Embodied Awareness

When I started working on the VALAWAI project, I had no idea what the people writing the proposal meant by “awareness”. In this research line, I explore what machine awareness is and how we can improve contextual awareness, especially in social robots.

A Difficult Definition

What was clear from the beginning was that “awareness” was different from “consciousness”, an unpopular term that comes with a far too heavy philosophical and scientific baggage for a PhD student. But I could not wrap my head around the difference between the two terms. In addition, the VALAWAI consortium made it clear that we should always talk about being aware of something. So in the end, the imaginary line that I ended up tracing to distinguish the two is the of itself: saying that someone is aware – without specifying what they are aware of – for me was a synonym of being conscious; everything else was “just awareness”.

Now that consciousness was out of the picture, another problem came up. I will provide an example. If I say that I am aware of the consequences, this means that I know what the consequences are. But surely awareness should be more than mere knowledge, at least that is what my gut feelings said.

Not wanting to fall into the rabbit hole that is the definition of “knowledge” and what it means to know something, and keeping myself well away from the risk of ending up having to define what “understanding” is, I decided to settle on “having access to information” as one of the requirements for machine awareness. From this, other requirements followed, giving a better idea of what the necessary aspects of an aware, embodied AI are.

Neurosciences to the Rescue

Luckily for me, the VALAWAI project proposal pretty much defined awareness as something that can be achieved through the implementation of the Dehaene-Changeux computational model for consciousness awareness, which in turn is based on Baar’s Global Workspace Theory. Wikipedia does a far better job at summarising it.