The Mind According to LIDA
By Stan Franklin and the Cognitive Computing
The Mind According to LIDA is a collection of descriptions, of increasing length and detail, designed to give the reader a quick overview of the LIDA model. Immediately below is a short Abstract, followed by a 2-page Executive Summary, and finally by a Brief Description document.
The Mind According to
LIDA – Abstract
How do minds work? Human minds, animal
minds, artificial minds? What are their underlying mechanisms? What
would answers to such questions even look like? We claim it's best to look for answers
not in terms of neurons or cell assemblies, or in terms of algorithms, but in
terms of systems-level cognitive models that attempt to conceptually account
for everything mental that occurs between an incoming sensory stimulus and a
resulting outgoing motor action. Our LIDA systems-level cognitive model, as yet
partly computational, maintains that all of our ongoing mental activity is
composed of an overlapping sequence of extraordinarily rapid cognitive moments
that we call cognitive cycles. Each such cognitive cycle, a mental building
block, is exceeding complex in itself, consisting of multiple memory systems
and executive systems, together with a host of processes acting upon each of them.
External or internal stimuli are first interpreted so as to update LIDA's
ongoing understanding of its current situation, not always an easy task. The
most salient portions of this understanding come to consciousness to enable learnings in multiple modes, along with the selection and
execution of an appropriate mental or motor response, thus completing the cycle.
Higher level cognitive processes, reasoning, planning, imagining, etc. are
accomplished by cascading these cognitive cycles. An appropriate fleshing out
of this brief description may begin to answer the question "what are the
mechanisms of minds?" But, what fleshing out?
This abstract is intended as an introduction, and an
enticement, to a following two page executive summary of the mind according to
LIDA. That summary is itself an introduction and enticement to a six page brief
description of the LIDA conceptual model. Finally, a guide to the LIDA
literature endeavors to lead the reader through articles and presentations describing
the latest intricacies of the LIDA model, its computational architecture, and
its implemented LIDA-based software agents and robots.
The Mind According to
LIDA – Executive Summary
The LIDA cognitive model attempts to answer the question,
"How do minds work, be they human, animal, or artificial?" Conceived of as a
cognitive prosthesis, the LIDA model is a conceptual theory intended to
facilitate thinking about thinking. It describes conceptually mechanisms
underlying mental activities. As a systems-level model, LIDA attempts to
account for a full range of mental processes from sensing incoming external or
internal stimuli, to producing outgoing motor responses, and every sort of
mental activity in between.
LIDA's basic building block is the cognitive cycle, its
version of the action-perception cycle of the psychologists and the
neuroscientists (Cutsuridis, Hussain,
& Taylor, 2011; Dijkstra, Schoner,
& Gielen, 1994; Freeman, 2002; Neisser, 1976). Emerging asynchronously from the actions of
its processes, each LIDA cognitive cycle consists of three phases. Its
understanding phase interprets the incoming stimuli in light of the current
situation and its past memories, and updates its current situational model. The
following attending phase chooses the most salient portion of the updated
current situation, the conscious content, to be broadcast globally (Baars,
Franklin, & Ramsoy, 2013; Baars, 1988). This
broadcast content enables the final acting and learning phase during which
memories of several sorts are encoded or reinforced, and a suitable
sensory-motor response is selected and executed. Though complex, each cognitive
cycle is quite brief, say 300-600 ms in humans (Koivisto & Revonsuo, 2010; Madl, Baars, & Franklin, 2011).
The LIDA model asserts that all cognitive processing
consists of an overlapping sequence of such cognitive cycles, with different
parts of overlapping cycles running simultaneously in parallel. Though the
model is asynchronous, seriality is maintained by the
conscious broadcasts of the cycles effectively producing the illusion of
continuity of consciousness. The model further asserts that all higher level
cognitive processing, for example, deliberating, reasoning, planning, etc., is
implemented via multiple cognitive cycles acting as cognitive "atoms."
Typical of cognitive models, LIDA is composed of various modules
each with its own processes. LIDA is not only asynchronous, it is almost
entirely local in its processing, with the conscious
broadcast being the only almost global process. Most of the model’s modules are
memory systems, including sensory, perceptual (recognition), spatial, episodic,
attentional, procedural, and sensory motor memories. A preconscious workspace
houses LIDA's current situational model, the global workspace hosts the
competition for consciousness, and there are modules for action selection and
motor plan execution. Even at the so fleeting cognitive cycle level, the LIDA
model is quite complex, with each of these modules claiming its own inner
structure and sophisticated processing.
Currently, the internal structures of several of the LIDA
modules are conceptually and computationally constructed as directed graphs
similar to semantic nets, but embodied. Nodes in the graph represent entities
such as objects, categories, actions, feelings, events, etc., while their links
represent various relationships among these entities. Other LIDA modules are
composed of more complex structures formed from subgraphs
of these same graphs. For mostly computational reasons, there is now a movement
afoot to move from nodes and links representations to sparse distributed vector
representations (Snaider & Franklin, 2012).
The LIDA Computational Framework is a generic and
customizable computational implementation of, as yet, much of the LIDA model,
programmed in Java (Snaider, McCall, & Franklin, 2011). The primary goal of
the Framework is to provide a generic implementation of the LIDA model, easily
customizable for specific problem domains, so as to allow for the relatively
rapid development of LIDA controlled software agents and/or robots. The
Framework permits a declarative description of the specific implementation in
which the architecture of the software agent is specified using an XML
formatted file. In this way, the developer need not define the entire agent in
Java, but can simply specify it using this XML file.
The Framework is intended to be ready customizable at
several levels depending upon the required functionality. At the most basic
level, developers can use the XML file to customize their applications. Within the Framework, several
small data structures can also be customized by adding one’s own versions of
them. Finally, more advanced users can also customize and change the
internal implementation of whole modules. For each module, the Framework
provides default implementations that greatly simplify the customization
In order to show how the computational LIDA architecture can
model human cognition in basic psychological tasks, we have used the LIDA
Framework to develop several cognitive software agents that replicate
experimental data from human subjects (Faghihi, McCall, & Franklin, 2012; Madl, et al., 2011; Madl &
Franklin, 2012). Our main goals with these agents were to substantiate some of
the claims of the LIDA model, and to move towards identifying a set of internal
parameters. Ideally, these internal parameters will remain constant when
disparate datasets from different experiments conducted on human subjects are
reproduced with LIDA agents. Finding such a set of parameters would provide
empirical evidence of the accuracy and usefulness of the conceptual cognitive
As demonstrated by the real world problem solving of LIDA's
predecessor IDA (Franklin, 2003; McCauley & Franklin, 2002), the LIDA model
is quite capable of controlling real-world software agents or robots (Franklin,
2001; Franklin & Jones, 2004)
Every comprehensive model of cognition must be grounded in
the underlying neuroscience. How is this grounding in neuroscience to be
accomplished in the LIDA Model? Perceptual symbols (Barsalou,
1999) in the form of nodes and links in LIDA's perceptual memory comprise the
common representational currency of the LIDA Model. To ground these perceptual
symbols in the underlying neuroscience, we think of them as representing not
neurons or cell assemblies, but rather wings of
chaotic attractors in an attractor landscape (Franklin, Strain, Snaider,
McCall, & Faghihi, 2012; Freeman, 1999; Harter, Graesser, & Franklin,
2001; Skarda & Freeman, 1987). When perturbed by
a previously learned exogenous stimulus such as one that may result from an
inhalation, the spiking trajectory of a cell assembly, such as an olfactory
bulb, falls into a wing of an attractor, and so recognizes an odor. Thus we
postulate non-linear dynamics as an intermediate theory serving to ground
comprehensive cognitive models such as LIDA in the underlying neuroscience.
A still brief, but much more detailed account of the LIDA
model is available.
Franklin, S., & Ramsoy, T. (2013). Global
workspace dynamics: Cortical "binding and propagation" enables conscious contents.
Frontiers in Consciousness Research, 4, 200. doi: 10.3389/fpsyg.2013.00200
J. (1988). A Cognitive Theory of Consciousness.
Cambridge: Cambridge University Press.
Barsalou, L. W. (1999). Perceptual symbol
systems. Behavioral and Brain Sciences, 22, 577–609.
Cutsuridis, V., Hussain, A., & Taylor, J. G. (2011).
Perception-Action Cycle: Models, Architectures, and Hardware (Vol. 1):
Dijkstra, T. M. H., Schoner, G.,
& Gielen, C. C. A. M. (1994). Temporal
stability of the action-perception cycle for postural control in a moving
visual environment. Experimental Brain Research,
McCall, R., & Franklin, S. (2012). A Computational Model
of Attentional Learning in a Cognitive Agent. Biologically
Inspired Cognitive Architectures, 2, 25-36.
(2001). Automating Human Information Agents. In Z. Chen & L. C. Jain
(Eds.), Practical Applications of Intelligent Agents (pp. 27–58 ). Berlin: Springer-Verlag.
(2003). An Autonomous Software Agent for Navy Personnel Work: A Case Study. In
D. Kortenkamp & M. Freed (Eds.), Human
Interaction with Autonomous Systems in Complex Environments: Papers from 2003
AAAI Spring Symposium (pp. 60–65). Palo Alto: AAAI.
Franklin, S., & Jones, D. (2004, October 22-24, 2004). A
Triage Information Agent (TIA) based on the IDA Technology Paper presented at
the AAAI Fall Symposium on Dialogue Systems for Health Communication Washington, DC, USA.
Strain, S., Snaider, J., McCall, R., & Faghihi, U. (2012). Global Workspace Theory, its LIDA model and the underlying
neuroscience. Biologically Inspired Cognitive
Architectures, 1, 32-43. doi:
J. (1999). Consciousness, Intentionality and Causality.
In R. Nunez & W. J. O. R. o. E.
e. Freeman (Eds.), Reclaiming Cognition (pp. 143–172). Exeter:
Freeman, W. J. (2002). The limbic
action-perception cycle controlling goal-directed animal behavior. Neural Networks, 3, 2249-2254.
Graesser, Arthur C., & Franklin, S. (2001). Bridging the gap: Dynamics as a
unified view of cognition. Behavioral and Brain Sciences, 24, 45–46.
& Revonsuo, A. (2010). Event-related brain
potential correlates of visual awareness. Neurosci Biobehav Rev, 34(6), 922-934. doi: 10.1016/j.neubiorev.2009.12.002
Madl, T., Baars, B. J., & Franklin, S. (2011). The Timing of the Cognitive Cycle. PLoS
ONE, 6(4), e14803. doi:
Madl, T., & Franklin, S. (2012). A
LIDA-based Model of the Attentional Blink. Paper presented at the
International Conference on Cognitive Modelling.
& Franklin, S. (2002). A Large-Scale Multi-Agent System
for Navy Personnel Distribution. Connection Science, 14, 371–385
%O Comments: special issue on agent autonomy and groups.
(1976). Cognition and Reality: Principles and Implications of Cognitive
Psychology San Francisco: W. H. Freeman.
Skarda, C., & Freeman, Walter J. (1987). How Brains
Make Chaos in Order to Make Sense of the World. Behavioral and Brain Sciences,
& Franklin, S. (2012). Extended Sparse Distributed Memory and Sequence
Storage. Cognitive Computation, 4(2), 172-180. doi: 10.1007/s12559-012-9125-8
McCall, R., & Franklin, S. (2011). The LIDA Framework as a General Tool for
AGI. Paper presented at the The Fourth Conference on
Artificial General Intelligence (Springer Lecture Notes in Artificial
Intelligence), Mountain View, California, USA.
The Mind According to
LIDA – Brief Account
Download the Brief Account here. (PDF, 20 Pages)