and the capacity to engage users, giving the same importance also to low-resourced languages.

This year, Florin together with his team started the implementation of complex Conversational AI frameworks (available for all languages, channels, and use-cases) designed to improve the AI based solutions within Bosch SO. The target is to continue the deployment of projects in each of the five Conversational AI pillars: Voicebots, Chatbots, Translation AI, Text/Email AI and Agent Assist AI. Prior to this role, Florin was involved in research, start-ups and other summits as keynote speaker - all these experiences giving him a 360 degree perspective on trends and challenges.

Creating Conversational Assistants: Challenges and Solutions

Conversational assistants are all around us – as smart speakers in our homes, on our smartphones, in our cars. The Conversational AI market is expanding dramatically and major technology companies are producing ever larger language models and datasets to support the development of conversational assistants. However, the use of these assistants is currently mainly restricted to one shot conversations involving commands or questions, or in some cases to fulfilling well-defined tasks. More open conversations are still within the realm of research, yet these systems will be needed for new areas of application such as conversational commerce, healthcare and mental health support, and the promotion of active and healthy ageing in older people. Traditional approaches based on best-practice guidelines are compared with new neural dialogue based approaches. The presentation concludes with a brief look at some recent projects in which I have been involved and the different ways in which they are addressing the challenges faced by designers and developers of conversational assistants.

Michael McTear is an Emeritus Professor at Ulster University with a special interest in spoken language technologies. He has been researching in the field of spoken dialogue systems for more than 20 years and is the author of several books, including Spoken Dialogue Technology: Toward The Conversational User Interface (Springer, 2004), Spoken Dialogue Systems (Morgan and Claypool, 2010), with Kristiina Jokinen, The Conversational Interface: Talking to Smart Devices (Springer, 2016), with Zoraida Callejas and David Griol, and Conversational AI ( Morgan & Claypool 2020). Michael has delivered keynote addresses and tutorials at many academic conferences and workshops, including SpeechTEK, Conversational Interaction, ProjectVoice, REWORK AI Assistant Summit, and the European Chatbot Conferences. Currently Michael is involved in several research and development projects investigating the use of conversational agents in socially relevant projects such as mental health monitoring, and home monitoring of older persons.

How to Successfully Scale Chatbot Development Across a Global Enterprise

The [email protected] program aims to empower all Bosch Rexroth countries to ramp up their own chatbots. The goal is to rapidly scale up the most impactful bots and roll out positive experiences globally. This mission comes with challenges such as decentral requirements, global qualification program, demand management, central development, central reporting and KPI management, and also the development of a concierge bot concept. The presentation will share insights and learnings from the parallel development of multiple chatbots, all based on Cognigy.AI as an enterprise solution.

Özge is a Data Analyst in Bosch Rexroth. She studied Industrial Engineering and IT and has a background in Online Marketing for 4 years. Her area of expertise is to create a data management strategy for the marketing, customer sales and communication behaviour. She collects, organizes, extrapolates, interprets, and builds Power BI-based Dashboards for Global Sales Management to make important sales and customer service-oriented decisions. Her team is also collecting chat data in order to implement machine learning and use that valuable data in order to improve their chatbots based on these agent-customer conversations. Additionally, she is the Linkedin Corporate Influencer of Bosch Rexroth regarding AIoT communication

How to Successfully Scale Chatbot Development Across a Global Enterprise

The [email protected] program aims to empower all Bosch Rexroth countries to ramp up their own chatbots. The goal is to rapidly scale up the most impactful bots and roll out positive experiences globally. This mission comes with challenges such as decentral requirements, global qualification program, demand management, central development, central reporting and KPI management, and also the development of a concierge bot concept. The presentation will share insights and learnings from the parallel development of multiple chatbots, all based on Cognigy.AI as an enterprise solution.

Oliver has 35 years of experience in software development. Before joining Bosch Rexroth in 2007, he worked for various software modeling solution providers. Early on, he specialized in the development and consulting of generative software development approaches based on abstract software architecture models.

Most recently, he developed a highly configurable online chat system based on Genesys that can be adapted to the different requirements of the many Rexroth countries. Building on this, he is the responsible Program Manager for [email protected], a global program that enables all Rexroth countries to develop their own chatbots. In addition, he has experience with machine learning with a focus on natural language understanding.

Conversational AI: Mind The Knowledge Gap

Using NLP, NLG, Knowledge Graphs and Deep learning technologies makes it possible to build Conversational AI systems. Gaps in knowledge used by such a system are a critical challenge in achieving high accuracy and realistic Conversational AI. The knowledge gaps exist due to the lack of knowledge about the subject of conversation, or the knowledge the AI systems are built on becomes outdated. In this talk, Prof Thakker will present his research as part of an Innovate UK-funded project for building a Conversational AI system to support UK Immigration casework. Immigration law advice is a specialist domain, and underlying subject knowledge frequently changes due to changes in the immigration rules. Such knowledge dynamism is a common characteristic in other specialist domains such as other areas of law advice, financial advice, and medical advice. Prof Thakker will present work as part of this project in building NLP, NLG and Knowledge graph pipelines for conversational AI and explore techniques to overcome knowledge gap challenges. He will share lessons learnt with broader applicability in other domains.

Professor Dhaval Thakker is a Full Professor in Artificial Intelligence and the Internet of Things (IoT) at the University of Hull, UK. Dhaval has over fifteen years of working experience in the European Union(EU) and industrial projects researching and delivering innovative solutions. He has been a PI on projects with over €1m funding from Innovate UK, and European Commission. This includes the SCORE (Smart Cities and Open Data REuse) project with nine European cities and three other Universities and an Innovate UK-funded project on “Artificial Intelligence and UK Immigration law”, where his research focuses on Conversational AI.

CAI Meet Unsupervised Learning: Tools and Best Practices for Optimising NLU Performance for Improved CX

This session examines the demands on AIOps teams to maintain accuracy and NLU performance against the overwhelming volume and variation of user queries, the difficulty of translating this into reliable training data (without breaking things) and doing so in a timely manner that addresses both the demands of the business (for new scope and functionality) and the desire to improve overall satisfaction of conversational self-service experiences.

Leveraging a real-world example, we will present solutions for: • identification and treatment of confusion in NLU design • measurement of the accuracy of your model over time • minimising “learning delay” • reduction of learning workloads for humans • application of changesets into new NLU models • regression protection

In this live demonstration, prepared for both business and technical teams, we will evaluate the NLU design and benchmark the bot accuracy, produce a change set for identification and treatment of areas of confusion, automatically push the change set into the NLU model and evaluate the regression impact to confirm that the changes can be made safely.

Chris has a 20+ year history of leading delivery of software and business change projects across a variety of industries and functions, including government, insurance, airline, utilities and contact centre. His experience includes 6+ years of delivering advanced Conversational AI and chatbot implementations. From running projects for small agile teams through to large scale enterprise transformation initiatives, Chris combines a background in enterprise architecture and engineering, with a deep understanding of business and technical needs, offering a unique ability to bridge the gap between senior stakeholders, practical operational needs, technical constraints and delivery.

CAI Meet Unsupervised Learning: Tools and Best Practices for Optimising NLU Performance for Improved CX

This session examines the demands on AIOps teams to maintain accuracy and NLU performance against the overwhelming volume and variation of user queries, the difficulty of translating this into reliable training data (without breaking things) and doing so in a timely manner that addresses both the demands of the business (for new scope and functionality) and the desire to improve overall satisfaction of conversational self-service experiences.

Leveraging a real-world example, we will present solutions for: • identification and treatment of confusion in NLU design • measurement of the accuracy of your model over time • minimising “learning delay” • reduction of learning workloads for humans • application of changesets into new NLU models • regression protection

In this live demonstration, prepared for both business and technical teams, we will evaluate the NLU design and benchmark the bot accuracy, produce a change set for identification and treatment of areas of confusion, automatically push the change set into the NLU model and evaluate the regression impact to confirm that the changes can be made safely.

Tom Port is the VP Sales at ServisBOT, the Conversational AI Platform helps businesses automate interactions across customer and employee journeys using AI. Prior to ServisBOT, Tom was a member of the founding team at AutoEntry a fintech start up that was successfully sold to the Sage Group in 2019. Tom's passions lie in deeply understanding the nuances of pain felt by large enterprises and how we can use technology to help solve them.

Kane Simms is one of the conversational AI industry's primary thought leaders and host of the renowned conversational AI and CX podcast, VUX World. Kane's been published in the likes of Harvard Business Review and voted in the top 10 voice AI influencers (Voicebot, SoundHound). Kane is a speaker, analyst and consultant who helps business leaders and teams realise the future of CX excellence using emerging AI and NLP technologies.

Tripta Kumari’s twenty-year UX career has involved designing for voice, mobile, web, and multi-modal platforms – creating experiences which feel natural and enjoyable. Nuance-trained in conversational UI and natural language interaction, she’s worked with companies including PlayStation, Intel, Vodafone, BT, Skype, Sony Ericsson, John Lewis, TomTom, Emirates and ING. She was one of the co-authors of 97 Things Every UX Practitioner Should Know (O’Reilly, 2021).

Jakob Rudbeck Mølgaard brings years of project management experience in the digital transformation and automation space from across a range of industries. As Head of Customer Onboarding at boost.ai, he is responsible for ensuring that new customers to the company’s conversational AI platform are primed to build and deploy virtual agents that reach their full potential and deliver exceptional customer experiences. During his time at boost.ai, Jakob has previously been responsible for successfully delivering virtual agent implementations across numerous markets and business sectors.

Sonia is the Data Science Engineering Lead for Natural Language Engineering at Lloyds Banking Group. Working at the intersection between machine learning, linguistics and software development, Sonia strives to unlock the value of unstructured data at the Enterprise level. Prior to starting at Lloyds, Sonia worked as a consultant Data Scientist for BJSS, working on projects spanning workforce planning, hospital wait time forecasting, Ethical AI, data science life cycle, and biomedical knowledge graphs.

Superhuman Conversational AI

AI has reached superhuman levels in various areas such as playing complex strategic and video games, calculating protein folding, and visual recognition. Are we close to superhuman levels in conversational AI as well? In his talk, Yariv will address this question, sharing some of the recent developments from Google Cloud AI, Google Brain Research, Deepmind, and Duplex - across speech recognition and generation, and natural language understanding.

For the past 10 years, Yariv has been leading products and product teams at Google Zurich. In his current role, he is leading the Zurich product team working on the Google Assistant. Prior to this role, Yariv worked on a wide range of products, including proprietary Google infrastructure, privacy and security, products designed for the Emerging Markets, and even the notorious non skippable ads on YouTube. Before joining Google, Yariv was an engineering manager in various Israeli start-ups and companies.

From Conversational Search to Real-World Tasks and Beyond: The Future of Multi-Modal Assistance

Conversational assistants are rapidly evolving with significant potential, but with fundamental challenges that need to be overcome to realize their potential. This talk presents current challenges and future directions for knowledge grounded conversational agents. It discusses current progress from experience developing systems and benchmarks in conversational search, with leading media organizations, and the development of the winning agent in the Alexa TaskBot challenge. It presents fundamental challenges in conversational knowledge-grounded language understanding and discusses issues on controllability of neural generative models. It highlights future directions including multi-modal understanding, new forms of mixed-initiative and rich interactions, and increasingly complex tasks.

Dr. Jeff Dalton is an Associate Professor (Senior Lecturer) in the School of Computing Science at the University of Glasgow where he leads the Glasgow Representation and Information Learning Lab (GRILL) (https://grilllab.ai). His research focuses on text understanding and conversational information seeking. He completed his Ph.D. at the University of Massachusetts Amherst in the Center for Intelligent Information Retrieval. Later in Google Research, he worked on Information Extraction as part of the Knowledge Discovery Team (Knowledge Vault) and on language understanding in the Assistant Response Ranking team. He is the lead organizer for the TREC Conversational Assistance Track (CAsT) (http://treccast.ai) benchmark, the longest running conversational search benchmark. He is the recipient of a prestigious UKRI Turing AI Acceleration Fellowship on Neural Conversational Assistants and received research awards from Google, Amazon, and Bloomberg. He is the faculty advisor for the UoG GRILLBot team that won the 2021/2022 Alexa Prize Taskbot Competition, the first grant challenge to include multi-modal aspects. He holds multiple patents in retrieval, information extraction, and question answering.

No Language Left Behind: Scaling a Machine Translation Pipeline to Low Resource Languages

Progress in machine translation has coalesced around a small subset of languages, leaving behind the vast majority of mostly low-resource languages. In No Language Left Behind, we took on the challenge of breaking the 200 language barrier, scaling on the engineering and modeling side, but also on evaluation. Our model achieves an improvement of 44% BLEU relative to the previous state-of-the-art, laying groundwork towards universal translation. This presentation will first give an overview of our approach. It will then dive deep into our automated data mining system, which was key in identifying precious training data for low-resource languages.

Alexandre Mourachko is a Research Engineering Manager at Meta AI. He currently leads research and engineering efforts for the No Language Left Behind project, which aims at breaking language barriers across the world by scaling machine translation to more low resource languages. Within this highly multidisciplinary endeavor, his research team focuses on bringing sentence representation learning to the next level, scaling data mining pipelines to web corpora, and mitigating toxicity and bias in translation model outputs. Before joining Meta, Alexandre was leading Machine Learning research teams fighting toxic behaviors in multilingual online gaming communities.

AI for Spoken Communication

Do you want to know how chat bots like Amazon Alexa works? Why does Alexa understand what you say and how does she know what to answer? In this event we will discuss AI building blocks that Alexa has and what kind of improvement does AI provide in contrast to rule-based solutions.

Asmita is a Software Development Engineer at Amazon Alexa, where she works on developing and productionizing natural language processing and speech models. Deeply drawn to the idea of multi-disciplinary collaboration, Asmita also has prior experience in applying machine learning in diverse domains such as healthcare, genomics and music recommendation. She has previously worked at the National University of Singapore, INRIA Lille-Nord Europe and Academia Sinica- Taiwan, where she did research, developed software and machine learning methods for these various applications. Asmita holds a Masters degree in Computer Science from the University of Oxford.

Transfer Learning for Conversational AI: a Multidimensional Approach

The emergence of deep learning and recent advances in using large pre-trained language models have given conversational AI a big impulse. For a range of applications, models can be fine-tuned with only a limited amount of application specific data. However, for scalable development of spoken dialogue interfaces with true conversational intelligence, ultimately this approach may very well prove to be insufficient. In this presentation, I will argue for a multi-dimensional dialogue modelling approach, and discuss experiments, in which its transfer learning potential is demonstrated for the case of adapting reinforcement learning-based dialogue policies to a new application.

Simon Keizer is a research scientist specialized in conversational AI. He has a BSc/MSc in Applied Mathematics and a PhD in Computer Science, both from the University of Twente (The Netherlands). Simon has been active in spoken dialogue systems research for more than 18 years, focusing on topics such as statistical dialogue management, reinforcement learning, and user simulation. He has been involved in several national and international research projects at Tilburg University, University of Cambridge, Heriot-Watt University Edinburgh and the Vrije Universiteit Brussel, and is currently a research engineer in the Cambridge Research Lab, Toshiba Europe Limited.