AI Assistant Summit schedule

Democratising Conversational AI

Personal assistants such as Alexa and Siri are everywhere, but have trouble handling any tasks more complex than setting alarms or playing songs on Spotify. PolyAI is a new London-based startup with a next generation platform for building voice-based agents. We use machine learning to handle complex tasks across different application domains, in a wide array of world languages. In this talk, I will present our Wizard-of-Oz data collection framework, which allows us to collect high-quality datasets and train our models at a fraction of the cost compared to traditional data labelling approaches.

Nikola Mrksic is the CEO and Co-Founder of PolyAI, a London-based startup developing the next generation machine learning platform for building conversational user interfaces. Our technology enables the development of machine learning powered chatbots or voice-based agents which perform tasks across many applications, in a wide array of world languages. Nikola completed his PhD at the University of Cambridge, working with Professor Steve Young. He also worked as a machine learning researcher with the Apple Siri team in Cambridge. Before that, he was the first technical hire at VocalIQ, a Cambridge-based dialogue systems startup acquired by Apple in 2015.

Opinion Mining Using Heterogeneous Online Data

Understanding public opinion is important in many applications, such as improving company's product or service, marketing research, recommendation systems, decision and policy making and even predicting results of elections. Social media is a very powerful tool to transfer information and express emotions for users and a rich source of data that enables researchers to mine public opinion. In this talk I will describe different lines of work within our group, including rumour stance and veracity classification classification, predicting well-being based on heterogeneous user generated data and target-dependent sentiment recognition. False information circulating on social media presents many risks as social media is used as a source of news by many users. Detecting rumourous content is important to prevent the spread of false information which can affect important decisions and stock markets. Rumour stance classification is considered to be an important step towards rumour verification as claims that attract a lot of scepticism among users are more likely to be proven false later. Therefore performing stance classification well is expected to be useful in debunking false rumours. In our work we classify a set of Twitter posts from conversations discussing rumours as either supporting, denying, questioning or commenting on the underlying rumours. We propose a LSTM-based sequential model that achieves state-of-the-art results on this task through modelling the conversational structure of tweets. The task of automatically assessing well-being using smartphones and online social media is becoming of crucial importance, as an attempt to help individuals self-monitor their mental health state. In the current work, a multiple kernel learning approach is proposed as a mental health predictor, trained on heterogeneous (text and smartphone) user-generated data. The results reveal the efficiency of the proposed model and sequential approaches for time series modelling (i.e., LSTMs) are proposed for future work. Opinion mining is usually achieved by determining the overall sentiment expressed. However, inferring the sentiment towards specific targets is limited by such an approach since a Social Media posts may contain different types of sentiment expressed towards each of the targets mentioned. Our work on target-specific sentiment recognition goes beyond tweet-level or single-target approaches, and proposes a multi-target-specific sentiment classification model, which explores the context around a target as well as syntactic dependencies involving the target.

I am a Computer Science PhD student at the University of Warwick supervised by Dr. Maria Liakata and Prof. Rob Procter. Currently I am based at the Alan Turing Institute in London. My background is Applied mathematics and Complexity science. I work in the area of Natural Language Processing. My research is focused on Rumour Stance and Veracity Classification in Twitter conversations. I am studying the benefits of utilising the conversation structure in supervised learning models.

What’s So Conversational About Conversational Interfaces?

Conversational interfaces have emerged as a new way to interact with AI assistants on smartphones, smart speakers, and other devices. However, to what extent are these systems truly conversational? And if not, do they really need to be anyway? In this talk I will outline the capabilities of current conversational interfaces and assess the extent to which they model important conversational phenomena such as follow-up questions, changes of topic, out-of-scope utterances, and mixed-initiative dialogue. Generally conversational interfaces are categorised as either task-oriented, for example, to assist with making a hotel booking, or as entertainment-based, for example, to engage in chit-chat. However, another category of conversational interface is emerging in which the assistant takes part in a conversation in order to solve a problem, for example, to provide support and advice in a healthcare context. I will discuss the technical and other challenges we face in developing this new type of conversational interface.

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, and The Conversational Interface: Talking to Smart Devices (Springer, 2016), with Zoraida Callejas and David Griol. Michael has delivered keynote addresses and tutorials at many academic conferences and workshops, and is a regular presenter at the annual SpeechTEK conference. 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, home monitoring of the elderly, and the management of obesity.

Natural UI

AI is powering the next paradigm shift in Human-Machine Interaction - the Natural User Interface. In his talk, Yariv Adan will provide an insider perspective on this paradigm shift, and will demo and explain the technology driving it, using the Google Assistant.

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.

Democratising Conversational AI

Personal assistants such as Alexa and Siri are everywhere, but have trouble handling any tasks more complex than setting alarms or playing songs on Spotify. PolyAI is a new London-based startup with a next generation platform for building voice-based agents. We use machine learning to handle complex tasks across different application domains, in a wide array of world languages. In this talk, I will present our Wizard-of-Oz data collection framework, which allows us to collect high-quality datasets and train our models at a fraction of the cost compared to traditional data labelling approaches.

Nikola Mrksic is the CEO and Co-Founder of PolyAI, a London-based startup developing the next generation machine learning platform for building conversational user interfaces. Our technology enables the development of machine learning powered chatbots or voice-based agents which perform tasks across many applications, in a wide array of world languages. Nikola completed his PhD at the University of Cambridge, working with Professor Steve Young. He also worked as a machine learning researcher with the Apple Siri team in Cambridge. Before that, he was the first technical hire at VocalIQ, a Cambridge-based dialogue systems startup acquired by Apple in 2015.

Can We End the Era of Unreadable Privacy Policies with AI?

With the length and complexity of today's privacy policies, it's no surprise that they are rarely read. So what if AI can read these policies on our behalf, simplify them, and let us know what to do about them? That is why we created Polisis, the first advanced AI platform for analyzing privacy policies, and PriBot, the first AI assistant for automatically answering questions about these policies. In this talk, I'll be making the case for using deep learning to alleviate the huge cost for understanding these complex documents in the case of both businesses and individuals. I'll also outline our vision towards turning these documents from a costly requirement into a competitive business opportunity.

Hamza Harkous is a Postdoctoral Researcher at École Polytechnique Fédérale de Lausanne (EPFL), Switzerland. He is the creator of Polisis, the first and most advanced AI platform for automated analysis of privacy policies. He also created PriBot, the first chatbot for privacy policies. He obtained his Ph.D., from EPFL in 2017, where he worked on data-driven usable privacy. His general interests are at the intersection of deep learning, privacy and human computer interaction. He is currently working on taking this research to the next deployment stage. This will provide the opportunity for businesses to automate the full workflow of privacy practices' assessment.

Empathic Design for Conversational Interfaces

We are in an age of blending horizons and blurring devices brought alive by ambient assistive companions. Everything is talking, everything has a brain...and we respond to these "presences" with many of the same feelings and reactions as if they were sentient beings. It's not our fault. It's 150,000 years of human hardwiring telling us that a speaking thing is not just an “it,” it’s alive and aware and connecting with us. With “ears” and “eyes” sensing, it becomes even more emotionally compelling and/or complex.

How might we design for Emotional Intelligence to make people feel that these experiences are useful, familiar, comfortable, positive, supportive and trust-worthy.

Marc is a Senior User Experience Designer at Google, focusing on multimodal (voice and touch) conversational experiences for the Google Assistant across a range of surfaces from mobile to wearables.

Creating Genuinely Interesting Conversational Partners

Conversational agents and digital assistants are increasingly becoming part of our lives. However, current agents are usually dry data delivery mechanisms bereft of any of the charm or wit that we prize within human interlocutors. What does it take to create conversational agents that can be genuinely interesting conversational partners? I suggest that the function of human conversation is not to deliver useful information, but to display mind-reading and perspective-taking abilities–to demonstrate ourselves to be astute socio-political actors. This can be achieved through showing we know another person's goals and desires, and by empathising and understanding. It can also be done through the creative demonstration of an ability to combine concepts in the minds of other that they had not yet noticed were related–using analogy, humour, and story-telling. Therefore, to create engaging conversational agents we must focus on the development of these empathic and creative abilities. We need to ensure that agent models can develop conversational common ground, emotional connection and empathy, and be creative and humorous. These are not simple tasks but they are necessary if digital agents are to become truly engaging conversational partners capable of developing longer term relationships with humans that humans seek to sustain.

Gary McKeown is a senior lecturer and social psychologist in the School of Psychology, Queens University Belfast. He has a primary research and theoretical interest in human communication and social interaction. His research profile is interdisciplinary with both theoretical, experimental and methodological papers in psychology journals and also many within the domains of social signal processing and affective computing. Often this research involves understanding natural human communication to inform the development of embodied conversational agents. Building on a long history of emotion and affective computing research in the School of Psychology – including the HUMAINE network, he had important roles in the SEMAINE, ILHAIRE project and within the SSPNet network. He has produced a theoretical account of the evolution of human communication with a particular emphasis on mind-reading and perspective-taking known as the Analogical Peacock Hypothesis; this has led to research work in understanding laughter, humour and storytelling. His research interests have led to many collaborations with non-academic partners, particularly commercial and industrial partnerships for the transfer of knowledge and techniques to commercial settings. From a methodological perspective, the fields of social signal processing and affective computing address challenging data collection issues from a behavioural science stance–often using dynamic scenarios with multiple streams of synchronised information. This requires new ways of thinking about data and novel statistical approaches to develop analyses that can usefully address social science questions. Dr McKeown has been actively involved in the creation of both new data gathering techniques and the statistical approaches required to address them.

Evolution of AI & Machine Learning in Customer Experience - Beyond Interfaces

The evolution of Machine learning is directly proportional to the customer believing in the non-existence of a machine in between. With deep learning and neural network implementation, the traditional ML models are becoming dated. Often when a new technology has its breakthrough; it’s impact is only felt in hindsight. But it's different with Artificial Intelligence and Machine Learning. Let’s talk about how ML powered Chatbots add value to the customer experience. From demand generation to fulfilment, all behind a seamless customer experience.

Adi is a Product Manager with over 11 years of experience across Fashion, E-commerce and Technology Industries. He is currently leading Product Management and Innovations for Vodafone UK in the space of Artificial Intelligence and Machine Learning. He is passionate about solving problems through design thinking, big data and automation. An avid reader of consumer technology and applied mathematics, he often spends his time thinking about how to merge the two to make consumer experiences worth their while. Adi holds an MBA degree from Lancaster University Management School.

Building an AI friend

Chatbots should have empathy. They need to have the ability to identify the user’s emotions from the conversation, to detect how emotions evolve over time, and to understand the user’s emotional needs. Artem will talk about how chatbots can simulate emotional connection with users using different modalities, including text, speech, and vision, and how we can use deep learning to build the best friend for a human.

Artem Rodichev is a Head of AI at Luka, a company that is best known for its recent app called Replika - an AI friend for emotional growth and awareness. Artem earned M.Sc. degree in Applied Mathematics and Computer Science from Moscow State University. Before Luka, he worked at top Russian IT companies such as Yandex, Mail.ru, and ABBYY developing NLP-driven products. Artem's area of research interest includes Conversational AI and Deep Learning techniques for modeling human conversation.

Zafeirios is a mathematician/computer scientist with a passion for neuroscience, biologically inspired A.I. and cognitive robotics. Currently, he works as a researcher in the Department of Computing of Imperial College London investigating the neural correlates of human time perception. Additionally, he works at Emotech LTD, where he is building the A.I. system of the personal robot Olly. He received his Ph.D. in computational neuroscience from Imperial College, where he focused on the relation between brain oscillations, our ability to select between multiple prospective actions and motor symptoms of Parkinson's disease.

Operator/AI Assistant Interaction: Challenges from a User Perspective

So called "cognitive assistants" are flourishing, but how to evaluate the benefit/cost ratio of introducing that technology into aeronautical operations? How far can cognitive assistant help operators such as pilots or mechanics at work? What impact can we expect on human-system performance? Our hypothesis is that cognitive assistants will be useful and accepted if and only if they implement a certain level of cognitive interaction with the human operator. First we question what kind of human cognitive functions can actually be supported to by a digital assistant to increase its performance at work. Then we propose a set of basic principles of cognitive interaction inspired by human sciences, e.g. cognitive philosophy and sciences of language. That definition is agnostic regarding technology. Our intention is to shape the evaluation and the specification of future cognitive assistants from a user perspective. Our purpose in this summit is to spark a discussion on the ambition and the feasibility of "natural" cognitive interaction between human operators and cognitive assistants.

Denys Bernard is an aeronautical engineer and doctor in Computer Sciences. He has relentlessly worked in research and development activities for decision support systems - both on board and on ground - particularly for aircraft maintenance. His main domains of interest are: diagnostic; planning; knowledge extraction and representation; decision making. He is now focused on the application of cognitive technologies as the foundation of virtual assistants for mechanics and cockpit crew.

Developing Behavioural Models for Intelligent User Interaction using Mobile and IoT Sensor Data

In this talk I will present our current work on extracting information about physical and emotional context through sensor data collected by means of mobile phones, wearables and other Internet of Things devices in order to deliver the right information at the right time to users. In particular, I will discuss how it is possible to extract behavioural signals from sensor data collected by means of off-the-shelf devices in order to drive the delivery of relevant information to users. I will then discuss the current challenges (and opportunities) in this emerging research area.

Mirco Musolesi is a Reader (equivalent to Associate Professor in the North-American academic system) in Data Science at University College London and a Turing Fellow at the Alan Turing Institute, the UK National Institute for Data Science and Artificial Intelligence. At UCL he leads the Intelligent Social Systems Lab. He held research and teaching positions at Dartmouth, Cambridge, St Andrews and Birmingham. He is a computer scientist with a strong interest in sensing, modelling, understanding and predicting human behaviour and social dynamics in space and time, at different scales, using the "digital traces" we generate daily in our online and offline lives. He is interested in developing mathematical and computational models as well as implementing real-world systems based on them. This work has applications in a variety of domains, such as intelligent systems design, ubiquitous computing, and digital health & wellbeing.

Kriti Sharma, is an artificial intelligence tech entrepreneur, currently building AI that solves real problems facing the world today, from access to emergency services in developing countries to an AI powered financial assistant that helps entrepreneurs around the world. Kriti deeply cares about the ethics of AI and often writes about AI and bot design principles.

Dr Janet Bastiman is Chief Science Officer at Storystream where she heads up the AI division. She has extensive experience in leading teams and building AI systems. With over 10 years experience at C-level, she has taken abstract concepts to innovative scalable products at multiple companies. A STEM polymath, Janet brings multidisciplinary ideas to the departments she leads, forming cutting edge research teams with an eye on delivery that is rarely seen. She has a bachelors and masters degree in Molecular and Cellular Biochemistry from Oxford, and her PhD was in the field of Computational Neuroscience.

Janet is a co-founder of the Tech Women London Meetup group and regularly speaks and writes on various technical subjects. She blogs on maths and technology at http://janjanjan.uk.

Kriti Sharma, is an artificial intelligence tech entrepreneur, currently building AI that solves real problems facing the world today, from access to emergency services in developing countries to an AI powered financial assistant that helps entrepreneurs around the world. Kriti deeply cares about the ethics of AI and often writes about AI and bot design principles.

The Brave New World of Data? - Ethical Questions and Case Examples

Automated information processing has revolutionised business, government and led to changes in society, for example in the way we communicate. Following the "data is the new oil" mantra, many companies offer free services in return for getting access to their users' personal data. Professional services and government departments are working on a digital agenda. In this talk, Jochen will consider the moral side of the new world of data. What ethical principles ought to underpin building systems, doing business, governing citizens from a data privacy perspective? What kind of ethical issues have already happened and will likely happen soon? For example, Eventually, automated 'black box' systems could be making decisions without transparency or recourse in the case of error. I conclude with some suggestions for remedy.