An analyst looks at a chart and realises that two groups have been combined incorrectly. The correction feels immediate: the explanation that had seemed plausible a moment earlier no longer fits. Asked how they noticed, the analyst recalls a label, a familiar pattern, and a question that came to mind.
That account describes a thought and offers clues about how it arose. It is not a complete record of every perceptual and memory process involved. Nor does the incompleteness show that the thought was an illusion or that conscious reasoning contributed nothing.
Thought is a broad term covering activities such as remembering, imagining, judging, planning, and reasoning. Different studies define narrower targets so they can investigate them. There is no single operational definition that settles every question about mental processes and subjective experience. A useful explanation keeps those questions connected without pretending they have already become one solved problem.
Separate Content, Experience, Process, and Report
The content of the analyst’s thought concerns the groups in the chart. Its experience may include words, an image, a sense of recognition, or a combination. The processes involved may include perception, retrieval of knowledge, comparison, and attention. The later explanation is a report about that episode.
These descriptions are related, but they are not interchangeable. A person can accurately report thinking about a mistake without knowing which processes produced the recognition. A researcher can measure a response time without establishing exactly what the moment felt like.
A practical distinction is:
| Question | Possible evidence | What remains to investigate |
|---|---|---|
| What was the person considering? | A report, choice, or task response | Completeness and context of the report |
| How did performance change? | Accuracy, timing, errors | Which processes produced the difference |
| What neural activity accompanied it? | Brain recordings and analysis | Interpretation and causal role |
None of these methods gives unrestricted access to the others. Combining them can narrow explanations, provided the limits of each remain visible. A brain measure should not automatically override a report, and a persuasive report should not substitute for a performance test.
Thinking Does Not Always Take the Form of Inner Speech
Some thinking is experienced as words, while other episodes involve imagery, sensory qualities, or less easily described content. The fact that a person explains an idea in language afterwards does not establish that it first occurred as a sentence.
Hurlburt, Heavey, and Kelsey’s research on inner speaking uses detailed sampling and interviews to examine reported experience. The approach takes first-person description seriously while making its methods and interpretive demands explicit. It does not provide a universal inventory of every person’s mental life.
For the analyst, recognising the mistaken grouping might precede a verbal explanation. On another occasion, talking through the categories could be an important part of finding the problem. Both possibilities deserve investigation rather than assuming all reasoning is silent speech or all insight bypasses conscious activity.
This matters in collaborative work. Someone who needs to sketch a relationship may be reasoning effectively even before they can provide a concise verbal account. Someone who speaks fluently may still need to check whether the proposed relationship holds.
Models Explain Selected Features of the Process
A computational model describes transformations among specified representations or states. That can help explain performance, generate predictions, or compare alternative mechanisms. It does not automatically establish that a thought is literally a software object or a notification from an otherwise hidden system.
Computation itself need not be deterministic, reversible, or completely observable. Those properties depend on the system and model. Defining computation narrowly and then presenting thought as its opposite creates a contrast that does little explanatory work.
Models can also describe different levels. A learning model might predict how choices change with feedback, while a neural model investigates activity that could support the relevant computation. Their relationship needs evidence; similarity in terminology is not enough.
The predictive-processing article examines this problem with prediction and error. A model developed for one kind of processing can suggest questions elsewhere, but its reach should be tested rather than declared universal.
For the chart, a useful model might predict which labels make a grouping error easier to detect. That claim can be evaluated through performance. It does not need to settle why awareness has a subjective quality before contributing something valuable.
Consider two explanations for the correction. The analyst may have noticed an inconsistent label, or they may have recalled an earlier case with the same grouping problem. Those explanations suggest different comparisons: changing label clarity, varying relevant experience, or asking for a judgement before additional discussion. More than one process may contribute. The point of a model is to produce distinctions that evidence can examine, rather than simply rename the event as a state transition.
Introspection Is Informative but Not a Complete Process Log
People can describe what they are considering and why they believe a conclusion. Those descriptions are important evidence about experience and reasoning. They can also be incomplete, mistaken, or influenced by information obtained after the event.
In Johansson and colleagues’ choice-blindness experiment, some participants did not detect a manipulation between their choice and the outcome presented, and gave explanations for that outcome. The finding demonstrates a limitation under the tested conditions. It does not show that every explanation of a thought is invented or that people never know their reasons.
For the analyst, compare the later account with available records. The order of notes, visible chart versions, and questions asked at the time may support or challenge the reconstruction. The aim is a better explanation, not a presumption that first-person evidence is untrustworthy.
The storytelling article considers how accounts can make events understandable while simplifying their causes. Reports of reasoning can be useful narratives whose important claims still need checking.
Neural Decoding Provides Evidence Under Specific Conditions
Brain recordings can contain information related to what a participant sees, hears, imagines, or attempts to communicate. Researchers can train models to recover aspects of that information. The result depends on the task, recording method, training data, evaluation, and participant.
In Tang and colleagues’ 2023 language-decoding study, a system reconstructed aspects of meaning from non-invasive brain recordings under carefully controlled conditions. Individual training and cooperation were important. The generated output was not a general transcript of every private thought.
The distinction between prediction and explanation matters here too. A model can predict content-related features without establishing the complete mechanism of thinking or resolving the relationship between neural activity and conscious experience.
Ask what the evaluation compares. Does the output recover broad meaning, exact wording, a category, or a choice among limited alternatives? How does it perform on new material? These questions make a decoding result more informative than the unqualified claim that a machine can read minds.
Also distinguish information used for prediction from a demonstrated cause. A recording can contain a feature that tracks the task without showing that the feature alone produces the experience. Testing a causal account requires an appropriate research design and comparison. Predictive success is valuable evidence, but its interpretation should remain matched to the question the method was designed to answer.
They also show why privacy implications deserve specific analysis. Neither dismissing the research as meaningless nor assuming unrestricted access to mental life accurately describes what a particular method can do.
Keep Empirical Questions Separate From Philosophical Conclusions
Whether conscious experience can be fully explained in computational or biological terms remains a philosophical and scientific question. A behavioural experiment can constrain accounts of a task without resolving every issue about consciousness, agency, or meaning.
A delay between measurable activity and a report does not, by itself, establish that all conscious decisions are causally irrelevant. A person’s inability to describe every processing step likewise does not prove that the reported experience never influenced what happened next.
Treat explanatory metaphors as tools with a purpose. Describing thinking as a changing state can help highlight dependence on context and previous experience. It becomes misleading when the metaphor is presented as a settled discovery that thoughts are nothing but transitions or that experience is merely an after-the-fact alert.
The overthinking article applies similar care to repeated thought. Naming a network or process does not provide a complete explanation of the content or its significance for a person.
Study a Particular Thought Through Better Questions
For the analyst’s correction, ask what information was available, what changed in the interpretation, which comparison revealed the error, and how the conclusion was checked. A short record made near the event can help preserve that context.
Then test the practical claim. If the groups were combined incorrectly, the calculation and resulting decision should change in identifiable ways. Confidence and a vivid experience of insight are reasons to investigate, not substitutes for checking the data.
This approach leaves room for several legitimate descriptions: an experienced recognition, a change in the model of the problem, a pattern of cognitive activity, and a reported conscious event. Understanding how they relate requires evidence about the particular process and clarity about what remains unresolved. That is a more useful starting point than declaring thought either transparent to its owner or wholly illusory.
