Improving social competences of virtual agents through artificial consciousness based on the Attention Schema Theory

Introduction

In the last decade, deep learning algorithms have enabled AI systems to perform a series of tasks, like speech and image recognition, as well as or better than humans. However, this technology is not going to be enough to deliver human-level intelligence without consciousness.

Importance of Consciousness

Without understanding the subjective awareness element, it may be impossible to build AI that has a human-like ability to focus its computational resources and intelligently control that focus and interact with people in a socially competent manner.

Project Overview

ASTOUND proposes an Integrative Approach For Awareness Engineering to establish consciousness in machines. The approach consists of an AI architecture for Artificial Consciousness based on the Attention Schema Theory (AST), a novel approach to social cognition that reconciles some of the current most debated cognitive neuroscience theories of consciousness.

Attention Schema Theory

According to the AST, the brain constructs subjective awareness as a schematic model of the process of attention, suggesting that an information-processing machine could attribute consciousness properties to others in a similar way.

Proposed Architecture

The AST-based architecture proposed by ASTOUND will combine:

1. An Attention Mechanism provided by the attentional layers in a deep neural architecture.
2. A Long Term Memory module allowing interplay between internal and external stimuli (data).
3. An Attention Schema that will determine empathic and trustworthy decision-making.

Implementation

ASTOUND will first implement this architecture into a virtual conversational agent (i.e. chatbot) to verify the hypothesis that an artificial consciousness based on AST will improve performance in a task of natural language understanding.

Future Insights

ASTOUND will provide insights into consciousness that are concrete enough, and mechanistic enough, that engineers can build upon it to facilitate future technologies. The study has the potential to be a toolbox for the construction of an EIC Portfolio in conscious AI.