Related materials: reification, anthropomorphism, teleologization, retrospectivization, consciousness as the capacity for distinction, ontological substitution in neuroscientific rhetoric.
Introduction
An explanation of a capacity must show the conditions and organization through which that capacity is exercised. Yet a description of a mechanism often introduces a new actor within it, already capable of doing precisely what the explanation was supposed to explain. Outwardly the explanation moves from the whole to a part, but logically it remains at its starting point.
The homuncular fallacy makes it possible to identify this lack of explanatory movement. Its subject is not the use of the words “brain,” “system,” “module,” or “agent” as such, but the hidden transfer of the capacity to be explained into the proposed mechanism.
Definition
The homuncular fallacy occurs when a mechanism introduced to explain a capacity of a subject or system is itself endowed with that same capacity without explaining how it operates.
Minimal structure:
a subject performs X
→ X is explained by an internal mechanism M
→ M also performs X
→ the origin of the capacity for X remains unexplained
The name derives from the image of a homunculus, a little person inside a person. But a literal little person is not required. Its role can be played by the brain, a particular brain region, a neural network, a module, a program, a representation, or any other part attributed a ready-made ability to see, understand, choose, or interpret.
Internal Image and Internal Observer
A classic example is constructed as follows:
a person sees an object because the brain constructs an internal image of it.
Constructing a particular internal state may be part of the mechanism of vision. But the mere presence of an image does not yet explain seeing. The question arises:
who or what sees that image?
If an internal observer is assumed, the explanation reproduces the original problem on a smaller scale:
external object
→ image in the brain
→ internal observer sees the image
It is now necessary to explain how the internal observer sees. If yet another image and observer are introduced inside it, an infinite regress begins.
The problem is not that internal states or representations are impossible. The problem arises when the word “representation” already presupposes someone to whom something is represented, without disclosing the position of that distinguishing subject.
Transferring Interpretation
The same error appears in the phrase:
the brain interprets a signal.
This may designate an established transformation of input: different signals produce different changes in states and behavior. In that case, “interprets” is used as functional shorthand that can be unpacked without invoking understanding.
But another construction is possible:
a subject understands the meaning of a signal because the brain understands what that signal means.
Here, the very capacity for distinguishing meaning that required explanation has already been placed inside the brain. Changing the grammatical subject creates the impression of a causal step, although the logical form has not changed:
subject distinguishes meaning
→ brain distinguishes meaning
→ neural network distinguishes meaning
→ module distinguishes meaning
The carrier becomes smaller, but the capacity remains ready-made at every level.
A Homunculus Without a Human Figure
A homuncular explanation need not sound anthropomorphic. It may be presented in technical language:
a central executive selects the relevant representation;
an evaluation system determines the significance of the input;
a conscious-access module decides which information to pass onward;
a recognition mechanism establishes what stands before the system.
Each formulation may be a legitimate name for a function. The error occurs if the name of the function is passed off as a description of its execution.
For example, “a recognition module recognizes a face” tells us where the researcher has placed the function in a model, but does not show how a distinction identifying the face is produced. Here the homunculus is not a humanlike creature but an opaque block that receives a task as input and outputs a ready-made result.
A Functional Name Is Not a Mechanism
It is convenient to name a component after the result of its work:
- decision-making system;
- attention mechanism;
- speech-comprehension module;
- error detector;
- memory storage.
But such a name moves from an observed capacity to its supposed carrier:
action X is observed
→ a part called “executor of X” is introduced
→ action X is explained by the existence of “executor of X”
If the component’s independent organization is not shown, circularity results. “The detector detects an error” may be shorthand for a procedure of comparison that has already been described. But if no criterion, comparison, or transformation is presented, the word “detector” merely hides the absence of an explanation.
Infinite Regress
The homuncular fallacy often leads to regress, but an explicit regress is not necessary. Sometimes reasoning stops at a word that a discipline accepts as sufficient:
who understands an utterance?
→ the subject
how does the subject understand?
→ the brain interprets meaning
how does the brain interpret?
→ that is a function of the neural mechanism of interpretation
The last line stops asking the question but does not answer it. The regress is hidden inside the functional label.
If the explanation continues in the same way, a chain results:
subject sees
→ brain sees an internal representation
→ subsystem reads the representation
→ neuron recognizes the signal
→ molecular mechanism determines its meaning
At each step the carrier becomes smaller, but the relationship to meaning, image, or choice remains without decomposition. The homunculus shrinks, while the unexplained capacity remains unchanged.
Not Every Decomposition Is Homuncular
Decomposing a capacity into subsystems is a normal research method. It contains no error when the capacities of the parts are simpler than the original capacity and their joint operation shows how the result arises.
For example, an explanation of sorting may involve comparison, swapping, and stopping operations. None of them must already know how to sort the entire array. Their organization produces the outcome that needed to be explained.
The distinction can be expressed as follows:
valid decomposition:
complex capacity X
→ simpler operations A, B, C
→ show how A, B, C jointly implement X
homuncular decomposition:
complex capacity X
→ internal part already capable of X
The words may nevertheless coincide. In engineering descriptions, “the agent chooses an action” can be an acceptable level of abstraction if the choice procedure is defined independently. Thus, the presence of a verb describing a subject’s action is a reason to examine an explanation, not sufficient proof of an error.
Homuncular Fallacy and Reification
These errors may accompany each other, but identify different transitions.
Reification occurs when a descriptive category becomes an entity, property, or cause:
observed dynamics
→ called “choice”
→ choice is placed inside the system as its property
The homuncular fallacy occurs when that internal “choice” is used to explain the capacity to choose:
system chooses
→ because an internal mechanism makes a choice
A category can be reified without constructing a homuncular explanation. Conversely, a homuncular fallacy can involve a genuinely existing part of a system if the capacity of the whole is attributed to that part without explanation.
Homuncular Fallacy and Anthropomorphism
Anthropomorphism transfers a human form of action to something nonhuman:
a neuron decides whether to transmit a signal.
The homuncular fallacy has a more specific logical structure:
a subject makes a decision because neurons make decisions.
The second sentence not only humanizes neurons but also transfers the capacity for decision that requires explanation to the level of the mechanism. Thus a statement may simultaneously be anthropomorphic, reifying, and homuncular, although the grounds for each classification differ.
reification:
a category became a property of an object
anthropomorphism:
an object acquired the form of human action
homuncular fallacy:
the capacity of the whole was attributed to its part without explanation
Where the Subject of Distinction Is Located
The error is especially significant in explanations of consciousness and meaning. If a subject is defined through the capacity to make distinctions, that capacity cannot be explained by introducing an internal mechanism that already distinguishes.
The phrase
the brain separates the significant from the insignificant
requires us to establish:
- relative to whom or what significance exists;
- which criterion produces the separation;
- whether the result is a physical difference between states or an experienced distinction of meaning;
- whether the observer’s criterion was imperceptibly placed inside the mechanism.
Physically different responses of a system to different inputs are not yet identical to a subject’s distinction of meaning. Claiming identity requires separate ontological grounds. Otherwise, the position of the researcher who sees the difference between outputs becomes an internal observer of the system.
How to Recognize the Error
Several questions are useful for examining an explanation:
- What exact capacity needs explaining?
- To which part or mechanism is it then attributed?
- Has the part’s capacity become simpler than the whole’s capacity?
- Is the structure of the operation shown, or merely its functional name?
- Can the action of the part be described without using a verb naming the original capacity?
- Will the proposed mechanism require another observer, interpreter, or decision-maker inside it?
- Has the researcher’s distinction been transferred into the system being researched?
A powerful diagnostic technique is to unpack the actor behind every verb:
the system represents an object — to whom?
the brain reads a signal — by means of what distinction?
the module evaluates the result — by whose criterion, or what criterion?
the mechanism chooses — between which alternatives, and how is the choice criterion formed?
If the answer again contains a hidden executor of the same capacity, the explanation remains homuncular.
How to Move Beyond a Homuncular Explanation
Replacing a subject-related verb with a technical term is not enough. One must either disclose how the operation works or honestly mark the explanatory boundary.
Instead of:
the brain recognizes a face,
one can describe an established dependency:
when images are presented, certain transformations of the input lead to stable differences in the system’s states and to different responses.
Such a description does not yet resolve the question of experienced recognition, but neither does it present that question as already resolved.
A complete explanation must show transitions between levels without inserting the result of the next level into the previous level as a ready-made capacity:
physical change
→ difference between system states
→ organization of behavior
→ subject's report of perception
If one needs to move from these dependencies to experience, understanding, or meaning, that transition must be formulated as a separate problem, rather than closed with the verb “interprets.”
Boundary of the Critique
Identifying a homuncular fallacy does not prove that the brain, neural networks, or internal representations do not participate in realizing a capacity. It establishes a narrower result: the proposed description has not yet explained the capacity if it has merely attributed it to an internal part.
Nor must every high-level description be unpacked down to its physical basis on every occasion. For a specific task, it is permissible to treat a functional block as given. What matters is not to conflate two assertions:
we provisionally treat function X as given at this level of the model;
we have explained the origin of capacity X.
The first may be methodologically sufficient. The second requires demonstrating an organization that does not presuppose X in ready-made form.
Brief Formula
The homuncular fallacy is explaining a capacity through an internal carrier to which that same capacity has already been attributed.
Its scheme:
the whole does X
→ because a part does X
→ the question of how X is possible has been moved inside the system
Valid decomposition does not make the subject smaller; it replaces the original capacity with simpler operations and shows how their organization produces the outcome being explained.
