How to Build a Brain: A neural architecture

for biological cognition

Chris Eliasmith

October 8, 2010
Contents

0.1 Writing tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

0.2 Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1 The science of cognition 9

1.1 The last 50 years . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 How we got here . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Where we are . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 The house of answers . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5 Nengo: An introduction . . . . . . . . . . . . . . . . . . . . . . 29

I How to build a brain 35

2 An introduction to brain building 36
2.1 Brain parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2 Theoretical neuroscience . . . . . . . . . . . . . . . . . . . . . . 42
2.3 A framework for building a brain . . . . . . . . . . . . . . . . . . 45
2.3.1 Representation . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 Transformation . . . . . . . . . . . . . . . . . . . . . . . 53
2.3.3 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.4 The three principles . . . . . . . . . . . . . . . . . . . . . 59
2.4 Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.5 Nengo: Neural representation . . . . . . . . . . . . . . . . . . . . 69

3 Biological cognition – semantics 79

3.1 The semantic pointer hypothesis . . . . . . . . . . . . . . . . . . 80
3.2 Semantics: An overview . . . . . . . . . . . . . . . . . . . . . . 84
3.3 Shallow semantics . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4 Deep semantics for perception . . . . . . . . . . . . . . . . . . . 91

1
CONTENTS 2

3.5 Deep semantics for action . . . . . . . . . . . . . . . . . . . . . . 99

3.6 Meaningful conclusions . . . . . . . . . . . . . . . . . . . . . . . 105
3.7 Nengo: Neural computations . . . . . . . . . . . . . . . . . . . . 109

4 Biological cognition – syntax 111

4.1 Structured representations . . . . . . . . . . . . . . . . . . . . . 111
4.2 Neural binding . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.2.1 Without neurons . . . . . . . . . . . . . . . . . . . . . . 113
4.2.2 With neurons . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3 Manipulating structured representations . . . . . . . . . . . . . . 122
4.4 Learning structural manipulations . . . . . . . . . . . . . . . . . 127
4.5 Clean-up memory and scaling . . . . . . . . . . . . . . . . . . . 130
4.6 Example: Fluid intelligence . . . . . . . . . . . . . . . . . . . . . 135
4.7 Deep semantics for cognition . . . . . . . . . . . . . . . . . . . . 140
4.8 Nengo: Structured representations in neurons . . . . . . . . . . . 145

5 Biological cognition – control 146

5.1 The flow of information . . . . . . . . . . . . . . . . . . . . . . . 146
5.2 The basal ganglia . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.3 Basal ganglia, cortex, and thalamus . . . . . . . . . . . . . . . . 151
5.4 Example: Fixed sequences of action . . . . . . . . . . . . . . . . 154
5.5 Attention and the routing of information . . . . . . . . . . . . . . 155
5.6 Example: Flexible sequences of action . . . . . . . . . . . . . . . 156
5.7 Timing and control . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.8 Nengo: Question answering . . . . . . . . . . . . . . . . . . . . 159

6 Biological cognition – memory and learning 160

6.1 Extending cognition through time . . . . . . . . . . . . . . . . . 160
6.2 Working memory . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3 Example: Serial list memory . . . . . . . . . . . . . . . . . . . . 165
6.4 Biological learning . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.5 Example: Learning new long-term strategies . . . . . . . . . . . 170
6.5.1 Learning new control strategies . . . . . . . . . . . . . . 170
6.5.2 Learning new syntactic manipulations . . . . . . . . . . . 170
6.6 Nengo: Neural dynamics . . . . . . . . . . . . . . . . . . . . . . 184
6.6.1 ***from comp neuro paper*** Neural integrator . . . . . 184
CONTENTS 3

7 The semantic pointer architecture (SPA) 194

7.1 A summary of the SPA . . . . . . . . . . . . . . . . . . . . . . . 194
7.2 Example: A SPA unified model . . . . . . . . . . . . . . . . . . . 194
7.3 A unified view: Symbols and probabilities . . . . . . . . . . . . . 195
7.4 Nengo: Learning (syntactic transformations???) . . . . . . . . . . 197

II Is that how you build a brain? 198

8 Evaluating cognitive theories 199
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
8.2 Quintessential cognitive criteria (QCC) . . . . . . . . . . . . . . 199
8.2.1 Representational structure . . . . . . . . . . . . . . . . . 199
8.2.1.1 Systematicity . . . . . . . . . . . . . . . . . . 200
8.2.1.2 Compositionality . . . . . . . . . . . . . . . . . 200
8.2.1.3 Productivity . . . . . . . . . . . . . . . . . . . 202
8.2.1.4 The massive binding problem . . . . . . . . . . 203
8.2.2 Performance concerns . . . . . . . . . . . . . . . . . . . 204
8.2.2.1 Syntactic generalization . . . . . . . . . . . . . 205
8.2.2.2 Robustness . . . . . . . . . . . . . . . . . . . . 206
8.2.2.3 Adaptability . . . . . . . . . . . . . . . . . . . 208
8.2.2.4 Memory . . . . . . . . . . . . . . . . . . . . . 209
8.2.3 Scientific merit . . . . . . . . . . . . . . . . . . . . . . . 210
8.2.3.1 Triangulation (contact with more sources of data)210
8.2.3.2 Compactness . . . . . . . . . . . . . . . . . . . 211
8.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.4 Nengo: Whole brain modeling . . . . . . . . . . . . . . . . . . . 212

9 Comparison to current cognitive theories 213

9.1 The state of the art . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.1.1 Production systems . . . . . . . . . . . . . . . . . . . . . 216
9.1.2 Synchrony-based approaches . . . . . . . . . . . . . . . . 217
9.1.3 Neural blackboard architectures . . . . . . . . . . . . . . 218
9.1.4 Rumelhart networks . . . . . . . . . . . . . . . . . . . . 221
9.1.5 ACT-R . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.2 Some concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.3 Important oversights . . . . . . . . . . . . . . . . . . . . . . . . 222
9.3.1 Vector processing . . . . . . . . . . . . . . . . . . . . . . 222
CONTENTS 4

9.3.2 ***from Jphil*** A puzzling oversight . . . . . . . . . . 222

10 Conceptual consequences 224

10.1 The same... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10.2 ... but different . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
10.2.1 ***bbs*** Neural plausibility (triangulation) . . . . . . . 225
10.2.2 ***bbs*** Robustness . . . . . . . . . . . . . . . . . . . 227
10.2.3 ****bbs****Scalability . . . . . . . . . . . . . . . . . . 229
10.3 Conceptual? ramifications of the SPA . . . . . . . . . . . . . . . 230
10.3.1 Representation . . . . . . . . . . . . . . . . . . . . . . . 231
10.3.2 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 232
10.3.3 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 233
10.3.4 Inference . . . . . . . . . . . . . . . . . . . . . . . . . . 234
10.4 The road ahead . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
10.4.1 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

A Mathematical derivations for the NEF 238

A.1 Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
A.1.1 Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . 238
A.1.2 Decoding . . . . . . . . . . . . . . . . . . . . . . . . . . 239
A.2 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
A.3 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

B Mathematical derivations for the SPA 244

B.1 Binding and unbinding HRRs . . . . . . . . . . . . . . . . . . . . 244
B.2 Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
B.3 Clean-up memory . . . . . . . . . . . . . . . . . . . . . . . . . . 248

C Mathematical derivations for models 249

C.1 Model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

Bibliography 250
CONTENTS 5

0.1 Writing tips

• Make it personal: talk about the people in the story, their ideas and quirks

(esp for history part)

• Make a case: let the reader be the judge, don’t ‘tell them how it is’.
• Take time to make a point: Writers like gladwell a simple argument that is
illustrated with anecdotes; anecdotes must be scientific research
• Use examples: Give an example and then generalize; but the generalization
is important
• No math: Wrong audience, provide references, put in appendix
• Pick an audience: Keep the audience consistent for this book: Steve (brother)?;
Paul T; Trevor Bekolay; Ronan Reilly
• Citations and notes: Keep them as out of the way as possible, as few notes
as possible, all end notes, including citations.
• Be sure to have specific, detailed connection to neural data – distinguish this
work from connectionism by biological realism.
• Size: 200-ish pages at 350 wds/pg = 70 000 wds.

0.2 Preface
• use chunks of proposal?
• Book is a bit of a throw back to the good ol days when people used to
propose ’unified theories of cognition’, or took themselves to do some kind
of whole-brain modelling. We have long since been lost in the myriad,
amazing, and important details of how specific task are accomplished, and
characterizing the mechanisms of relatively small parts of the brain. I think
we’re in a position to get back to those days. We are no doubt wrong, but
good science establishes a fruitful research program... blah?
CONTENTS 6

• really emphasize the teamwork aspect: ‘dream team’ of students & post-
docs.
• joke: You may be thinking ’sure, give just about any couple a bit of privancy
and about nine months, and they’ll build a brain’. Certainly! And they’ll
build a much better one than I can. But, of course I’m not going to tell you
how to build a brain the way nature does. I’m going to talk about this in a
way that emphasizes a mechanistic understanding of the resulting system.
We want to know how to intervene, why certain thigns happen, etc.
• architect metaphor: Though my name is on the cover of this book, taking
credit for all of the contents is like an architect taking credit for building a
building. I have had a hand in the design, but much of the important work
has been done by others. Especially those in my lab... etc.
• this book is a *beginning*, not an end of a research program! (said in intro
to chp 3 also)
• Admittedly a grand ambition, blah. i’m an engineer and a philosopher, put
them together and you get theo neuro
• remember how the point of cog sci was to understand the whole system...
the reason most researchers get into these areas is to understand the brain...
but everything is highly focussed, partly understandable, but don’t want to
mis the ’big picture’... ala Newell.
• Introduce ‘classical’, and uncertainty about what a cognitive system is:
• The goal of the first chapter is to identify the criteria relevant for distinguish
cognitive from non-cognitive systems, and situate them historically.
• some impressive numbers about nengo, number of neurons running on desk-
top, biggest models run, GPUs and parallel processing, etc.?
• rapid compare/contrast with other simulators? (see reviews)
• mention that the nef is a zeroth order theory in the previous book... this is
similarly a start?
• have a section on ‘all the math you need to read this book’: 1. discussion
of scalars and vectors, as ’arrows’ or just as points x,y on a graph, as col-
lections of numbers; 2. distinguishing linear from nonlinear functions and
CONTENTS 7

functions in general; 3. integration 4. The math that goes beyond anything

described here is in the appendices 5. anything on dynamics? super simple
differential equation 6. dot product? 7. any other notation?

As we will see in chapter two, there have been several ‘non-classical’ attempts at
satisfying these criteria. Whether or not these suggestions will ultimately supplant
classical architectures, or whether they will merely provide domain specific exten-
sions to such architectures, remains to be seen. What is clear, however, is that the
field remains wide open for suggestions as to how we should understand the ba-
sic architecture of cognition. The main purpose of this book is to introduce one
such possible contender. I call it the “semantic pointer” architecture. A detailed
description of this architecture will have to remain until later in the book, after the
appropriate groundwork is delayed. Nevertheless, it is useful to keep some of the
main motivations for positing this particular view in mind. The first motivation is
that most work in cognitive systems seems to be somewhat disconnected from the
enormous amount of data that we have about the biological underpinnings of real,
living cognitive systems. Now, there has been a realization in some quarters, for
instance in many of the projects funded by the EU, that better understanding real
biological systems will help us build better cognitive systems. Nevertheless, there
are no broadly accepted, systemmatic methods for relating biological data to our
high-level understanding of the complex dynamics, and sophisticated representa-
tional properties of cognitive systems. In chapter 3, I will describe one systematic,
quantified method for realizing exactly this goal.
A second motivation for positing the semantic pointer architecture comes from
the realization that once we take seriously the specific computational and rep-
resentational properties of neural systems, we must adapt our understanding of
higher-order cognition to be consistent with those properties. In other words, the
architecture I will propose will adopt cognitively relevant representations, com-
putations, and dynamics that are natural to implement in large-scale, biologically
realistic neural networks. In short, the semantic pointer architecture is centrally
inspired by understanding cognition has a biological process.
Perhaps unsurprisingly, before I can clearly describe this architecture, it will
be essential to more completely characterize the other alternatives for understand-
ing cognitive systems. I provide this characterization in two parts: in the remain-
der of this chapter I will briefly describe the recent history of cognitive science; in
the next chapter I will more specifically characterize several of the more influential
alternatives currently available for understanding cognitive systems. Throughout
this discussion, I’ll be attempting to clearly characterize what we mean by a cog-
CONTENTS 8

nitive system, and identify agreed-upon properties of such systems. In chapter 3,

I will outline the promised systematic method that allows us to move smoothly
from single cell to abstract dynamical descriptions of a neural mechanism, which
is called the Neural Engineering Framework (NEF). In chapter 4, I will adopt this
framework in order to demonstrate how we can characterize biologically relevant
computations and biologically relevant representations for implementing a neu-
rally inspired cognitive architecture. In chapter 5, I describe the semantic pointer
architecture in detail. In the remaining two chapters I evaluate the proposed ar-
chitecture by comparing it to past alternatives. And discuss how research in the
science of cognitive systems should proceed.
Chapter 1

The science of cognition

Questions are the house of answers. – Alex, age 5

1.1 The last 50 years

“What have we actually accomplished in the last 50 years of cognitive systems

research?” was the pointed question put to a gathering of experts from around
the world. They were in Brussels, Belgium, at the headquarters of the European
Union funding agency, and were in the process of deciding how to divvy up about
70 million euros of funding. The room went silent. Perhaps the question was
unclear. Perhaps so much had been accomplished that it was difficult to know
where to start the answer. Or, perhaps even a large room full of experts in the field
did not really know any generally acceptable answers to that question.
The point of this particular call for grant applications was to bring together
large teams of researchers from disciplines as diverse as neuroscience, computer
science, cognitive science, psychology, mathematics, and robotics, in order to un-
ravel the mysteries of how biological cognitive systems are so impressively robust,
flexible, adaptive, and intelligent. This was not the first time such a call had been
made. Indeed, over the course of the last four or five years this agency has funded
a large number of such projects, spending close to half a billion euros. Scientif-

9
CHAPTER 1. THE SCIENCE OF COGNITION 10

Figure 1.1: The iCub. The iCub is an example of a significant recent advance in
robotics. See ???youtube ref??? for videos. ???better picture, permission???

ically speaking, important discoveries have been made by the funded researchers
(ff: examples and refs???). However, these discoveries tend not to be of the kind
that tell us how to better construct integrated, truly cognitive systems. Rather, they
are discoveries in the specific disciplines that are taking part in these “Integrated
Projects.”
For instance, there have been sophisticated new robotic platforms that have
been constructed. One example is the iCub (figure 1.1), which is approximately
the size of a two-year-old child, and has over 56 degrees of freedom.1 The iCub
has been broadly adopted by researchers in motor control, emotion recognition
and synthesis, and active perception (refs???). But, of course, iCub is not a cog-
nitive system. It may be a useful testbed for a cognitive system, no doubt. It may
be a wonder of high-tech robotics engineering, indeed. But, it is not a cognitive
system.
So the pointed question still stands: “What have we actually accomplished in
the last 50 years in cognitive systems research?” That is, what do we now know
about how cognitive systems work, that we did not know 50 years ago? Pes-
1 A degree of freedom is an independent motion of an object. So, moving the wrist up and

down is one degree of freedom, as is rotating the wrist, or moving a finger up and down.
CHAPTER 1. THE SCIENCE OF COGNITION 11

simistically, it might be argued that we do not know too much more than we knew
50 years ago. After all, by 1963 Newell and Simon had described in detail the
program GPS (General Problem Solver). This program, which was an extension
of work that they have started in 1959, is the first in a line of explanations of
human cognitive performance that relied on production systems. Production sys-
tems are, historically, by far the most influential approach to building cognitive
systems. Simply put, a production system consists of a series of productions, or
if-then rules, and a control structure. The job of the control structure is to match a
given input to ‘if’ part of these productions to determine an appropriate course of
action, captured by the ‘then’ part.
In 1963, GPS had all of these features. And, it put them to good use. This
“program that simulates human thought” (???ref pg number) was able to solve
elementary problems in symbolic logic entirely on its own, and went through steps
that often matched those reported by people solving the same problem. GPS could
be given a novel problem, analyze it, attempt a solution, retrace its steps if it
found a dead end (i.e., self-correct), and eventually provide an appropriate solution
(Newell and Simon, 1976). (???1963 ref???)
The success of GPS lead to several other cognitive architectures, all of which
had production systems as their core. Best known amongst these are Soar(Newell,
1990) and ACT (Adaptive Character of Thought; Anderson, 1983). Despite the
many additional extensions to the GPS architecture that were made in these sys-
tems, they share a reliance on a production system architecture. The dominance
of the representational, computational, and architectural assumptions of produc-
tion systems has resulted in such systems being called ‘classical’ approaches to
cognitive systems.
While classical approaches had many early successes, such as finding convinc-
ing solutions to cryptarithmetic problems and solving complex planning tasks,
the underlying architecture does not seem well-suited to interacting with a dy-
namic, real-world environment, or explaining the evolution of real cognitive sys-
tems through time (Eliasmith, 1995). In fairness, more recent work on ACT and
ACT-R (Anderson, 1996), has taken dynamics more seriously, explaining reac-
tion times across a wide variety of psychological tasks (Anderson et al., 2004).
Nevertheless, explaining reaction times addresses only a very small part of cogni-
tive system dynamics in general, and the ACT explanations rely on a 50ms “cycle
time,” which itself is not explained.
In fact, those centrally concerned with the dynamics of cognition largely es-
chew production systems (Port and van Gelder, 1995)(add ref schoner???). If you
want to build a real-world cognitive system – one that actually interacts with the
CHAPTER 1. THE SCIENCE OF COGNITION 12

physics of the world – then the most unavoidable constraints on your system are
dynamic interactions with the world through perception and action. Roboticists,
as a result, seldom use production systems to control their systems. Instead, they
carefully characterize the dynamics of their system, attempt to understand how
to control such a system when it interacts with the difficult-to-predict dynamics
of the world, and look to perception to provide guidance for that control. If-then
rules are seldom used. Differential equations, statistics, and signal processing are
the methods of choice. Unfortunately, it is unclear how to use those same methods
for characterizing high-level cognitive behavior, like language, complex planning,
and deductive reasoning – behaviors for which traditional approaches have the
most success.
In short, there is a broad gap in our understanding of real, cognitive systems:
on the one hand there are the approaches taken to understanding dynamic, real-
world perception and action; on the other hand there are the approaches taken to
understanding higher-level cognition. Unfortunately, these approaches are not the
same. Nevertheless, it is obvious that perception/action and high-level cognition
are not two independent parts of one system. Instead, these two aspects are, in
some fundamental way, integrated in cognitive animals such as ourselves.2 In-
deed, a major theme of this book is to suggest that it is through biology that we
will be able to understand this integration. But for now, I am only concerned to
point out that classical architectures are not obviously appropriate for understand-
ing all aspects of real cognitive systems. This, then, is why we cannot simply say,
in answer to the question of what has been accomplished in the last 50 years, that
we have identified the (classical) cognitive architecture.
However, this does not mean that identifying such architectures is without
merit. On the contrary, one undeniably fruitful consequence of the volumes of
work that surrounds the discussion of classical architectures is the identification
of criteria for what counts as a cognitive system. That is, when proponents of clas-
sicism were derided for ignoring cognitive dynamics, one of their most powerful
responses was to note that their critics had no truly cognitive systems to replace
theirs with. This resulted in a much clearer specification of what a cognitive sys-
tem was. So, I suspect that there would be agreement amongst most of the experts
gathered in Brussels as to what has been accomplished lo these 50 years. Indeed,
the accomplishments are not in the expected form of an obvious technology, a
breakthrough method, or an agreed upon theory. Instead, the major accomplish-
ments have been in clarifying what the questions are, in determining what counts
2 For standard discussions at this point see (refs???).
CHAPTER 1. THE SCIENCE OF COGNITION 13

as a cognitive system, and in figuring out how we are most likely to succeed in
explaining such systems (or, perhaps more accurately, how we are not likely to
succeed).
If true, this is no mean feat. Indeed, it is even more true in science than else-
where that, as economic Nobel laureate Paul A. Samuelson has observed, “good
questions outrank easy answers.” If we actually have more thoroughly identified
criteria for distinguishing cognitive from non-cognitive systems, and if we really
have a good sense of what methods will allow us to understand how and why sys-
tems can successfully meet those criteria, we have accomplished a lot. Ultimately,
only time will tell if we are on the right track. Nevertheless, I believe there is an
overall sense in the field that we have a better idea of what we are looking for in
an explanation of a cognitive system than we did 50 years ago – even if we do not
yet know what that explanation is. Often, progress in science is more about iden-
tifying specific questions that have uncertain answers than it is about proposing
specific answers to uncertain questions.
The goal of this first chapter, then, is to identify these cognitive criteria and
articulate some questions arising from them. These appear in sections 1.3 and
1.4 respectively. First, however, it is worth a brief side trip into the history of the
cognitive sciences to situate the concepts and methods that have given rise to these
criteria.

1.2 How we got here

In the previous section, I identified the ‘classical’ approach to understanding

cognition. And, I contrasted this approach with one that is more centrally inter-
ested in the characterization of the dynamics of behavior. However, much more
needs to be said about the relationship between classical and non-classical ap-
proaches in order to get a general lay-of-the-land in cognitive systems theorizing.
Indeed, much more can be said than I will say here (see, e.g., Bechtel and Graham,
1999). My intent is to introduce the main approaches in order: 1) to identify the
strengths and weaknesses of these approaches, both individually and collectively;
1) to state and clarify the cognitive criteria mentioned earlier; and, ultimately, 3)
to outline a novel theory of biological cognition in later chapters.
In the last half century, there have been three major approaches to theoriz-
CHAPTER 1. THE SCIENCE OF COGNITION 14

ing about the nature of cognition. Each approach has relied heavily on a pre-
ferred metaphor for understanding the mind/brain. Most famously, the classical
approach (aka ‘symbolicism’, or Good Old-fashioned Artificial Intelligence (GO-
FAI)), relies on the “mind as computer” metaphor. Under this view, the mind is
the software of the brain. Jerry Fodor, for one, has argued that the impressive the-
oretical power provided by this metaphor is good reason to suppose that cognitive
systems have a symbolic “language of thought” which, like a computer program-
ming language, expresses the rules that the system follows (Fodor, 1975). Fodor
claims that this metaphor is essential for providing a useful account of how the
mind works. Production systems, which I have already introduced, have become
the preferred implementation of this metaphor.
A second major approach is often called ‘connectionism’ (aka the Parallel Dis-
tributed Processing (PDP) approach or New-fangled Artificial Intelligence (NFAI)).
In short, connectionists explain cognitive phenomena by identifying large net-
works of typically identical nodes, that are connected together in various patterns.
Each node performs a simple input/output mapping. However, when grouped
together in sufficiently large networks, the activity of these nodes is interpreted
as implementing rules, analyzing patterns, or performing any of several other
cognitively-relevant behaviors. Connectionists, like the symbolicists, rely on a
metaphor for providing explanations of cognitive behaviors. This metaphor, how-
ever, is much more subtle than the symbolicist one; these researchers presume
that the functioning of the mind is like the functioning of the brain. The subtlety
of the “mind as brain” metaphor lies in the fact that connectionists also hold that
the mind is the brain. However, when providing cognitive descriptions, it is the
metaphor that matters, not the identity. In deference to the metaphor, the founders
of this approach call it “brain-style” processing, and claim to be discussing “ab-
stract networks” (Rumelhart and McClelland, 1986). In other words, their models
are not supposed to be direct implementations of neural processing, and hence
cannot be directly compared to the kinds of data we gather from real brains. This
is not surprising since the computational and representational properties of the
nodes in connectionist networks bear little resemblance to neurons in real biolog-
ical neural networks.3
The final major approach to cognitive systems in contemporary cognitive sci-
ence can be called ‘dynamicism,’ and is often identified with ‘embedded’ or ‘em-
bodied’ approaches to cognition. Proponents of dynamicism also rely heavily
on a metaphor for understanding cognitive systems. Most explicitly, van Gelder
3 As discussed in chapter 10 of (Bechtel and Abrahamsen, 2001).
CHAPTER 1. THE SCIENCE OF COGNITION 15

employs the Watt Governor as a metaphor for how we should characterize the
mind (van Gelder, 1995). It is through his analysis of the best way to characterize
this dynamic system that van Gelder argues for understanding cognitive systems
as non-representational, low-dimensional, dynamic systems. Like the Watt Gov-
ernor, van Gelder maintains, cognitive systems are essentially dynamic and can
only be properly understood by characterizing their state changes through time.
The “mind as Watt Governor” metaphor suggests that trying to impose any kind
of discreteness, either temporal or representational, will lead to a mischaracter-
ization of cognitive systems. This same sort of analysis – one which highlights
the importance of dynamics – highlights the essential coupling of cognitive sys-
tems to their environment (van Gelder and Port, 1995). Dynamic constraints are
clearly imposed by the environment on the success of our behavior (we must see
and avoid the cheetah before it eats us). If our high-level behaviors are built on
our low-level competencies, then it is not surprising that identifying this important
role of the environment has lead several researchers to emphasize the fact that real
cognitive systems are embedded within a specific environment, with specific dy-
namics. Furthermore, they have argued, the nature of that environment can have
significant impacts on what cognitive behaviors are realized (ref andy??? oth-
ers???). Because many of the methods and assumptions of dynamicsm and em-
bedded approaches are shared, in this discussion I group both under the heading
of ‘dynamicism.’
Notably, each of symbolicism, connectionism, and dynamicism, rely on metaphor
not only for explanatory purposes, but also for developing the conceptual foun-
dations of their preferred approach to cognitive systems. For symbolicists, the
properties of Turing machines become shared with minds because both are com-
putational systems. For connectionists, the character of representation changes
dramatically under their perferred metaphor. Mental representations are taken to
consist of “sub-symbols” associated with each node, while “whole” representa-
tions are real-valued vectors in a high-dimensional property space.4 Finally, be-
cause the Watt Governor is best described by the mathematics of dynamic systems
theory, which makes no reference to computation or representation, dynamicists
claim that our theories of mind need not appeal to computation or representation
either (van Gelder, 1998).
I have argued elsewhere that our understanding of cognitive systems needs
4 See, for example, (Smolensky, 1988). Notably, there are also connectionist models that take
activities of individual nodes to be representations. These are still very much unlike symbolic
representations.
CHAPTER 1. THE SCIENCE OF COGNITION 16

to move beyond such metaphors (Eliasmith, 2003). We need to move beyond

metaphros because, in science, metaphorical thinking can sometimes unduly con-
strain available hypotheses. This is not to deny that metaphors are incredibly
useful tools at many points during the development of scientific theory. It is only
to say that, sometimes, metaphors only go so far. Take, for instance, the develop-
ment of the current theory of the nature of light. In the nineteenth century, light
was understood in terms of two metaphors: light as a wave, and light as a par-
ticle. Thomas Young was the best known proponent of the first view, and Isaac
Newton was the best known proponent of the second. Each used their favored
metaphor to suggest new experiments, and develop new predictions.5 Thus, these
metaphors played a role similar to that played by the metaphors discussed above
in contemporary cognitive science. However, as we know in the case of light, both
metaphors are false. Hence the famed “wave-particle duality” of light: sometimes
it behaves like a particle; and sometimes it behaves like a wave. Neither metaphor
by itself captures all the phenomena displayed by light, but both are extremely
useful in characterizing some of those phenomena. So, understanding what light
is required moving beyond the metaphors.
I believe that the same is true in the case of cognition. Each of the metaphors
mentioned above has some insights to offer regarding certain phenomena dis-
played by cognitive systems. However, none of these metaphors is likely lead
us to all of the right answers. Thus, we ideally want a way of understanding cog-
nitive systems that draws on the strengths of symbolicism, connectionism, and
dynamicism, but does not depend on the metaphors underlying these approaches.
In fact, I believe we are in a historically unique position to affect this kind
of understanding. This is because we currently have more detailed access to the
underlying biology of cognition than ever before. We can record large-scale blood
flow in the brain during behavior using fMRI, we can image medium-sized net-
works of cells during the awake behavior of animals, we can record many individ-
ual neurons inside and outside of the functioning brain, and we can even record
the opening and closing of individual channels, which are about 200 nanometers
in size, on the surface of a single neural cell.
This kind of information is exactly what we need to free ourselves of high-
level guiding metaphors. To be clear, the preceding sentence is not the claim that
we should get rid of all metaphors. I by no means think that is plausible. Instead,
I am advocating that we reduce our reliance on metphors as theoretical arbiters.
5 Fora detailed description of the analogies, predictions, and experiments, see (Eliasmith and
Thagard, 1997).
CHAPTER 1. THE SCIENCE OF COGNITION 17

That is, we need to remove them from a role in determining what is a good ex-
planation and what is not. If we can relate our cognitive explantions to specific
biological measurements, then it is that data, not a metaphor, that should be the
determiner of the goodness of our explanation. If we can identify the relationship
between cognitive theories and biological data, then we can begin to understand
cognitive systems for what they are: complex, dynamic, biological systems. The
three past approaches I have discussed provide few, if any, hints as to the nature
of this relationship.
One reason such hints are scarce is that, historically speaking, theories regard-
ing cognitive systems mainly come from psychology, an offshoot of philosophy
of mind in the late 19th century. At the time, very little was known of the biology
of the nervous system. There was some suggestion that the brain was composed
of neurons, but this was highly contentious. Certainly there was little understand-
ing of how cells could be organized to control our simple behaviors, let alone our
complex, cognitive ones.
So, cognitive theorizing has proceeded without much thought about biology.
In the early days of psychology, most such theories were generated by the method
of introspection: i.e., sitting and thinking very carefully, so as to discern the com-
ponents of cognitive processing. Unfortunately, different people introspected dif-
ferent things, and so there was a crises in ‘introspectionist’ psychology. This crisis
was resolved by ‘behaviorist’ psychologists who simply disallowed introspection.
The only relevant data for understanding cognitive systems was data that could be
gleaned from the ‘outside’, i.e., from behavior.
While much is sometimes made of the difference between ‘philosophical’ and
‘psychological’ behaviorism, there was general agreement on this much: internal
representations, states, and structures are irrelevant for understanding the behavior
of cognitive systems. For psychologists, like Watson and Skinner, this was true
because only input/output relations are scientifically accessible. For philosophers,
like Ryle, this was true because mental predicates (like ‘believes’, ’wants’, etc.),
if they were to be consistent with natural science, must be analyzable in terms of
behavioral predicates. In either case, looking inside the “black box” that was the
system of study, was prohibited.
Interestingly, engineers of the day respected a similar constraint. In order to
understand dynamic physical systems, the central tool they employed was clas-
sical control theory. Classical control theory, perhaps notoriously, only charac-
terizes physical systems in terms of their input/output relationship. As a result,
classical control theory is limited to designing non-optimal, single-variable, static
controllers and depends on graphical methods, rules of thumb, and does not allow
CHAPTER 1. THE SCIENCE OF COGNITION 18

for the inclusion of noise.6 While the limitations of classical controllers and meth-
ods are now well-known, they nevertheless allowed engineers to build systems of
kinds they had not systematically built before: goal-directed systems.
While classical control theory was practically useful, especially in the 1940s
when there was often a desire to blow things up (the typical goal at which such
systems were directed), some researchers thought the theory had more to offer.
They suggested that classical control theory could provide a foundation for de-
scribing living systems as well. Most famously, the interdisciplinary movement
founded in the early 1940s known as ‘cybernetics’ was based on precisely this
contention.7 Cyberneticists claimed that living systems, like classical control sys-
tems, were essentially goal-directed systems. Thus, closed-loop control should
be a good way to understand the behavior of living systems. Given the nature of
classical control theory, cyberneticists focused on characterizing the input/output
behavior of living systems, not their internal processes. Unfortunately for cyber-
neticists, in the mid-50s there was a massive shift in how cognitive systems were
viewed.
With the publication of a series of seminal papers, 8 the ‘cognitive revolu-
tion’ took place. One simplistic way to characterize the move from behaviorism
to ‘cognitivism’ is that it became no longer taboo to look inside the black box.
Quite the contrary: internal states, internal processes, and internal representations
became standard fare when thinking about the mind. Classical control theory no
longer offered the kinds of analytic tools that mapped easily onto this new way
of conceiving complex biological behavior. Instead, making sense of the insides
of that black box was heavily influenced by concurrent successes in building and
programming computers to perform complex tasks. Thus, many early cognitive
scientists saw, when they opened the lid of the box, a computer. As explored in
detail by Jerry Fodor, “[c]omputers show us how to connect semantical [meaning-
related] with causal properties for symbols” (Fodor, 1987, p. 18), thus computers
have what it takes to be minds. Once cognitive scientists began to think of minds
as computers, a number of new theoretical tools became available for character-
izing cognition. For instance, the computer’s theoretical counterpart, the Turing
machine, suggested novel philosophical theses, including ‘functionalism’ (the no-
tion that only the mathematical function computed by a system was relevant for
its being a mind or not) and ‘multiple realizability’ (the notion that a mind could
6 For a succinct description of the history of control theory, see (Lewis, 1992).
7 For a statement of the motivations of cybernetics, see (Rosenblueth et al., 1943).
8 These papers include, but are not limited to Newell et al. (1958), Miller (1956), Bruner et al.

(1956), and Chomsky (1959).

CHAPTER 1. THE SCIENCE OF COGNITION 19

be implemented (i.e. realized), in pretty much any substrate – water, silicon, inter-
stellar gas, you name it – as long as it computed the appropriate function). More
practically, the typical architecture of computers, the von Neumann architecture
(which is a predecessor of the architecture of a production system), was thought
by many to be relevant for understanding our cognitive architecture.
Eventually, however, adoption of the von Neumann architecture for under-
standing minds was seen by many as poorly motivated. Consequently, the early
1980s saw a significant increase in interest in the connectionist research program.
As mentioned previously, rather than adopting the architecture of a digital com-
puter, these researchers felt that an architecture more like that seen in the brain
would provided a better model for cognition. It was also demonstrated that a
connectionist architecture could be as computationally powerful as any symbol-
icist architecture (ref???). But despite the similar computational power of the
approaches, the specific problems at which each approach excelled were quite dif-
ferent. Connectionists, unlike their symbolicist counterparts, were very successful
at building models that could learn and generalize over the statistical structure of
their input. Thus, they could begin to explain many phenomena not easily cap-
tured by symbolicists, such as object recognition, reading, concept learning, and
other behaviors crucial to cognition.
For some, however, connectionists had clearly not escaped the influence of the
mind-as-computer metaphor. Connectionists still spoke of representations, and
thought of the mind as a kind of computer. These dynamicists, who we have met
before, suggested that if we want to know which functions a system can actu-
ally perform in the real world, we must know how to characterize the system’s
dynamics. Consequently, since cognitive systems evolved in dynamic environ-
ments, we should expect evolved control systems, like brains, to be more like the
Watt Governor – dynamic, continuous, coupled directly to what they control –
than like a discrete-state Turing machine that computes over ‘disconnected’ rep-
resentations. As a result, dynamicists suggested that dynamic systems theory,
not computational theory, was the right quantitative tool for understanding minds.
They claimed that notions like ‘chaos,’ ‘hysteresis,’ ‘attractors,’ and ‘state-space’
underwrite the conceptual tools best-suited for describing cognitive systems.
So, as we have just seen, each of the three positions grew out of critical evalu-
ation of previous positions. Connectionism was a reaction to the over-reliance on
computer architectures for describing cognition. Dynamicism was a reaction to an
under-reliance on the importance of time and our connection to the physical envi-
ronment. Even symbolicism was a reaction to its precursor, behaviorism, which
had ruled out a characterization of cognition which posited states internal to the
CHAPTER 1. THE SCIENCE OF COGNITION 20

agent. Consequently, each of the metaphors have been chosen to emphasize dif-
ferent aspects of cognition, and hence driven researchers in these areas to employ
different formalisms for describing their cognitive theories. Simply put, symbol-
icists use production systems, connectionists use networks of simple nodes, and
dynamicists use sets of differential equations to explain our most complex behav-
iors.
While we can see the progression through these approaches as rejections of
their predecessors, it is also important to note what was preserved. Symbolicism
preserved the commitment to providing scientific explanations of cognitive sys-
tems. Connectionism retained a commitment to the notions of representation and
computation so central to symbolicist approaches. And dynamicism, perhaps the
most self-conscious attempt to break away from previous methods, has not truly
broken with tradition. Rather, dynamicists have most convincingly argued for a
shift in emphasis: they have made time a non-negotiable feature of cognition.9
Perhaps, then, a successful break from all of these metaphors will be able to
relate each of the central methods to one another in a way that preserves the central
insights of each. And, just as importantly, such a break from past approaches
should allow us to see how our theories relate to the plethora of data we have about
brain function. In my more impetuous moments, I have been known to argue that
the methods developed in this book are one such break. I have such moment in
section ???. Before such arguments can be made with any force, however, much
work needs to be done. First, for instance, I need to use the historical background
just presented to identify what I earlier claimed was the major advance of the the
last 50 years: identifying criteria for distinguishing cognitive from non-cognitive
systems.
1.3 Where we are
It will strike many as strange to suggest that there is anything like agreement
on criteria for identifying cognitive systems. After all, this area of research has
been dominated by, shall we say, “vigorous debate” between proponents of each
of the three approaches. It is true that there are instances of symbolicists calling
connectionism “quite dreary and recidivist” (Fodor, 1995). Nevertheless, I believe
9 However, it should be noted that Newell, one of the main developers of production systems,
was quite concerned with the timing of cognitive behavior. He dedicated a large portion of his
magnum opus, Unified Theories of Cognition, to the topic. Nevertheless, time is not a neces-
sary feature of production systems, and so his considerations seem, in may ways, after the fact
(Eliasmith, 1996).
CHAPTER 1. THE SCIENCE OF COGNITION 21

that there is some agreement on what the target of explanation is. By this I do not
mean to suggest that there is an agreed-upon definition of cognition. But rather
that ‘cognition’ is to most researchers what ‘pornography’ was to justice Potter
Stewart: “I shall not today attempt further to define [pornography]; and perhaps I
could never succeed in intelligibly doing so. But I know it when I see it.”
Researchers seem to know cognition when they see it, as well. There are many
eloquent descriptions of various kinds of behavior in the literature, which most
readers – regardless of their commitments to symbolicism, connectionism, or dy-
namicism – recognize as cognitively relevant behavior. It is not as if symbolicists
think that constructing an analogy is a cognitive behavior and dynamicists dis-
agree. This is why I have suggested that we may be able to identify agreed-upon
criteria for the characterization of cognitive systems. To be somewhat provoca-
tive, I will call these the ‘Quintessential Cognitive Criteria,’ or ‘QCC’ for short.
I should note that this section is only a first pass and summary of considerations
for the proposed QCC in table 1.1. I return to a much more detailed discussion of
them before explicitly employing them in chapter 8.
Criteria are, of course, not necessary and sufficient conditions that a system
must meet to be cognitive. They are not, in other words, definitive. Instead,
they are standards against which a system can be judged to determine to what
extent it has properties relevant for cognition. Since criteria are not necessary
and sufficient conditions, employing them makes it natural to claim that some
systems are ‘more cognitive’ than others. This suggests that there is a ‘cognitive
continuum.’ A rough ordering of some common research subjects in terms of their
‘cognitiveness’ might be: humans, chimpanzees, monkeys, cats, rats, turtles, flies,
and worms.10 If we appropriately identify the QCC, they should be consistent with
this kind of intuitive ordering. In other words, to place a system on the cognitive
continuum, we can appeal to the number and degree to which the QCC are met:
the more and the better the QCC are met, the more cognitive the system will be.
So what are the QCC? Let us turn to what have researchers said about what
makes a system cognitive. Here are examples from proponents of each view:

• Dynamicism (van Gelder, 1995, p. 375-6): “[C]ognition is distinguished

from other kinds of complex natural processes... by at least two deep fea-
tures: on the one hand, a dependence on knowledge; and distinctive kinds
of complexity, as manifested most clearly in the structural complexity of
10 As a matter of interest, I conducted an informal poll of 20 researchers spanning the cognitive
disciplines, and 18 gave this response. One response was incomplete, and the other placed turtles
ahead of rats.
CHAPTER 1. THE SCIENCE OF COGNITION 22

natural languages.”.

• Connectionism (McClelland and Rumelhart, 1986, p. 13): Rumelhart and

McClelland explicitly identify the target of their well-known PDP research
as “cognition.” To address it, they feel that they must explain “motor con-
trol, perception, memory, and language”.

• Symbolicism (Newell, 1990, p. 15): Newell presents the following list of

behaviors in order of their centrality to cognition: 1) problem solving, deci-
sion making, routine action; 2) memory, learning, skill; 3) perception, mo-
tor behavior; 4) language; 5) motivation, emotion; 6) imagining, dreaming,
daydreaming.
These examples do not provide criteria for identifying cognitive systems, but
rather attempt to identify which particular aspects of behavior must be explained
in order to successfully explain cognition. However, from such lists we can distill
the QCC that capture the intuitions motivating these researchers. Notice that there
are several commonalities among the lists. First, language appears in all three.
This is no surprise as language is often taken to be the pinnacle of human cogni-
tive ability. But, there are other important shared commitments evident in these
lists. For instance, each identify the importance of adaptability and flexibility. For
van Gelder adaptability is evident through his identification of the dependence of
cognitive behavior on knowledge. For the other two, explicit mention of memory
and learning highlight a more general interest in adaptability. As well, while not
in this specific quote from van Gelder, dynamicism is built on a commitment to
the centrality of action and perception to cognition (Port and van Gelder, 1995).
Perhaps surprisingly, then, it is the connectionists and symbolicist who clearly
identify motor control and perception as important to understanding cognition. In
any case, it is clear that all three agree on the important role of these, more basic,
processes.
While these are simple lists, I believe we can see in them the motivations for
subsequent discussions explicitly aimed at identifying what it takes for a system to
be cognitive. Let us briefly consider some of these more direct discussions. One
of the first, and perhaps the most well-known discussions, is provided by Fodor
and Pylyshyn in their 1988 paper “Connectionism and cognitive architecture: A
critical analysis.” While mainly a critique of the connectionism of the day, this
paper also provides three explicit constraints on what it takes to be a cognitive
system. These are productivity, systematicity, and compositionality. Productivity
is the ability of a system to generate a large number of representations based on a
CHAPTER 1. THE SCIENCE OF COGNITION 23

few basic representations (a lexicon) and rules for combining them (a grammar).
Systematicity refers to the fact that some sets of representations (generated pro-
ductively) come together. For instance, they suggest that cognitive systems cannot
represent ‘John loves Mary’ with out thereby being able to represent ‘Mary loves
John’. Finally, compositionality is the suggestion that the meaning of complex
representations is a direct ‘composition’ (i.e. adding together) of the meanings of
the basic representations.
More recently, Jackendoff has dedicated his book to identifying challenges for
a cognitive neuroscience of cognition (Jackendoff, 2002). In it he suggests that
there are four main challenges to address when explaining cognition. Specifically,
Jackendoff’s challenges are: 1) the massiveness of the binding problem (that very
many basic representations must be bound to construct a complex representation);
2) the problem of 2 (how multiple instances of one representational token can
be distinguished); 3) the problem of variables (how can roles (e.g. ‘subject’) in
a complex representation be generically represented); and 4) how to incorporate
long-term and working memory into cognition. Some of these challenges are
closely related to those of Fodor and Pylyshyn, and so are integrated with them as
appropriate in the QCC (see table 1.1).
The Fodor, Pylyshyn, and Jackendoff criteria come from a classical, symboli-
cist perspective. In a more connectionist-oriented discussion, Don Norman sum-
marizes several papers he wrote with Bobrow in the mid-70s, in which they argue
for the essential properties of human information processing (???ref). Based on
their consideration of behavioral data, they argue that human cognition is: ro-
bust (appropriately insensitive to missing or noisy data, and damage to its parts),
flexible, and relies on ‘content-addressable’ memory. Compared to symbolicist
considerations, the emphasis in these criteria has moved from representational
constraints to more behavioral constraints, driven by the ‘messiness’ of psycho-
logical data.
Dynamicists can be seen to continue this trend towards complexity in their
discussions of cognition. Take, for instance, Gregor Schoner’s discussion in his
article “Dynamical systems approaches to cognition” (???ref). In his opening
paragraphs, he provides examples of the sophisticated action and perception that
occurs during painting, and playing in a playground. He concludes that “cognition
takes place when organisms with bodies and sensory systems are situated in struc-
tured environments, to which they bring their individual behavioral history and to
which they quickly adjust“ (p. 101). Again, we see the importance of flexibility
and robustness, with the addition of an emphasis on the role of the environment.
Before presenting a final summary of the QCC, I believe there are additional
CHAPTER 1. THE SCIENCE OF COGNITION 24

criteria that a good theory of cognition must meet. I suspect that these will not be
controversial, as they are a summary of insights that philosophers of science have
generated in their considerations of what constitutes a good scientific theory in
general (refs???). I take it as too obvious to bother stating that each approach as-
sumes that we are trying to construct a scientific theory cognitive systems. Never-
theless, these criteria may play an important role in distinguishing good cognitive
theories from bad.
Two of the most important considerations for good scientific theories are those
of unity and simplicity. Good scientific theories are typically taken to be unified:
the more sources of data, and the more scientific disciplines that they are consis-
tent with, the better the theory. One of the reasons Einstein’s theory of relativity
is to be preferred over Newton’s theories of motion is that the former is consis-
tent with more of our observations. That is, we can triangulate our data sources
in such a way as to prefer one over the other. In addition, good theories tend to
simplicity. That is good theories can be stated compactly. The reason that the
heliocentric theory of our solar system is to be preferred over a geocentric one is
that, in the latter, we need to specify not only the circular paths of the planets,
but also the many infamous ‘epicycles’ of each planet in order to explain their
motions. In contrast, the heliocentric theory needs to specify one simple ellipse
for each planet.
Though this discussion has been brief, I believe that these considerations, cou-
pled with the earlier evidence of a convergence of an understanding of cognitive
systems, provide a reasonably clear indication of several criteria that researchers
in each of the three approaches would agree to. As a result, table 1.1 summarizes
the QCC we can, at least on the face of it, extract from this discussion. As a re-
minder, I do not expect the mere identification of these criteria to be convincing.
A detailed discussion of each is presented in chapter 8.

1.4 The house of answers

While the QCC should prove useful for evaluating a characterization of cognition,
they are not obviously useful for directing such a characterization. Instead, I
think we need to turn to specifying a few central questions that have arisen in the
last 50 years of cognitive science research. As a result, in this section I suggest
four questions which, if answered in detail, should go a long way to providing a
characterization of a cognitive system that addresses most, if not all, of the QCC.
These questions are:
CHAPTER 1. THE SCIENCE OF COGNITION 25

1. Representational structure
a. Systematicity
b. Compositionality
c. Productivity (the problem of variables)
d. The massive binding problem (the problem of two)
2. Performance concerns
a. Syntactic generalization
b. Robustness
c. Adaptability
d. Memory
3. Scientific merit
a. Triangulation (Contact with more sources of data)
b. Compactness
Table 1.1: Quintessential Cognitive Criteria (QCC) for theories of cognition.

1. How are semantics captured by the system?

2. How is syntactic structure encoded and manipulated by the system?

3. How is information flexibly routed through the system in response to task

demands?

4. How are memory and learning accounted for by the system?

A long-standing concern when constructing models of cognitive systems is how
to characterize the relationship between the representations inside the system, and
the objects in the external world which they purportedly represent. That is, how do
we know what a representation means? Of course, for any system we construct,
we can simply define what the meaning of particular representation is. Unfortu-
nately, this is very difficult to do convincingly: consider trying to define a map-
ping between dogs-in-the-world and a state in your head that acts like our concept
‘dog’. The vast psychological literature on concepts points to the complexity of
this kind of mapping. Most researchers in cognitive science are well aware of,
and dread, addressing this problem, which is often called the ‘symbol grounding
problem’ (ref?). Nevertheless, any implemented model of a cognitive system must
make some assumptions about how the representations in that system get their
meaning. Consequently, answering this first question will force anyone wishing
to characterize cognition to, at the very least, state their assumptions about how
CHAPTER 1. THE SCIENCE OF COGNITION 26

internal states are related to external ones. Whatever the story, it will have to plau-
sibly apply not only to the models we construct, but also the natural systems we
are attempting to explain.
The second question addresses what I have already noted is identified nearly
universally as a hallmark of cognitive systems: the manipulation of structured
representations. Whether we think internal representations themselves are struc-
tured (or, for that matter, if they even exist), we must face the undeniable ubiquity
of behavior involving the manipulation of language-like representations. Con-
sequently, any characterization of cognition will have to tell a story about how
syntactic structure is encoded and manipulated by a cognitive system. Answering
this question will unavoidably address at least the first five criteria in the QCC.
Another broadly admired feature of cognitive systems is their incredible, rapid
adaptability. People put into a new situation can quickly survey their surround-
ings, identify problems, and formulate plans for solving those problems – often
within the space of a few minutes. Performing each of these steps demands coor-
dinating the flow of huge amounts of information through the system. Even seem-
ingly mundane changes in our environment, such as switching from a pencil to a
marker while writing, pose difficult control problems. Such a simple switch can
alter the weight of the hand, change what are valid configurations of the fingers,
and modify the expected visual, auditory, and proprioceptive feedback signals. All
of these pieces of information can influence what is the best motor control plan,
and so must be taken into account when constructing such a plan. Because the
sources of such information can remain the same (e.g. visual information comes
from the visual system), while the destination of the information may change (e.g.,
from arm control to finger control), a means of routing that information must be in
place. More cognitively speaking, if I simply tell you that the most relevant infor-
mation for performing a task is going to switch from something you are hearing
to something you will be seeing, you can instantly reconfigure the information
flow through your brain to take advantage of that knowledge. Somehow, you are
re-routing the information you use for planning to come from the visual system
instead of the auditory system. We do this effortlessly, we do this quickly, and we
do this constantly.
The fourth question focuses our attention on another important source of cog-
nitive flexibility: our ability to use past information to improve our performance
on a future task. The timescale over which information might be relevant to a
task ranges from seconds to many years. Consequently, it is not surprising that
the brain has developed mechanisms that also range over these timescales. Mem-
ory and learning are behavioral descriptions of the impressive abilities of these
CHAPTER 1. THE SCIENCE OF COGNITION 27

mechanisms. Considerations of memory and learning directly address several of

the performance concerns identified in the QCC. Consequently, any characteri-
zation of a cognitive system has to provide some explanation for how relevant
information is propagated through time, and how the system can adapt to its past
experience.
As mentioned earlier, I believe the great success of cognitive science is the
improved ability to formulate questions that need to be addressed when studying
cognition. The four broad questions I have described here are really intended to
highlight much larger classes of more specific, and hence more useful, questions.
(And those of you who read the table of contents may notice that they also align
with the aims of chapters 3 through 6, respectively.) The purpose of only iden-
tifying these four as central to characterizing a cognitive system, is to avoid the
possibility of getting lost in the details of a laundry list of detailed questions rele-
vant to cognition. Nevertheless, compiling such a list can be helpful. In table 1.2
I provide a small snippet of such a list.
Having lists of questions like this can prove helpful for guiding our discussion
of cognitive systems, but it does not suggest a way we might go about answering
them. However, I think we can take an approach in cognitive science that has been
espoused as crucial for other sciences. That is, we can give a good characterization
of a cognitive system if we show how to build one.
Richard Feynman, the famous physicist, wrote a few of his last, perhaps most
dear, thoughts on the blackboard in his office shortly before he died. After his
death, someone had the foresight to take a photograph of his blackboard for poster-
ity, which you can see at http://abbynuss1.tripod.com/id32.htm. In large
letters, in the top left corner, he wrote:
What I cannot create I do not understand.
I would like to propose to adopt this as a motto for understanding cognition. It
suggests that creating a cognitive system would provide one of the most convinc-
ing demonstrations that we truly understand such a system. Of course, in this case,
as in the case of many other physical systems, ‘creating’ the system amounts to
creating simulations of the underlying mechanisms in detail. Although I will not
argue for this point in great depth – though many others have (refs? dretske?,
mech people) – but, hopefully an explanation of a cognitive system that begins
with neurons as component parts, characterizes those basic mechanisms quantita-
tively, and demonstrates how they can be arranged to give rise to a wide variety
of cognitive behaviors will prove compelling evidence for the claim that such a
characterization provides us with a improved understanding of cognitive systems.
CHAPTER 1. THE SCIENCE OF COGNITION 28

Table 1.2: A partial laundry list of important questions that should be addressed
by an account of cognition.
Semantics questions
Can the semantic account be the same for all neural representations (i.e., from light intensity up to ful
Are there amodal representations, how is their content determined?
What is the role and importance of hierarchies in visual/motor systems to representational content?
What is the relation between an increase in hierarchical level and an increase in representational com
Does our semantic characterization (have to) solve the frame problem?
Syntax questions
What kind of neurally realistic architectures can support syntactic structure?
What kinds of evidence can we bring to bear to distinguish approaches to neural binding?
Does syntactic binding use the same mechanism as perceptual binding? What is the relation between
Is our chosen syntactic mechanism fast enough to account for on-the-fly structure processing?
Can our syntactic mechanism scale to allow efficient processing of vocabularies of the size employed
Control questions
How does such a seemingly distributed system act in such a unified way?
How can we efficiently deal with the curse of dimensionality?
How, in a generic sense, are neurons organized to give rise to dynamics and computations that are use
Does the architecture decompose complex control problems (if so, how), or have specialized controlle
What is the precise nature of the subtle intercommunication between perceptual, cognitive, and motor
Memory and learning questions
How does the wide variety of timescales relevant for cognition arise from a brain, whose individual c
How are internal signals that drive adaptation generated?
What kind of architecture can be adaptable in the many ways that brains are? (from learning fine-mot
What kinds of memory are there, and how do these map onto brain mechanisms?
How much of our proposed architecture can be learned and how much (and what precisely) must be i
Other questions
How do ‘low-level’ properties of systems affect ‘high-level’ function? (What do we mean by low and
How do so few cell types do so many different things?
How do ‘levels’ of description relate given the vast range of spatial and temporal scales of available d
Why do the same brain areas ‘light up’ during vastly different tasks?
Does the architecture support the right kinds of interactions with its environment over the right time f
CHAPTER 1. THE SCIENCE OF COGNITION 29

This, in a nutshell, is the goal I’m striving towards. In the next chapter, I begin at
the bottom with neural mechanisms, so we can begin to work our way towards an
understanding of the neural basis of cognition.

1.5 Nengo: An introduction

This is the first of the 10 tutorials that are included in this book. There is one tuto-
rial at the end of each chapter, and each provides a instructions for constructing a
simulation related to a key concept encountered in that chapter. All of the tutorials
use the Nengo software package, which is actively supported by my lab. Tutorials
can be skipped without interrupting the flow of the rest of the book.
Nengo (Neural ENGineering Objects) is a graphical neural simulation envi-
ronment developed in my lab. In this section, I describe how to install and use
Nengo, and guide you through simulating a single neuron. The simulation you
will build is shown in Figure 1.2. Anything you must do is placed on a single
bulleted line. Surrounding text describes what you are doing in more detail. All
simulations can also simply be loaded through the File menu. They are in the
‘Building a Brain’ subdirectory.

• mention youtube flythrough/demos or other videos?

Installation
Nengo works on Mac, PC, and Linux machines and can be installed by download-
ing the software from http://nengo.ca/.

• To install, unzip the downloaded file where you want to run the application
from.

• In the unzipped directory, double-click ’Nengo’ to run the program.

An empty Nengo world appears. This is the main interface for graphical model
construction.

Building a model
The first model we will build is very simple: just a single neuron.
CHAPTER 1. THE SCIENCE OF COGNITION 30

• Right-click anywhere on the background and choose New Network. Set the
Name of the network to ‘A single neuron’ and click OK.
All models in Nengo are built inside ‘networks’. Inside a network you can put
more networks, and you can connect networks to each other. You can also put
other objects inside of networks, such as neural populations (which have single
neurons inside of them), and input and outputs which go to and from those neural
populations. For this model, you first need to create a neural population.
• Right-click inside the network you created and select Create new->NEFEnsemble.
In the dialog box that appears, you can set Name to ‘neuron’, Number of
nodes to 1, Dimensions to 1, Node Factory to ‘LIF Neuron’, and Radius to
1. Click OK.
A single neuron is now created. This leaky integrate-and-fire (LIF) neuron is
a simple, standard model of a spiking single neuron. It resides inside a neural
‘population’, even though there is only one neuron. To readjust your view, you
can zoom using the scroll wheel and drag the background to shift within the plane.
• To see neuron you created double-click on the ‘population’ you just created.
This shows a single neuron called ’node0’. To get details about this neu-
ron, you can right-click it and select Configure and look through the parameters,
though they won’t mean much without a good understanding of single cell mod-
els. In order to simulate input this neuron, you need to add another object to the
model that generates that input.
• Close the NEFEnsemble Viewer (with the single node in it) by clicking the
‘X’ in the upper right of that window.
• Right-click in the Network Viewer window and select Create new->Function
Input. Set Name to ‘input’, and Output Dimensions to 1. Click Set Func-
tions. In the drop down select Constant Function. Click Set. Set Value to
0.5. Click OK. Click OK. Click OK.
The object that will create a constant current to inject into the neuron has now
been created. It has a single output labeled ’origin’. To put that into the neuron,
you need to create a ‘termination’ (i.e., input) on the neuron.
• Right-click the ‘neuron’ population and select Add decoded termination.
Set Name to ‘input’, Weights Input Dim to 1, and tauPSC to 0.02. Click Set
Weights, double-click the value and set it to 1. Click OK.
CHAPTER 1. THE SCIENCE OF COGNITION 31

Figure 1.2: A single neuron Nengo model. This is a screen capture from Nengo
that shows the finished model described in this section.

This takes the input and puts it directly into the neuron without changing the value.
The tauPSC is a name for the ‘synaptic time constant’, which I will discuss in the
next chapter. It is measured in seconds, so 0.02 is equivalent to 20 milliseconds.
A new element will appear on the left side of the population. This is where you
can hook the input function to.

• Click and drag the ‘origin’ on the input function you created to the ‘input’
on the neuron population you created.

Congratulations, you’ve constructed your first neural model. Your screen should
look like Figure 1.2.

Running a model
You want to make sure your model works, so you need to use to the parts of
Nengo that let you ‘run’ the model. There are two ways to run models in Nengo.
The first is a ‘non-interactive mode’, which lets you put ‘probes’ into parts of the
network. You then run the network for some length of time, and those probes
gather data (such as when neurons are active, what their internal currents and
voltages are, etc.). You can then view that information later using the built in data
CHAPTER 1. THE SCIENCE OF COGNITION 32

viewer or with another program such as Matlab®. The other way to run models
is in ‘interactive mode’, which is more hands-on, so I will use it in most of the
examples in this book. An example of the running output in interactive mode for
this model is shown in Figure 1.3.

• Right-click on the Network Viewer background and select Interactive Plots.

This pulls up a new window with a simplified version of the model you made. It
has a ‘play’ button and other controls that let you run your model. First, you need
to show some of the information generated by the model.

• Right-click on the ‘neuron’ population and select spike raster.

A spike raster shows the times that the neurons in the population fire an action po-
tential spike (signifying a rapid change in its membrane voltage), which is largely
how neurons communicate with one another. You can drag the background, and
any of the objects around within this view to arrange them in a useful manner.

• Right-click on the ‘neuron’ population and select value.

This graph will show what effects this neuron will have on the current going into
a neuron that receives its output spikes shown in the raster. Each small angular
pulse is called a post-synaptic current or PSC. This is the current that is caused in
the neuron after the synapse by this neuron’s spike.

• Right-click ‘input’ and select control.

• Right-click on ‘input’ and select value.

This shows the value of your control input over time.

• Click the play arrow in the bottom right.

The simulation is now running. Because the neuron that you generated was ran-
domly chosen, it may or may not be active with the given input. Either way, you
should grab the slider control and move it up and down to see the effects of in-
creasing or decreasing input. Your neuron will either fire faster with more input
(an ‘on’ neuron) or it will fire faster with less input (an ‘off’ neuron). Figure 1.3
shows an on neuron with a fairly high firing threshold (just under 0.69 units of in-
put). All neurons have an input threshold below (or above for off neurons) which
they will not fire any spikes.
CHAPTER 1. THE SCIENCE OF COGNITION 33

Figure 1.3: Running the single neuron model. This is a screen capture from the
interactive plot mode of Nengo running a single neuron.

You can test the effects of the input and see if you have an on or off neuron,
and where the threshold is. You can change the neuron by pausing the simulation,
and returning to the main Nengo window. To randomly pick another neuron, do
the following:
• Right-click on the ‘neuron’ population and select Configure.
• Click on the grey rightward pointing arrow that is beside i neurons (int). It
will point down and i 1 will appear.
• Double-click on the 1 (it will highlight with blue), and hit Enter. Click
Done.
You can now return to the interactive plots and run your new neuron by hitting
play. Different neurons have different firing thresholds. As well, some are also
more responsive to the input than others. They are said to have higher sensitivity,
or ‘gain’. You can also try variations on this tutorial by using different neuron
models. Simply create another population with a single neuron and choose some-
thing other than ‘LIF neuron’ from the drop down menu. If you would like the
neuron to spike less regularly, you can add a noise generator to your LIF neu-
ron.11
11 Todo this, open the population, right-click the neuron you are using and select Configure.
Under noise right-click EMPTY and select Replace. Under PDF right-click EMPTY and select
CHAPTER 1. THE SCIENCE OF COGNITION 34

Congratulations, you have now built and run your first biologically plausible
neural simulation using Nengo. You can save and reload these simulations using
the File menu.

Replace. Select GaussianPDF and set the variance to 1.0. Click Create, then click OK, then set
frequency to 1000. Then click Create, then click OK, then click Done. You can increase the noise
by increasing the variance in the Configure->noise->NoiseImplPDF->PDF object.