Skip to main content
Go to the home page of the European Commission (opens in new window)
English en
CORDIS - EU research results
CORDIS

The Global Latent Workspace: towards AI models of flexible cognition

Project description

AI-based architecture for robust and flexible cognition

Advancements in deep learning have allowed artificial intelligence (AI) to achieve human-level performance in various sensory, perceptual, linguistic, and cognitive tasks. However, to enable more robust and flexible cognition, there is a need for innovative cognitive architectures inspired by the human brain. The Global Workspace Theory (GWT) is a cognitive architecture that qualitatively explains conscious and unconscious processes. In this context, the ERC-funded GLoW project aims to develop a roadmap using unsupervised neural translation across multiple latent spaces, resulting in a unique, amodal global latent workspace (GLW) framework. This GLW framework will be directly implemented in deep learning models of increasing complexity. The project will also evaluate the correlation between the GLW framework and brain networks.

Objective

Recent advances in deep learning have allowed Artificial Intelligence (AI) to reach human-level performance in many sensory, perceptual, linguistic or cognitive tasks. There is a growing need, however, for novel, brain-inspired cognitive architectures to achieve more robust and flexible cognition. The Global Workspace Theory refers to a large-scale system integrating and distributing information among networks of specialized modules to create higher-level forms of cognition and awareness. It is one of the dominant neuroscientific accounts of higher-level brain function. We argue that the time is ripe to consider explicit implementations of this theory using deep learning techniques. We propose a roadmap based on unsupervised neural translation between multiple latent spaces (neural networks trained for distinct tasks, on distinct sensory inputs and/or modalities) to create a unique, amodal global latent workspace (GLW). Sensory inputs that are broadcast in this GLW acquire meaning by connecting them to (or translating them into) the relevant semantic knowledge and representations from language, memory or decision systems: the semantic grounding property. Conversely, language and semantic representations are grounded in the sensory environment via the same broadcast/translation operation. Finally, broadcasting sensory and semantic inputs to the relevant effector domains can create motor affordances and support embodied cognition. Together, the grounding and affordance properties infuse meaning in AI processes, which can then be combined sequentially (via attentional selection) to enable flexible cognitive functionsi.e. System-2 AI. The interdisciplinary project will directly implement the GLW framework in deep learning models of growing complexity, and evaluate their correspondence with brain networks. It will provide an explicit evaluation of the Global Workspace Theory, and push the limits of current deep learning systems towards next-generation AI.

Fields of science (EuroSciVoc)

Keywords

Programme(s)

Topic(s)

Funding Scheme

HORIZON-ERC - HORIZON ERC Grants

See all projects funded under this funding scheme

Call for proposal

(opens in new window) ERC-2022-ADG

See all projects funded under this call

Host institution

CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE CNRS
Net EU contribution
€ 2 499 366,00
Address
RUE MICHEL ANGE 3
75794 Paris
France
Region
Ile-de-France Ile-de-France Paris
Activity type
Research Organisations
Links
Total cost
€ 2 499 366,00

Beneficiaries (1)