A neural cognitive architecture
Introduction
The mind is extraordinarily complex, and to understand it sophisticated tools are needed. One set of tools for studying the human mind, which emerges from the behaviour of the brain, is cognitive architectures. A cognitive architecture is the fixed or slowly varying structure that forms the framework for the immediate processes of cognitive performance and learning (Newell et al., 1990). This paraphrase is a cornerstone of research in cognitive architectures, and is consistent with many architectures (see Section 2.1).
Human cognition is based on the behaviour of neurons. While Newell et al. (1990) based their architecture on a rule based system, he also partitioned cognition into several bands including the neural band. The coarse topology of neurons is slowly varying, so the coarse human neural topology is a cognitive architecture, a neural cognitive architecture.
One of the main goals of the Human Brain Project (HBP) is to devise a complete simulation of the human brain (Markram, 2012). Simulation can happen using a variety of primitives, but the HBP is particularly interested in spiking point neuron models. This simulation will then be used to help decode the function of the human brain (Amunts et al., 2016).
Cognitive architectures are theories of how the mind works, but as typically used they are formal languages that can be used to develop new models that can be executed on a standard computer. Systems written in the architecture’s language are then run and this behaviour is a model of cognitive behaviour. Each of these written systems instantiate the architecture, though no instance to date (or in the foreseeable future) instantiates a full model of a human. As these instances are executable programs, they are verifiable.
The architecture develops over time to become more effective and more closely approximate cognitive behaviour with new versions extending capabilities and adding constraints. Systems developed in these new frameworks more accurately reflect human cognitive behaviour.
This paper will propose a neural cognitive architecture. The proposal will include a sketch of an initial version of the architecture. This will include a development language, making use of spiking point neurons, that could be used in the near future in simulated spiking neurons on, for instance, the HBP’s EBrains platforms. The paper will then propose ways to move forward from the initial architecture to develop more effective versions that more closely approximate neural and cognitive behaviour.
Section snippets
Literature review
Cognitive architectures have been a research area for at least thirty years, and have been productive for furthering understanding of the mind, and for practical applications of agents, task analysis, and cognitive modelling. More recently, neural cognitive architectures have been proposed (see Section 2.2); these show that neurons are capable flexible processing and memory devices. The HBP is a large interdisciplinary project with a primary goal of simulating the entire brain using spiking
The proposed architecture
The long term goal of the proposed architecture is to build a neural network that closely approximates a human architecture in both neural detail and cognitive function. This raises at least three major questions. 1. What is the basic neural unit? 2. How are those units connected? 3. How does that neural topology generate cognitive function?
There are many neural models including simple rate coded neurons, point models and complex compartmental models (Brette et al., 2007). What is the best
Architecture evolution and exploration
The initial architecture is obviously far from a complete architecture. It will need to develop including more refined neural systems that more closely approximate cognitive function. The author does not want to prejudge this evolution, but would like to propose some plausible future steps. Moreover, others are encouraged in participating in the development of this architecture.
One simple neural mechanism to improve is to have better CAs. CAs have several behaviours, implied by biological,
Conclusions
It is difficult to develop systems in neurons that perform tasks. An instantiated complete neural cognitive architecture is an enormous task. There are, however, known features of neural behaviour that are important that can simplify the search. This paper advocates that the neural cognitive architecture make explicit assumptions and note the evidence behind these assumptions. Almost necessarily, systems developed in the near future will not be complete brains as it is difficult to simulate all
Declaration of Competing Interest
None.
Acknowledgements
This work was supported by This work has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 720270 (the Human Brain Project).
References (40)
- et al.
The human brain project: creating a european research infrastructure to decode the human brain
Neuron
(2016) Intelligence without representation
Artificial Intelligence
(1991)- et al.
Processing with cell assemblies
Neurocomputing
(2010) The symbol grounding problem
Physica D
(1990)- et al.
Soar: An architecture for general cognition
Artificial Intelligence
(1987) - et al.
Working memory 2.0
Neuron
(2018) Neural reuse of action perception circuits for language, concepts and communication
Progress in Neurobiology
(2018)- et al.
The atomic components of thought
(1998) - et al.
A neural-level model of spatial memory and imagery
ELife
(2018) - et al.
Synaptic modifications in cultured hippocampal neurons: Dependence on spike timing, synaptic strength, and postsynaptic cell type
Journal of Neuroscience
(1998)
Simulation of networks of spiking neurons: A review of tools and strategies
Journal of Computational Neuroscience
The NEURON book
A large-scale model of the functioning brain
Science
The neural binding problem (s)
Cognitive Neurodynamics
Overview of the SpiNNaker system architecture
IEEE Transactions on Computers
NEST (neural simulation tool)
Scholarpedia
Engines of the brain: The computational instruction set of human cognition
AI Magazine
The organization of behavior: A neuropsychological theory
Supercomputers ready for use as discovery machines for neuroscience
Frontiers in Neuroinformatics
Cited by (10)
An amalgamation of cognitive aspects in software engineering: A content analysis
2024, Expert Systems with ApplicationsBio-inspired computational object classification model for object recognition
2022, Cognitive Systems ResearchCitation Excerpt :Cognitive Architectures (CAs) are approaches that model the behavior of intelligent agents by dividing the cognitive process into defined parts through the integration of various functional parts for the purpose of achieving certain cognitive behaviors. The structure of a cognitive architecture depends on the approach that inspires it by (Cervantes et al., 2020; Huyck, 2020). Since this model is part of a bio-inspired cognitive architecture (Jaime et al., 2015), it is important to mention how different CAs try to solve the problem of object recognition, as the process of recognition can involve different types of cognitive functions.
Unexpected Aspects of Expectancy in Music: A Spreading Activation Explanation
2023, Psihologijske TemeThe architecture of non-local semantics for artificial general intelligence
2022, International Journal of Applied Systemic StudiesAlgorithms for monitoring the functioning of nonequilibrium information processing systems
2021, Journal of Physics: Conference Series