Preserving and sharing knowledge through AI-powered digital avatars

An interactive digital avatar that can react and converse just like the person it is based on – sharing their knowledge and expertise, potentially even after death – may seem like science fiction, but an innovative technology platform developed by a Toronto Metropolitan University (TMU) researcher is enabling users to create digital versions of themselves.  

Led by The Creative School professor Hossein Rahnama, Augmented Eternity is currently in beta testing with a select group of about 25 people. The AI-powered application is trained to emulate an individual through their data. To ensure privacy, users choose what to allow access to: their social media, phone call patterns, emails and more, providing their data and knowledge to an algorithm that can create interactive chatbots and digital avatars, or what the industry calls “deepfakes.” Individuals can have different profiles, curating the information available to select audiences. Someone could have a version that offers professional expertise to clients and a second one that interacts with friends and family. Professor Rahnama offers the example of a lawyer who creates an avatar and then can provide clients the opportunity to access the avatar to get answers to routine legal questions. Others may want to capture their knowledge and personalities to leave an interactive legacy for their heirs or associates. 

“It has been in science fiction literature for the past hundred years. But something has changed, and that’s the amount of computing power we have today and the relatively cheap storage that we have for data using cloud computing, and very specific approaches in AI,” said professor Rahnama, who is also a visiting professor at MIT Media Lab. 

The technology is performing well in the beta test, but before it hits the market, professor Rahnama wants to address ethical issues, such as post-death control over the avatars, avoiding misrepresentation and being able to audit, in a transparent way, why and how an avatar might make a certain comment or suggestion. 

Separating fiction from the real innovation of Augmented Eternity and his other AI research is a focus on privacy, ethics and data portability and governance, allowing the individual to control and own their information rather than large companies. “Privacy should go hand-in-hand with AI. So a lot of our work at the university is about enabling data to be shared in a privacy-preserved manner.” 

An Augmented Eternity-powered avatar is not meant to be a decision-making tool, professor Rahnama explains, but instead acts as a support system that can offer insights from someone else’s perspective. One challenge for creating these avatars is the sheer volume of data needed to make an accurate facsimile of a person. Professor Rahnama says his team is focusing on younger generations who tend to already have large digital footprints. In cases where mining existing extensive data is not an option, interviews and other approaches can be used to supplement what is available to support avatar development. 

Professor Rahnama projects the beta test will conclude in fall 2023. Augmented Eternity has been featured in the recent German documentary My Avatar and Me (external link, opens in new window)  and in the 2021 Canadian-made documentary A.rtificial I.mmortality (external link, opens in new window) . 

The researcher and entrepreneur, who successfully launched the company Flybits based on his PhD work, hopes to follow a similar path with Augmented Eternity. He sees this kind of knowledge sharing, where individuals own and control their data, as the future of a more democratized, decentralized internet. “I truly think that services like this will define the new internet, allowing us to connect to each other better, more effectively with less bias,” said professor Rahnama. 

It has been in science fiction literature for the past hundred years. But something has changed, and that’s the amount of computing power we have today.

Learn more about the Augmented Reality project (external link, opens in new window) . 

For more information, see professor Rahnama’s paper “ (PDF file) A Neural-Symbolic Approach for User Mental Modeling: A Step Towards Building Exchangeable Identities (external link, opens in new window) ” with co-authors Marjan Alirezaie and Alex Pentland from the Proceedings of the Association for the Advancement of Artificial Intelligence 2021 Spring Symposium on Combining Machine Learning and Knowledge Engineering.

Learn more about The Creative AI Hub (opens in new window) , recently launched at TMU’s The Creative School. Professor Rahnama is the hub’s inaugural director.