Threshold Models

library(consciousnessModelR)

Purpose

This article explains the theoretical background behind consciousness_threshold(). Threshold models describe conscious access as a transition: below some level, processing may remain weak, local, unstable, or unavailable for report; above that level, information may become more stable, accessible, reportable, or globally available.

The purpose of this chapter is not to suggest that consciousness is literally a single numerical threshold. Rather, the purpose is to show how threshold logic appears in many theories of perception, attention, and conscious access.

The guiding question is:

What changes when a signal crosses from sub-threshold processing to access-like processing?

Why thresholds matter

Thresholds are common in models of perception and consciousness. A weak stimulus may influence behavior without being consciously reported. A stronger or more stable stimulus may become available for report, memory, reasoning, or action.

In Global Workspace Theory, this transition is often discussed in terms of ignition: a signal becomes sufficiently strong and stable to enter a global workspace and become broadly available (Baars 1988; Dehaene 2014). In perceptual research, related ideas appear in discussions of detection, confidence, visibility, and reportability (Sandberg et al. 2010).

Thresholds are therefore useful because they represent a theoretical boundary between different kinds of processing. They help distinguish between:

• weak processing and strong processing;
• local processing and global access;
• non-reportable influence and reportable awareness;
• unstable activation and stable availability.

However, threshold models must be used carefully. A threshold can clarify a theory, but it can also oversimplify consciousness if it is treated as a direct measure of subjective experience.

Thresholds as theoretical assumptions

A threshold is never neutral. Choosing a threshold means making an assumption about what level of activation counts as access-like.

A low threshold makes access easier. Many signals will be classified as above threshold. A high threshold makes access more selective. Fewer signals will qualify.

This matters because different theories may imply different access conditions. Some theories emphasize strong global broadcast. Others emphasize recurrent local processing, confidence, higher-order representation, or integrated information. Thresholds can be useful across these frameworks, but their interpretation depends on the theory being modeled.

In consciousnessModelR, a threshold should be understood as an educational proxy, not as a claim about the real boundary of consciousness.

Relation to the package

The function consciousness_threshold() takes a data frame with an activation-like column and adds threshold-related variables.

Theoretical concept	Package representation
Signal strength	Activation column
Access boundary	threshold
Above-threshold processing	above_threshold
Distance from access boundary	threshold_distance
Awareness-like classification	Activation greater than or equal to threshold

The function does not create consciousness. It classifies simulated activation values relative to a user-defined boundary.

Applying a threshold

The following example begins with a global workspace simulation. We then apply a threshold to the activation values.

sim <- simulate_global_workspace(
  n_processes = 8,
  steps = 100,
  ignition_threshold = 0.75,
  seed = 3
)

thresholded <- consciousness_threshold(
  sim,
  activation_col = "activation",
  threshold = 0.70
)

head(thresholded)
#>   step process activation winner is_winner broadcast ignited threshold
#> 1    1      P1  0.2713568     P2     FALSE 0.0000000    TRUE       0.7
#> 2    1      P2  0.9122034     P2      TRUE 0.9122034    TRUE       0.7
#> 3    1      P3  0.4384098     P2     FALSE 0.0000000    TRUE       0.7
#> 4    1      P4  0.3643226     P2     FALSE 0.0000000    TRUE       0.7
#> 5    1      P5  0.5750702     P2     FALSE 0.0000000    TRUE       0.7
#> 6    1      P6  0.6770616     P2     FALSE 0.0000000    TRUE       0.7
#>   above_threshold threshold_distance
#> 1           FALSE        -0.42864324
#> 2            TRUE         0.21220340
#> 3           FALSE        -0.26159021
#> 4           FALSE        -0.33567744
#> 5           FALSE        -0.12492979
#> 6           FALSE        -0.02293842

The output includes new columns showing whether each activation value is above the selected threshold and how far it is from the threshold.

Counting threshold crossings

table(thresholded$above_threshold)
#> 
#> FALSE  TRUE 
#>    11   789

This table counts how many simulated activation values fall below or above the threshold. In theoretical terms, these values can be interpreted as below-threshold and above-threshold processing events.

The result should not be interpreted as a count of conscious experiences. It is a count of simulated events that meet a simplified access criterion.

Comparing different thresholds

Threshold choice strongly affects interpretation.

low <- consciousness_threshold(sim, threshold = 0.50)
medium <- consciousness_threshold(sim, threshold = 0.70)
high <- consciousness_threshold(sim, threshold = 0.90)

c(
  low = sum(low$above_threshold),
  medium = sum(medium$above_threshold),
  high = sum(high$above_threshold)
)
#>    low medium   high 
#>    794    789    770

Interpretation of the comparison

A lower threshold produces more awareness-like events. A higher threshold produces fewer events.

This illustrates an important theoretical point: claims about access-like processing often depend on where the boundary is placed. If the threshold is permissive, many signals qualify. If the threshold is strict, only the strongest or most stable signals qualify.

For teaching, this is useful because it shows that threshold-based conclusions are sensitive to model assumptions.

Threshold distance

The column threshold_distance provides more information than a simple yes-or-no classification.

A value above zero means the activation is above threshold. A value below zero means the activation is below threshold. Values close to zero represent borderline cases.

head(thresholded[, c("step", "process", "activation", "threshold", "above_threshold", "threshold_distance")])
#>   step process activation threshold above_threshold threshold_distance
#> 1    1      P1  0.2713568       0.7           FALSE        -0.42864324
#> 2    1      P2  0.9122034       0.7            TRUE         0.21220340
#> 3    1      P3  0.4384098       0.7           FALSE        -0.26159021
#> 4    1      P4  0.3643226       0.7           FALSE        -0.33567744
#> 5    1      P5  0.5750702       0.7           FALSE        -0.12492979
#> 6    1      P6  0.6770616       0.7           FALSE        -0.02293842

This is useful because conscious access may not always be cleanly binary. Some perceptual states may be ambiguous, weak, partial, or unstable. A distance value helps represent the idea that some signals are close to the boundary.

Binary versus graded models

Threshold models often appear binary: a signal is either above threshold or below threshold. But consciousness research also includes graded views. On graded views, awareness may vary continuously in strength, confidence, clarity, or stability.

This creates a conceptual tension.

A binary threshold is useful for modeling reportability or access. A graded model may better represent partial awareness, uncertainty, or varying confidence. In practice, both views may be useful depending on the research question.

In consciousnessModelR, above_threshold gives a binary classification, while threshold_distance preserves a more graded interpretation.

Thresholds, confidence, and reportability

Thresholds are closely connected to reportability. A person may process a stimulus without being able to report it. As activation, confidence, or stability increases, the stimulus may become reportable.

However, reportability is not identical to consciousness. Some theories treat report as a useful behavioral indicator of conscious access. Others argue that report may reflect additional cognitive processes beyond experience itself.

This distinction matters because consciousness_threshold() models an access-like boundary, not subjective experience. Crossing a threshold means that a simulated activation value has met a formal criterion. It does not mean that experience has been produced.

Thresholds and global workspace models

In Global Workspace Theory, threshold logic is often tied to ignition. A signal becomes globally available when it becomes strong and stable enough to enter the workspace (Dehaene 2014).

The following example compares the model’s built-in ignition variable with a separate threshold classification.

gw <- simulate_global_workspace(
  n_processes = 8,
  steps = 100,
  ignition_threshold = 0.75,
  seed = 12
)

gw_thresholded <- consciousness_threshold(
  gw,
  activation_col = "activation",
  threshold = 0.75
)

table(
  ignited = gw_thresholded$ignited,
  above_threshold = gw_thresholded$above_threshold
)
#>        above_threshold
#> ignited FALSE TRUE
#>    TRUE    20  780

This comparison can help students see that different threshold rules may classify events in different ways depending on how the model is structured.

Relation to other functions

consciousness_threshold() can be applied whenever a simulation output contains an activation-like variable.

Related function	Use of threshold logic
simulate_global_workspace()	Classifies process activation relative to an access threshold
attention_competition_model()	Can classify priority values as above or below a selection threshold
broadcast_network()	Can classify node activation as locally or broadly available
simulate_information_integration()	Can classify component activation or integration scores if adapted
plot_consciousness_sim()	Visualizes activation patterns before or after thresholding

This makes consciousness_threshold() a general interpretive tool across the package.

Applying a threshold to a broadcast model

Threshold logic can also be applied to network activation.

net <- broadcast_network(
  n_nodes = 10,
  steps = 50,
  source_node = 1,
  seed = 4
)

broadcast_thresholded <- consciousness_threshold(
  net$time_series,
  activation_col = "activation",
  threshold = 0.40
)

head(broadcast_thresholded)
#>   step node activation source_node threshold above_threshold threshold_distance
#> 1    1   N1       0.90          N1       0.4            TRUE               0.50
#> 2    1   N2       0.25          N1       0.4           FALSE              -0.15
#> 3    1   N3       0.25          N1       0.4           FALSE              -0.15
#> 4    1   N4       0.00          N1       0.4           FALSE              -0.40
#> 5    1   N5       0.25          N1       0.4           FALSE              -0.15
#> 6    1   N6       0.00          N1       0.4           FALSE              -0.40

Here, above-threshold activation can be interpreted as a node becoming sufficiently activated by the broadcast signal. Again, this is an access-like classification, not a claim about experience.

What the model captures

The function captures several useful ideas:

• activation can be continuous while interpretation is categorical;
• thresholds represent assumptions about access;
• lower thresholds produce more access-like events;
• higher thresholds produce more selective classifications;
• distance from threshold helps represent borderline cases.

These features make the function useful for teaching how theoretical assumptions shape interpretation.

What the model does not capture

The function is intentionally simple. It does not model:

• subjective experience;
• perceptual confidence in detail;
• verbal report;
• metacognition;
• recurrent neural processing;
• higher-order representation;
• biological detection thresholds.

It also does not determine the true threshold of consciousness. There may be no single universal threshold.

Thresholds and caution

Thresholds are powerful because they simplify complex processes. They are also risky for the same reason.

A threshold can make a model easier to interpret, but it may hide important complexity. Real conscious perception may involve multiple interacting factors: attention, recurrence, confidence, memory, reportability, context, and system-level organization.

Therefore, threshold outputs should be described carefully. It is better to say:

The signal crossed an awareness-like access threshold in the model.

rather than:

The signal became conscious.

Educational use

This chapter supports several classroom questions:

• How does changing the threshold change the number of access-like events?
• Are borderline cases important?
• Is conscious access binary or graded?
• What is the difference between activation and reportability?
• Can threshold logic be used without reducing consciousness to a number?
• How does thresholding interact with global workspace or broadcast models?

These questions help learners understand thresholds as modeling assumptions rather than empirical facts.

Key takeaway

Threshold models describe conscious access as a transition. They are useful for representing the difference between weak/local processing and stronger/access-like processing.

consciousness_threshold() provides a simple educational tool for applying this idea to simulation outputs. It does not measure consciousness, but it helps clarify how threshold assumptions shape the interpretation of attention, activation, broadcast, and access.

References

Baars, Bernard J. 1988. A Cognitive Theory of Consciousness. Cambridge University Press.

Dehaene, Stanislas. 2014. Consciousness and the Brain. Viking.

Sandberg, Kristian, Bert Timmermans, Morten Overgaard, and Axel Cleeremans. 2010. “Measuring Consciousness: Is One Measure Better Than the Other?” Consciousness and Cognition 19 (4): 1069–78.