EntityBot is a digital assistant that knows users’ actual needs

The digital assistant EntityBot could help make computer usage easier by making suggestions based on the user’s actual needs, such as recommending relevant information.

This video is a part of the FCAI success stories series. In the video series, we explain why fundamental research in AI is needed, and how research results create solutions to the needs of people, society and companies.

Professor Giulio Jacucci from the University of Helsinki, in collaboration with Prof. Samuel Kaski’s group at Aalto University, leads a research group dedicated to the development of a digital assistant that recommends entities related to the user’s tasks.

These entities are information and apps such as documents, files, emails opened, and keywords. Machine learning will determine the necessary entities to be suggested for each of the user’s tasks.

The user can interact with the model to improve the estimate by selecting the entity on the User Interface.

“The digital assistants mostly used today are based on only limited data. For instance, Amazon will only recommend products to you. EntityBot has no limitations on recommendable data about services, devices and applications”, says Jacucci.

EntityBot is able to learn what we are doing and recognize the task that we are working on, as well as how, for instance, contact information, applications, and documents relate to it. Its recommendations are always based on what the user is currently doing.

In the model of the assistant, the necessary information is gathered during 14 days of monitoring computer usage. The data is collected by 24/7 digital activity monitoring of the test subjects, with a screen capture made every 2 seconds.

The data collection of course raises ethical questions about privacy, and Jacucci agrees that more research is needed about the potential risks of the collected data ending up in the wrong hands. The possible risks and mishandling of similar systems potentially developed in the future have to be identified, and the privacy of the subjects preserved.

“Google’s data collecting is to push adverts. Our new systems are different in that they are not handled by computing giants. The aim is to own our own models on what we do and use them for our own advantage”, says Jacucci.

The system has to be transparent to help the user understand why certain recommendations are issued.

The model might be of special use to the elderly. Currently, the research group is studying its usage within aging workers.

EntityBot will be published this year in ACM TOCHI and will be presented at the ACM CHI Conference 2022.

Jacucci, G., Dae, P., Vuong, T., Andolina, S., Klouche, K., Sjöberg, M., Ruotsalo, T.,  and Kaski. S., 2021, Entity Recommendation for Everyday Digital Tasks, to appear in ACM Transactions on Computer-Human Interaction.