Exorcising the Homunculus
There’s no one behind the curtain
The following article is from Free
Inquiry magazine, Volume 21, Number 2.
Among the issues probed by philosophers, perhaps none
strikes closer to home than inquiries into the origins of human action.
How do our thoughtful decisions arise, and how are they translated into
overt words and deeds? Many of our actions appear to be produced
automatically, without deliberation. We need not focus on the precise
control of every muscle as we execute a practiced golf swing, and we
rarely find ourselves weighing each individual word choice when we utter a
request for a condiment during dinner. In the midst of such automatic
behaviors, decisions seem to be made for us, enacted by reflexes and
learned habits. There are some actions, however, for which we claim
authorship. We carefully weigh the options before selecting a particular
golf club from our bag, or we purposely edit an amusing tale that might
offend those with whom we dine. These controlled behaviors feel
more effortful to produce, and we sense that they stem from our conscious
The mental faculty responsible for the selection of
our controlled actions has traditionally been dubbed “the will.” From
the cogitava of Aquinas to the noumenally free will of Kant,
philosophers have posited this central faculty as the helm of the body and
the mind. The will is said to be where desires and goals are translated
into action and where ingrained habits are suppressed in favor of reasoned
plans. It has been heralded as the seat of moral thought.
By some accounts, the mind is modular, with the will
fairly isolated from other mental faculties. The sources of emotion, world
knowledge, memory, learning, and imagination are at the will’s disposal,
but are not part of it. Indeed it is thought the will can function, if
perhaps clumsily, without these other faculties. The will is also often
seen as essentially atomic—a fundamental aspect of the self that cannot
be decomposed into other psychological processes.
Unfortunately, asserting that such a will is the
source of our actions does little to advance our understanding; it merely
introduces a regress concerning the locus of decision making. We started
with the question of how the individual makes a choice, and we are
left with the question of how the will makes a choice. The will
takes on the character of a homunculus—a little man who resides in the
mind and acts as the “central executive” of the cognitive enterprise.
The homunculus seems to be the ruler of the mind, the maker of choices,
and the kernel of identity, but it is truly a useless hypothetical
construct that explains nothing about the origin of our actions. If we are
to understand our own controlled behavior, the homuncular will must be
exorcised from the mind.
That exorcism is currently under way, using the tools
of modern experimental psychology and cognitive neuroscience, but it is
not yet complete. While a detailed and substantiated account of the
mechanisms of controlled behavior is still forthcoming, mounting evidence
suggests that what has been traditionally called “the will” is neither
largely independent of other mental systems nor functionally localized and
atomic in the brain. Instead, current speculations see “executive
control”—the array of processes that, for many cognitive scientists,
replaces traditional notions of “the will”—as emerging from the
concerted interaction of disparate neural circuits.
While it is easy to introspect and speculate about
the varied sources of our thoughts, actually reverse engineering the brain
to determine its primary working parts is an incredibly difficult task.
The brain has about a hundred billion neurons, each connected to about a
thousand other cells; simply producing any sort of a wiring diagram is a
monumental project. Even more difficult is finding ways to watch the brain
in action. Still, scientists have produced a number of techniques, both
behavioral and neuroscientific, for collecting clues concerning the
architecture of the mind, and these have been used to discern the
structure of executive control.
Experiments have shown that controlled behaviors tend
to be slower than their more automatic counterparts, especially when one
is trying to inhibit an automatic response. The classic example of this is
the Stroop task, in which you are asked to name the color of the ink of a
printed word.1 You will tend to be slower to name the ink
color if the the printed word is the name of a color (e.g., “green”).
The idea is that executive control processes have difficulty overcoming
your predisposition simply to read the word that appears before your eyes.
Control is dynamic, allowing you to switch from one
behavior to another rapidly and flexibly. But this comes at a cost. You
will tend to be slow at a newly engaged task until you “get into the
groove” by performing it a bit. Also, the dynamic nature of executive
control makes control easy to lose: distractions can cause us to
slip into automatic patterns of behavior. Control processes seem to
require active maintenance of behavior-guiding information.2
Such a “working memory” system appears to be central to executive
control, providing the means to keep in mind what needs to be done.
Working memory is also critical to the reasoning and planning processes
required to consciously select actions—maintaining intermediate
conclusions and goals while the mental gears grind.
In the brain, the frontal lobes appear to be critical
for executive control. In addition to housing higher motor areas, the
frontal cortex appears to play a central role in working memory,
maintenance of goals, selective attention, and strategy generation.
Patients with frontal lobe damage exhibit a wide range of behavioral
deficits. They have difficulty with fine movements and complex chains of
motions. They show problems in flexible problem-solving, attention, and
memory. Frontal lesions can produce a tendency to persist with a pattern
of behavior after it has ceased to be appropriate, a lack of appreciation
for the risks involved in certain actions, and impaired learning. Frontal
patients have difficulty inhibiting automatic responses (e.g., they show
excessive Stroop interference), they have trouble maintaining mental focus
over a delay, and they spontaneously produce fewer actions, in general. In
short, the frontal cortex seems to be important for selecting actions
appropriately, producing flexible action sequences, keeping relevant
information in mind, and overcoming habitual behaviors.3
There is evidence to suggest that frontal systems
develop slowly as we mature and deteriorate with old age. Behavioral
experiments certainly support this lifetime pattern of performance on
executive control tasks. Children and seniors show increased difficulty
inhibiting automatic responses and flexibly switching between tasks.4
One might be tempted to see the frontal cortex as the
home of the traditional will, but reality seems to be less simple than
that. It is far from clear that frontal systems are capable of executive
control on their own. In addition, the frontal cortex appears to be
essential for certain psychological processes unrelated to executive
The frontal cortex appears to be important for
reasoning and problem solving, but other areas seem critical for these
higher functions, as well. Knowledge about the world, mediated in part by
circuits in the temporal lobes, is important for problem solving. Visual
and spatial reasoning rely on systems in the occipital and parietal lobes
of the brain. Still, the frontal cortex seems to play a central role in
such activities. Evidence for this includes the observation that
prefrontal activity is consistently prompted by a wide range of
standardized I.Q. test tasks.5
Traditional conceptions of the will assume that
reason and the will are tightly bonded, but posit a greater isolation
between the will and the emotional faculties. Emotion, after all, seems to
drive many of the kinds of automatic responses that executive control must
overcome. There is now reason to believe, however, that emotional systems
are critical for effective, reasoned decision-making. This evidence
primarily stems from patients with ventromedial frontal lobe damage. These
patients score well on standard I.Q. tests, but show marked deficits in
practical reasoning skills. They can become stuck on even simple problems,
endlessly exploring the options without ever coming to a conclusion. They
behave as if they are insensitive to risks, even though they can
intelligently discuss such risks. Their reasoning seems to be decoupled
from their actions. Importantly, these patients also show a pronounced
absence of physiological responses to emotional situations (e.g., sweating
or increased heart rate). These cases have caused some neuroscientists to
suggest that healthy emotional responding plays a critical role in guiding
reasoning processes and in connecting deliberation to action.6
Emotional systems allow conclusions to be reached in a timely manner, and
they give conviction to our thoughts. Thus, counter to traditional
accounts, emotion appears to be critical for executive control. The
armchair philosopher might argue that, since we can conceive of a rational
decision-maker free of emotion, the will must be conceptually independent
of emotional systems. The brain does not seem to be convinced by this
Psychological processes not primarily identified with
executive control also seem critically dependent on the frontal cortex.
For example, various kinds of learning and memory seem to require frontal
support. While memory for recent autobiographical events is often
associated with the medial temporal lobe, both brain imaging studies and
work with patients have suggested that the frontal cortex actively
contributes to storage and retrieval of such memories.7
Frontal working memory systems also appear to be important for explicit
forms of learning, such as learning from direct instruction.8
Components of Will
In order to banish the homunculus, we must find a way
to reduce executive control functions to a collection of more simple
psychological mechanisms. While efforts to decompose the circuits of
controlled behavior are just beginning, some early findings and
speculations are both fascinating and promising.
Some regions of the frontal cortex contain an
abundance of cells that are richly interconnected. Computer modeling has
shown that such a configuration of connectivity often results in the
active maintenance of patterns of neural firing over time.9
In other words, these recurrent connections may form the basis of a
working memory system. Neural recordings in nonhuman primates have found
such sustained activity in frontal cells, with the activity coding for the
properties of recent events that are important for guiding behavior.10
For example, a monkey might be trained to note where a small dot
flashes on a screen, only shifting its gaze to that location after a bell
sounds. While waiting for the bell, some of these frontal cells actively
encode the location of the previously presented dot, holding the location
in mind until it is time to act.
Neurons in the frontal cortex send connections to
areas throughout the brain. Thus, the “control signals” held in a
frontal working memory system are in a good position to modulate
processing in a fairly global way: focusing attention in perceptual
systems, priming responses in motor systems, and inhibiting inappropriate
automatic processes. Some mechanism is needed, however, to determine the
kind of control information that should be actively maintained. While
there is some evidence to suggest that certain areas of the frontal cortex
play a role in the adjustment and manipulation of working memory contents,
another candidate for this role lies deep in the brain.
Hiding far beneath the folds of the cerebral cortex,
the basal ganglia receive extensive projections from the frontal cortex
and, through the thalamus, send extensive connections back. Certain cells
in this limbic area can learn to identify events and actions that are
predictive of reward.11 For example, if food consistently
follows a ringing bell, these neurons will initially fire when food is
presented, but will, over time, stop firing for the food and start firing
for the bell. Computer scientists have produced model neural circuits that
are capable of learning to predict reward in this way and, critically,
these models can use such predictions to learn to select appropriate
actions for complex tasks.12 (A world-class backgammon playing
program sold by IBM learned its strategy using a neural network model of
this kind.13) Through its connections with the frontal
cortex, the basal ganglia may also learn to identify the properties of
working memory contents that are predictive of reward and may, thus,
signal frontal systems to actively maintain exactly those control signals
that are needed for success.18
It is interesting to note that the basal ganglia are
also thought to be involved in the basic initiation of actions and in the
coordination of action sequences. Parkinson’s disease selectively
affects this limbic structure, and patients stricken with this ailment
have trouble starting motions and switching between well-learned motor
sequences. Thus, it is likely that the basal ganglia are involved in both
automatic and controlled aspects of the motor system.
Another brain area of importance for executive
control is the anterior cingulate cortex. Nestled in the deep fold that
separates the brain’s hemispheres, the cingulate appears most active in
situations in which there is substantial conflict between competing
options—situations where the brain is having a hard time settling on one
action or another. Scalp electric field measurements have detected a
signal from roughly this area that appears when a behavioral error is
made. This error-related negativity appears even when subjects are
not told they have made an error but, instead, realize it themselves.19
It is not clear if this cingulate activity is the result of reward
prediction information sent there from the basal ganglia or if it is an
independent measure of task difficulty.20 In either case,
such performance-monitoring information would be useful for selecting both
actions and working-memory contents. Some theories see the anterior
cingulate as the locus of action-selection processes, modulated by frontal
systems. Other theories see the frontal cortex as more central. In either
case, the anterior cingulate cortex and the frontal areas work together to
suppress automatic responses in favor of explicitly selected actions.
What emerges is a general story in which the frontal
cortex sends actively maintained control signals to much of the rest of
the brain. The nature of these signals is selected primarily by circuits
in the limbic system, based on predictions of reward. The control signals
maintained in working memory, along with conflict-related processing in
the anterior cingulate, give rise to the selection of appropriate actions
for the current situation.
Cognitive neuroscientists have begun breaking down
executive control into its functional parts. Much of this research is
still speculative, and detailed accounts of such processes as strategy
generation and complex action planning have yet to receive much attention.
There is still much work to be done.
The Power of the Brain Compels You
The traditional view of the will as a kind of little
man in your head needs to be replaced by a detailed account of how neural
tissue gives rise to controlled behavior. Preliminary attempts to
understand the mechanisms of executive control have found that they do not
form an isolated psychological faculty, but are heavily dependent on other
psychological processes, including emotional response. Initial attempts to
dissect the mental executive have identified critical roles for a frontal
working memory system and a limbic reward-prediction system. The
scientific exorcism of the homunculus continues, hoping to produce a clear
view of how mere flesh can give rise to our most deliberate and considered
1. J. Ridley Stroop, "Studies of Interference in Serial Verbal Reactions," Journal of Experimental Psychology 28(1935): 643-62.
2. Jonathan D. Cohen, Todd S. Braver, and Randall C. O'Reilly, "A Computational Approach to Prefrontal Cortex, Cognitive Control, and Schizophrenia: Recent Developments and Current Challenges," Philosophical Transactions of the Royal Society of London B 351 (1996): 1515-27.
3. Angela C. Roberts, Trevor W. Robbins, and Larry Weiskrantz, eds., The Prefrontal Cortex: Executive and Cognitive Functions (Oxford: Oxford University Press, 1998); and Mark Wheeler, Donald T. Stuss, and Endel Tulving, "Frontal Lobes and Memory Impairment," Journal of the International Neuropsychological Society 1(1995): 525-36.
4. Norman A. Krasnegor, G. Reid Lyon, and Patricia S. Goldman-Rakic, eds., Development of the Prefrontal Cortex: Evolution, Neurobiology, and Behavior (Baltimore: Brookes Publishing, 1997).
5. John Duncan, Rudiger J. Seitz, Jonathan Kolodny, Daniel Bor, Hans Herzog, Ayesha Ahmed, Fiona N. Newell, and Hazel Emslie, "A Neural Basis for General Intelligence," Science 289, 5478 (2000): 457-60.
6. Antonio R. Damasio, Descartes' Error: Emotion, Reason, and the Human Brain (New York: G. P. Putnam, 1994).
7. Lars Nyberg, Endel Tulving, Reza Habib, Lars-Goran Nilsson, Shitij Kapur, Sylvain Houle, Roberto E.L. Cabeza, and Anthony Randal McIntosh, "Functional Brain Maps of Retrieval Mode and Recovery of Episodic Information," NeuroReport 7(1995): 249-52.
8. David C. Noelle, "A Connectionist Model of Instructed Learning," Ph.D. thesis, University of California at San Diego, Department of Computer Science and Engineering, Department of Cognitive Science, 1997.
9. Daniel J. Amit, Modeling Brain Function: The World of Attractor Neural Networks (Cambridge, U.K.: Cambridge University Press, 1989).
10. Earl K. Miller and Jonathan D. Cohen, "An Integrative Theory of Prefrontal Cortex Function," Annual Review of Neuroscience, in press.
11. Wolfram Schultz, Paul Apicella, and Tomas Ljungberg, "Responses of Monkey Dopamine Neurons to Reward and Conditioned Stimuli During Successive Steps of Learning a Delayed Response Task," Journal of Neuroscience 13(1993): 900-13.
12. P. Read Montague, Peter Dayan, and Terrence J. Sejnowski, "A Framework for Mesencephalic Dopamine Systems based on Predictive Hebbian Learning," Journal of Neuroscience 16(1996): 1936-47.
13. Gerald Tesauro, "Temporal Difference Learning and TD-Gammon," Communications of the ACM 38(3) (1995).
18. Todd S. Braver and Jonathan D. Cohen, "On the Control of Control: The Role of Dopamine in Regulating Prefrontal Function and Working Memory," in S. Monsell and J. Driver, eds., Attention and Performance XVII (Cambridge, Mass.: MIT Press, 2000).
19. William J. Gehring, B. Goss, Michael G. H. Coles, David E. Meyer, and Emanuel Donchin, "A Neural System for Error Detection and Compensation," Psychological Science 4(1993): 385-90.
20. Matthew Botvinick, Leigh Nystrom, Kate Fissell, Cameron S. Carter, and Jonathan D. Cohen, "Conflict Monitoring Versus Selection-for-Action in Anterior Cingulate Cortex," Nature 402(1999): 179-81; and Clay Holroyd, Jesse Reichler, and Michael G.H. Coles, "Is the Error-Related Negativity Generated by a Dopaminergic Error Signal for Reinforcement Learning? Hypothesis and Model," in Cognitive Neuroscience Society Annual Meeting Program, p. 45, Washington, D.C., 1999.
Noelle is a postdoctoral research associate at the Center
for the Neural Basis of Cognition, a joint project of Carnegie
Mellon University and the University of