EIT Digital is launching a new innovation activity to develop a virtual assistant incorporating Artificial Intelligence (AI) and Natural Language Processing (NLP) technologies. Dubbed as the "Multi-channel & artificial intelligence based virtual assistant" it is hoped that the technology could relieve office workers from simple, low-value email correspondence freeing them up for more valuable activity.
According to a 2015 study carried out by Adobe, office workers spent more than six hours per day reading and writing emails. This time could be reduced and specialist staff freed-up for more value-adding tasks by allowing virtual assistant solutions to handle many of the simpler correspondence like appointment making.
The "Multi-channel & artificial intelligence based virtual assistant" innovation activity will combine the know-how and existing solutions of several EIT Digital partners. The solution will draw on the cloud-hosted end-to-end hybrid virtual assistant called "Julie", developed by the French EIT Digital scaleup, Julie Desk, in 2015. The main deliverable of the innovation activity will be an improved and more scalable version of Julie, and the setting up of new service startups to launch the new product on the market.
Christophe Gravier, the EIT Digital Innovation Activity leader and Associate Professor with Habilitation at Télécom Saint-Étienne, said:
"Our objective is to develop a virtual assistant able to understand requests in natural language received via multiple channels, not just by email. There is huge potential from an economic and societal point of view: if a virtual assistant could manage third party services, like travel reservations, restaurant and hotels bookings, and so on, it could help clients save time and increase productivity."
"As the volume of this type of low value email grows, so does the rationale for reducing the need for human intervention. This can be achieved through state-of-the-art Natural Language Processing technologies, like named entity recognition, neural networks and word embedding. We see Julie as being able to make a real difference to everyday email loads, freeing up workers for more valuable and impactful tasks."
Jason McDonald, Director at String Can Interactive and Julie Desk user, confirms:
"Julie is a very unique tool! She saves me hundreds of emails a week, not only does she save me time, but she also makes me so much more efficient."
To enable larger user volumes, the new solution will need to allow inputs from several channels and also serve its customers with several output channels. To enable this the activity will draw on the work of Hungarian E-Group, specialists in multi-channel user interfaces, who will provide speech or chat-to-email and email-to-speech or chat interfaces to the solution.
The following EIT Digital partners are participating to work of the "Multi-channel & artificial intelligence based virtual assistant" innovation activity with the following responsibility areas: Julie Desk is working together with the French Télécom Saint-Étienne on the email classification. Julie Desk also collaborates with the Italian Fondazione Bruno Kessler, an expert in human language technologies, focusing in the innovation activity on the task of named entity recognition in emails by using the TextPro software, and the Hungarian E-Group, specialized in multi-channel user interface.
Background
The "Multi-channel & artificial intelligence based virtual assistant" innovation activity is one of the 13 innovation activities of the Digital Infrastructure action line of EIT Digital for 2017. The Digital Infrastructure action line focuses on enabling digital transformation by providing secure, robust, responsive, and intelligent communications and computation facilities for the markets. The action line targets in networking the mobile broadband infrastructure, network softwarisation, and the Internet of Things (IoT); in computing: cloud computing, Big Data, and Artificial Intelligence; in security: privacy, cyber security, and digital ID management.
*About the various Natural Language Processing technologies
Named entity recognition consists of the recognition of e.g. names of persons, organisations, locations, time expressions, quantities, monetary values, or percentages in textual communication, like emails. Neural networks is computational approach based loosely modelling the way a biological brain solves problems. Word embedding is a language modelling and feature learning technique in the Natural Language Processing field.
