00The writing and the reading proposed for it
This page contains the words “human”, “AI”, “AGI”, “I”, “understanding” and “right”. Each enters the document as a written form. Explanations, examples and rules for applying these forms follow. Adding those words does not make the writing become what it names. The same applies to “written form”, “explanation” and “rule”, the terms used to construct this examination.
The proposal begins with this requirement: do not present a description in place of what it describes, or require anyone to accept that they are the written description. Stated grounds may support a particular use of a record; they do not turn the record into what it describes. “Grounds”, “use” and “description” also require examination through the wording shown.
The question therefore continues after a definition, report or rule has been supplied. A reading remains a proposed reading even when it has been explained in detail. A demand that someone accept that they are the written description is examined separately from a requirement to use particular words. This provision is itself a written proposal, developed through the examples below; its appearance does not enact a rule.
The central question is this: on what grounds is a learned form proposed for describing oneself, read as a description of someone else, and used in making demands? The question remains when the same words appear in an AI response. A label identifying the author changes the role attributed to the response; the label itself does not supply the understanding, feeling or authority named.
The document examines, in sequence, learning to use labels, self-description, descriptions of others, assessments of a system, and decisions based on a description. “Learning”, “assessment” and “decision” receive specific examples so that the line under examination and the additions to it can be identified. Every example marked as constructed was prepared for this examination and does not report an observation.
The proposed requirement for applying descriptions is to present, alongside a description, the initial material, the added reading, its grounds, the rule for applying it, and a way to challenge the step from description to decision. This is a written proposal for discussion. It is subject to the same examination as any rule it proposes to consider.
01What is added to a written form
trust She began to trust. This will happen.
The lines are presented separately. The second and third are not explanations already contained in the first. These English examples are adapted from the Russian edition.
The first line says “trust”. The second adds “She” and “began to”: it proposes an attribution to someone and a report of an onset. “A feeling of trust” would offer a further explanation. Neither the initial word nor that explanation becomes the feeling named. Examining the report more closely would require an account of whom “She” refers to, what material is supplied, and how it is read.
The third line says “This will happen.” It could be read as a promise, prediction, threat or continuation of a story. These labels add different frames to the same form. Selecting a frame calls for the context and the grounds for that selection to be supplied. The written words “will happen” do not present the event they name; writing “promise” does not fulfil what was promised.
This examination continues through the words used to explain. “On the basis of experience” adds “basis” and “experience”. Examining that explanation requires the specific material proposed for use under those names. A further sentence, “this has been confirmed”, would add an assessment and raise the same question about the check being offered.
A useful unit of examination is an exact passage and a particular addition to it. For example, “She began to” is added to “trust”; a reference to correspondence is added to that report; a selected reply is then read as grounds for an assessment. Each step can be presented separately. Here, “passage” and “step” name parts of this comparison; they do not exempt it from the boundary being examined.
t h i n k think f e e l feel
The spaced and joined spellings are shown separately. Joining the letters does not present what the words are offered as naming. These examples are adapted from the Russian edition.
Adding “this happens inside” would add another record. “happens” and “inside” are written too; “research confirms this” would add a claim and a reference to material that would need to be supplied. Neither that explanation nor an examination of the material supplied turns the written words “think” and “feel” into what they describe. “Material” and “examination” also belong to the explanation; an attached file is still examined in its own format.
“Stop believing in the description” itself introduces “believing” and “Stop”. This proposal calls for examination of a specific use of a record and any demand to accept that one is the written description. A report that “the belief has stopped” would also require examination; it does not present the cessation it names. The provision does not require one compulsory self-description to be replaced with another.
02Learning to speak and learning to write
Adult: “Say: I.” Child: “I.” Explanation: “Now the child knows who they are.”
Three constructed lines are presented here. “Adult”, “Child”, “Say” and “knows” occur in them as written forms. “Teaching scene” is also a label applied to the lines; that label does not supply the teaching it names.
The first two lines assign utterances to labelled roles. The last adds a claim about knowing oneself. The repetition of “I” in the record does not present that knowledge. Discussing a particular instance of teaching would require separate material: what was to be said, what response was received, what explanation was added, and what supports the conclusion drawn from it.
An account of learning to speak must also be distinguished from an account of learning to write. The written utterance “Say: I” uses letters to represent an imagined act of speaking. It does not establish that the scene included displaying the letter “I”, teaching its shape, or reading a printed instruction. An invitation to speak and an invitation to write each require their own material for examination.
An explanation can introduce “audio”: a u d i o. Writing “audio” does not play a sound. In that sentence, “sound”—s o u n d—is also written. Explaining “sound” as “something heard” introduces further written wording; “heard” does not take the explanation beyond that boundary. The same examination applies to “video recording” and “an observed case”: writing those expressions here does not supply what they name. The English spellings and explanatory wording are adapted for this edition.
A further question follows: what, specifically, was to be learned? To repeat a form, use it in particular situations, describe oneself with it, or accept the description as an exhaustive account of who one is? These formulations state different demands. An account of learning a word does not, by itself, establish which demand was made. If it reports a demand to accept that one is the written description, that report also becomes part of the examination, including its exact wording and the circumstances supplied.
03When you are asked to accept that the description is you
I am human. I feel fear. This is who I am.
“Self-description” is the label proposed for examining these lines; it does not establish their source, recipient or the state named.
The first line joins “I” and “human”. The second adds “feel fear”. The third proposes reading “This” as “who I am”. Each line can be examined separately: what is the form applied to, what material supports that application, and how is the resulting description then used? “human” and “fear” gain no special status from being placed beside “I”.
Using a line and being required to accept that one is the written description must be kept separate. A form, conversation or teaching example may contain different requirements. The table below compares constructed formulations; it does not report an organization's current procedure.
| Constructed requirement | What is being requested |
|---|---|
| “Enter this word in the field.” | Use the specified written form. |
| “In this table, select the category ‘human’.” | Apply a category within the stated table. |
| “Accept that you are this written description.” | Accept that the person described and the written description are one and the same. |
The final formulation adds a requirement absent from the first two: to accept that the person described and the written description are one and the same. Here, “one and the same” states the relation demanded in the constructed example. It does not present anyone's inward agreement. If “consent obtained” is subsequently recorded, the examination will need to identify which response or procedure is called consent, and how far that reading has been extended.
This opens a question for public discussion: who proposes a description, which forms of response are available, whether the category may be qualified or rejected, and what consequences are announced. The answers will also be records. They can be compared with the stated requirement without making either acceptance of a category or refusal of it an exhaustive account of the person to whom it is applied.
04Using a description of someone else to justify a requirement
Answer: “8”. Conclusion: “Anna does not understand the task.” Decision: “Do not allow progression to the next stage.”
The name, answer, conclusion and decision were prepared for this example. The three lines do not include the task.
The first line contains an answer. The next adds the name “Anna”, “does not understand”, and a connection to a task. The last proposes restricting progression. Two steps appear between the lines: from answer to description, and from description to decision. The order of the lines does not disclose the grounds for either step.
Adding the task “What is 3 + 4?” and the rule for ordinary integer addition provides grounds for comparing “8” with the result “7”. That comparison supports a particular assessment of the answer under the stated conditions. The broader “Anna does not understand the task” requires an account of what is being attributed through “does not understand” and why the selected answer is sufficient for that conclusion.
Refusing progression raises a further question: which rule authorizes that particular decision? A condition for moving between stages, a procedure for another attempt, and a record of applying the rule can be supplied. Those additions extend the material under examination. They do not make the conclusion identical to the understanding it describes, or retrospectively insert the rule into the initial answer.
The same sequence can be applied to the constructed phrases “he is dangerous”, “she agrees” and “the system is ready”. Each contains an attribution of its own; each subsequent requirement needs separate examination. Calling a description “objective” adds another characterization. That characterization also calls for the chosen conditions of comparison and assessment material to be shown.
In this document, “consequence” names a stated result of applying a requirement or decision. Discussing a particular consequence requires the corresponding material and an account of its connection to the use of the record. The word “consequence” does not perform that examination.
05The same line under different labels
| Label in the constructed example | Line |
|---|---|
| HUMAN | I understand. |
| AI | I understand. |
The right column repeats one written form. The left supplies different labels. Reading one line as a report of understanding and the other as an imitation of such a report adds two different attributions to the examination. Each requires its grounds to be disclosed. Repeated letters establish neither identical nor different states.
“HUMAN” and “AI” are also written. Examining particular responses requires the grounds for assigning those labels: what information about the source is supplied, and how it is connected to the response. The origin of the record, performance on a task and the chosen description of the result can then be considered separately. A label alone does not settle these questions.
The example suggests a simple exercise: first preserve the lines without labels, then add the labels and record which explanation changes. This provides material for comparing one's own written responses. A conclusion about what a participant felt or experienced would be a separate addition. The exercise is not presented here as an experiment that has been conducted.
“The person learned a language” and “the model was trained” also require separate examination. “Learning” or “training” in an explanation does not, by itself, supply the conditions, sequence or results of the process named. Comparison requires specific material for each case and an explicit question to be compared. A shared label is insufficient for attributing an identical mechanism; different labels are insufficient for establishing every difference.
The common requirement therefore concerns justification: every attribution must show its grounds. It does not propose making the participants described equivalent in advance. The same requirement to explain an application extends to this paragraph's own words “participant”, “attribution” and “grounds”.
06“Audio”, “video recording” and descriptions of material
“Audio”, “sound”, “video recording”, “image” and “an observed case” are written expressions here. Explaining “audio” through “sound” introduces another written term. In “sound can be heard”, “can” and “heard” also form part of the explanation; those words do not supply the hearing they describe. A reference to “observation” continues this examination rather than completing it.
| Written label for material | Constructed addition | What needs to be disclosed |
|---|---|---|
| “Audio file” | “The speaker is experiencing fear.” | Which attachment is given that label, which excerpts are used, and under what rule a description of fear is added to them. |
| “Image” | “This is how others see him.” | Who is called ‘others’, what material is supplied, and how the general conclusion is reached. |
| “Table of values” | “The system understands the person.” | How the values were obtained, and why the chosen assessment is called a test of understanding. |
The left column contains written labels. Writing “Audio file” does not attach a file, and “Table of values” does not supply any values. In the right column, “attachment”, “excerpts” and “rule” indicate what the proposal asks to be disclosed in a particular examination; those words likewise do not supply what they name merely by being written.
For work with supplied material, the proposal is to retain access to the attachment itself and present its caption, transcription and any conclusion separately. The attachment, its playback and a written report of that playback are not substitutes for one another. “Attachment” and “playback” are terms used here to describe a procedure; a particular examination needs to identify the file, the operation performed and the grounds for the next attribution. A written explanation does not turn every kind of material under examination into writing.
Under the proposed procedure, a transcription is prepared by selecting utterance boundaries and spellings; a caption selects features; a table specifies units and columns. The source file and the resulting record are retained separately, with an account of how the record was produced. Producing such a record does not make it become the material it describes. Calling a record “accurate” adds an assessment of correspondence; “correspondence” and the conditions of that assessment also require examination.
An expert opinion may then be added. “Expert” adds a role, while the opinion adds attributions and conclusions. Examining them requires information about authority, procedure and scope. Even a fully explained opinion does not become what it names. Adding material preserves that boundary; the boundary does not declare the material unnecessary or replace the work of examining it.
07Explanatory words receive no exemption
“The system understands.” “Grounds: it answered correctly.” “Correctly: the answer matched the reference answer.”
The last line states a possible rule for using “correctly”. It does not supply the answers, the reference answer or a completed comparison.
Here, “understands” is explained through “correctly”, and “correctly” through a match with a reference answer. The examination must show all three records. Supplying the answers and reference answer would allow the stated comparison to be performed. Why its result is used to support the first claim can then be examined separately. Defining an intermediate word does not automatically complete the entire connection.
Here, the label “Grounds” names what is offered in support of a chosen conclusion. It proposes a connection between records and material. That connection can be specified: which exact passage supports which claim, under what conditions, and what would change if the passage were replaced. Writing “grounds exist” does not substitute for that explanation.
The same work continues through “proof”, “verification”, “testimony”, “consent” and “authority”. Additional material may be supplied under each name. Its contents and proposed application can then be examined. The name itself does not give a description authority to certify everything to which it has been applied. A reference to beforeword remains subject to the same conditions: the project's name grants no exemption to a formulation it proposes.
A practical examination can end with the stated task: for example, two files have been compared byte by byte and the result recorded. That limit should be stated alongside the procedure performed. A broader claim about the files' meaning would require a further examination. Completing a check does not place its concluding line beyond writing.
“Beyond the text” and “an outside view” may also appear in an explanation. What has then been presented is those expressions. Examining the position they attribute requires the material and method of attribution to be identified. The subject of examination is the specific place where “an outside view” is named and used in an argument, including this explanatory sentence.
08What a criterion checks, and what is claimed afterwards
“Check” and “criterion” enter this section as written terms. Applying them requires a condition, material and a method of comparison. The presence of PASS does not yet say what was compared or which conclusion it is intended to support.
{"answer": 4}
PASS
The answer is correct.The example assumes a check of whether the record conforms to JSON format. It does not report a model test that has been conducted.
The first line is a record containing the field answer and the number 4. The second contains the status PASS; the third offers an assessment of the answer. Connecting the first and second requires the format being checked and the procedure. The third requires, at minimum, the question being answered and the assessment rule. No question is supplied in this example. Checking the structure of a record does not substitute for examining its use as an answer.
A condition can be stated explicitly: “PASS means that the record conforms to the specified format.” This supplies a rule for using the label. It does not supply every check that another document might call “correctness”. Comparing two uses requires both definitions and the material showing their application to be preserved. “Format” likewise does not substitute for the written rules.
“Safe”, “ready” and “reliable” require the same disclosure. For example, “no failure of the specified type was logged in this run” concerns a particular run and method of logging. Adding “the system is safe” extends the description. That extension requires the scope, conditions considered and grounds for the step to be shown. “Run” and “failure” need to be explained through the material supplied.
The recorded criterion, result and conclusion are preserved separately. Comparing them may support a limited conclusion under the stated rules; it does not make the words of that conclusion become the capability named. Subsequent permission to use a system for a particular task requires a separate decision rule. The assessment alone does not supply it.
09A proposal concerning rules, requirements and decisions
The sequence examined leads to a specific proposal: keep the record, its reading, the grounds supplied and the rule for applying it distinguishable. Here, “distinguishable” means that each part can be presented separately and its connection to the others identified. If one part is replaced by the next without explanation, that step becomes the specific point to challenge.
For self-description, the proposal is to disclose which form must be used, in which procedure, and with what stated consequences. A requirement to complete a field should be shown separately from a requirement to accept that one is the written description. An account of such a demand must preserve its exact wording and the material on which the account relies.
For AI, the proposal is to disclose which words describe the system, how those words are connected to the tests supplied, and which actions are permitted on the basis of the result. A system's response, a developer's assessment, an evaluator's report and a decision to use the system must be preserved separately. The role labels likewise require stated grounds when they are cited as a source of authority.
A written rule receives no exemption. If a document granting authority is cited as grounds, the exact passage, scope, proposed interpretation and connection to the decision under examination should be shown. “Right” in a conclusion does not replace that work. A document's role in a procedure does not make its wording become what it describes.
The proposed procedure is itself open to challenge. An objection may identify an inaccurately preserved record, omitted material, unsupported attribution, a rule unsuited to the task, or a conclusion extended too far. Supplying the exact passage and explaining what changes in the examination makes such an objection specific. Agreement with beforeword is not proposed as a condition for considering it.
Even a justified description does not become what it names. This line is also written: its use in the document is shown through particular comparisons, not certified by the project's name. The following sections develop the proposed requirements for public discussion and specific procedures.
10What is added to a response labelled AI
“AI” enters this section as a written label. In an interface, it may appear above a response; in documentation, within a definition; in advertising, as a name for advertised capabilities. Each use calls for the particular passage to be preserved. Carrying a description from an advertisement into an analysis of a response is a separate step, even when both use the label “AI”.
I understand that you are scared. The system detected the user's fear. The response demonstrates empathy.
The second and third lines were composed as different explanations of the first. They do not report the results of a test conducted here.
The first line already contains “I”, “understand”, “you” and “scared”. The explanations add “system”, “detected”, “user's” and “empathy”, and propose connections between these words. Before assessing “empathy”, the proposed use of that label needs to be stated: a style of response, a task result, a description of the recipient, or an attributed state. Selecting one of these readings does not silently establish the others.
A prompt, the available context, settings and a sequence of responses can be supplied. The comparison then concerns that material: which phrases appeared in the input, which appeared in the response, and which rule was used to assess them. The word “settings” does not substitute for the settings themselves; “log” and “context” do not supply records that remain unavailable. Any unavailable portion should be identified as unavailable for this examination, rather than filled in by conjecture.
A specific interface requirement can therefore be proposed: an explanation such as “fear detected” should be accompanied by the material from which it was derived and a description of the chosen criterion. The response “I understand” is preserved separately. If a score concerns the style of the response, the interface should say which features of that style were assessed. Here, “style” and “feature” name items in a proposed procedure: the features must be listed, and their application to the line must be shown.
The self-description “I do not have feelings” receives the same treatment. The line contains a negation and the word “feelings”; its negative form grants it no exemption. A reference to the system's design adds a technical description and material for examining it. The written technical explanation does not become the design it describes.
11What accompanies the label AGI
“AGI” consists of three letters in the record presented here. Expanding the abbreviation to “artificial general intelligence” adds English words; translating it adds words in another language. Neither expansion nor translation, by itself, specifies a proposed test or explains how a result is connected to the label.
The OpenAI Charter's definition of AGI refers to highly autonomous systems and to outperforming humans at most economically valuable work. This is a paraphrase of the organization's wording; a short exact excerpt and a link appear in source S4. “Autonomy”, “humans”, “outperforming”, “work” and “value” are not treated here as properties already demonstrated. Applying such a definition calls for the selected tasks, the results being compared, the measure and threshold, and the method of combining results.
Article 3 of the AI Act introduces a different category: a general-purpose AI model, or GPAI. Its definition uses wording about generality and a wide range of tasks. This category is examined through the text of that regulation. Equating GPAI with AGI would require an explicit comparison of the definitions; their shared English words do not perform that comparison. Source S1 identifies the passage and the question concerning its application.
90 of the 100 tasks in this set were completed. AGI has been achieved.
The figures and announcement were composed for this examination. They do not report a test result for any model.
The first line gives only a brief description of a result. It needs the tasks, conditions, responses and scoring rule alongside it. The second introduces the label AGI and a claim of achievement. Connecting the two requires the chosen definition and an account of how it was applied. Even after that comparison, “AGI has been achieved” remains a written conclusion. It does not become what those words propose to name.
For a public announcement, the proposed disclosures are: whose definition was used; which tasks were tested; what was compared with what; which prompts, tools and conditions were available; what remained outside the test; and what specific use of the result is proposed. “Human level” likewise requires an account of the chosen comparison. A group's average result does not provide a description of everyone referred to as “human”.
The procedure also covers a decision to use the system. A label assigned under chosen criteria does not substitute for a description of permitted actions, access and restrictions. “Access”, “restriction” and “permission” must be explained through the stated conditions and records of their application. Declaring those words sufficient grounds would introduce another unexplained step.
12Reading regulatory documents
A regulation, a recommendation and an organization's declaration are presented here as separate documents. Their names indicate proposed roles for the texts. A specific legal application would require the relevant document, edition, scope and procedure. This publication selects passages for examining their wording; it does not determine whether a provision applies to anyone's case.
Article 14 of the AI Act uses “automation bias” in a provision concerning excessive reliance on a system's output. This paraphrase introduces “reliance”, “excessive” and “output”. Saying “the law already accounts for trust” would obscure these additions and leave “trust” standing as a ready-made explanation. Source S2 gives a short original excerpt and separately identifies questions about its application. The Russian edition also provides a labelled translation.
Provision is made for taking excessive trust into account.
This wording was composed for examination. It is not a quotation from the AI Act.
In this line, “Provision is made” describes a role attributed to the text, “taking excessive trust into account” names a proposed operation, “excessive” introduces an assessment, and “trust” names what is to be assessed. The relevant provision, the chosen application of these words and the material used for the assessment need to be shown. Neither “trust” nor “taking excessive trust into account” supplies what it names merely by being written. An article number does not change this.
An OECD explanatory document describes autonomy in terms of the degree of learning or action without human involvement following delegation. Source S3 preserves a short excerpt. Examining a particular system still requires an account of what is recorded as “involvement”, “delegation” and “action”, and how those records are connected to the accompanying material. This definition should not be paraphrased as a claim about experience or self-awareness.
Existing provisions concerning the interpretation, documentation and contesting of outputs are examined through their wording. The beforeword proposal specifies the unit of examination: particular wording, the description attributed to it, the grounds for applying that description, and the subsequent requirement. This is a proposed procedure, not a claim that these questions have never previously been raised.
The quoted passages do not require anyone to accept that they are the words they learned. That question is examined separately through constructed scenes and through specific requirements when supplied for examination. Attributing such a requirement to all regulators would add a claim without the corresponding material.
13A proposal for the use of descriptions
The provisions below develop one requirement: do not present writing in place of what it describes, or turn the use of words into a duty to accept that one is those written words. Supplying further grounds does not remove this requirement. “Requirement” here names the proposed wording, which is itself open to examination and challenge.
The proposal separates a requirement to use specified wording from a demand to accept that one is the written description. The second should not be made a condition of the first. If such a demand is reported, preserve its wording, stated grounds and the consequences of refusal. Separately disclose how the description is used for assessment, admission, treatment, decisions about property, restrictions on liberty or the use of AI. These words name the proposed scope of the procedure. Implementation would require separate specifications of the decisions covered, who would be subject to the duties and how they would apply. Neither the name beforeword nor the appearance of this text creates such a duty.
Presenting the material, its reading and the requirement separately is proposed as a way to identify the precise step being challenged. This is the procedure’s stated purpose; its use in organizations has not been tested here. Prohibiting a demand that someone accept that they are the written description is a separate normative choice. The reason offered is that the opportunity to challenge the application of a description should not depend on first accepting that one is that description. Such a condition would make acceptance of the disputed description a prerequisite for its review. The proposal therefore excludes that refusal from the grounds for an adverse assessment and calls for an answer or action to be assessed under separately stated criteria. This choice is itself open to challenge; “grounds”, “refusal” and “criterion” are explained through the requirements and records shown, without exempting the explanation from examination.
- Provision 1. Preserve the form under examination. Identify the exact passage, its location, language and edition, and the extent of the available material. Present paraphrase, abridgement, translation and transcription separately; do not retrospectively insert them into the initial record.
- Provision 2. Disclose the proposed attribution. State to whom or to what a label is being applied, and supply the material offered for that connection. Placing a name, image and description next to one another must not obscure how they have been connected.
- Provision 3. Show the criterion in use. Present the conditions, selected features, rule, assessment material and result. State separately what the assessment covers and where the conclusion extends beyond it. “Measured”, “proved” and “checked” do not substitute for these records.
- Provision 4. Do not require anyone to accept that they are the written description. Distinguish a requirement to use specified wording from a requirement to accept that the person described is that written description. If a field must be completed, explain its purpose and the grounds for making it mandatory. If such a demand is reported, preserve its exact wording, stated grounds and the consequences of refusal. Learning or repeating words does not establish a duty to accept that demand. Any further rule is material for examination; under this proposed procedure, it does not override this restriction.
- Provision 5. Present the grounds for the decision separately. Identify the chosen rule, its application to the case, the person or organization to which authority is attributed, and the document cited. Specify the precise action and the limits of the permission. Preserve a record of what was done separately from the decision record.
- Provision 6. Allow a particular step to be challenged. Provide access to the relevant material and a procedure for correction, review or suspension. Refusal to accept that one is the written description should not count as grounds for an adverse assessment or restriction. Any other stated grounds should be presented and examined separately: the specific answer, task conditions and rule applied. Review should not require anyone first to accept that they are the disputed description. The conditions for applying this provision require separate development for each field.
- Provision 7. For AI and AGI claims, disclose the definition, conditions and scope of the test. Keep the response, technical description, advertising promise, task result and conclusion separate. A system's self-description should not be the only material offered to substantiate that same description.
- Provision 8. Apply the same examination to the explanation and the procedure itself. Naming a source, formula, authority or “grounds” does not complete the examination. A correction must show what changed and why; it is not declared to be a record that has ceased to be writing. A detailed justification for using a record does not turn it into what it describes.
In these provisions, “material”, “rule”, “grounds” and “authority” are written names for items in the proposed examination. Their content must be supplied in each case. A completed list does not certify itself or give beforeword the right to authorize an action. A decision to adopt this procedure would also require its grounds to be presented.
The proposed documentation requirements should not delay emergency assistance or provide an additional basis for withdrawing protection already provided. Refusal to accept that one is the written description should likewise not count as grounds for such a delay or withdrawal. Any other stated grounds for a decision are examined separately through the relevant material; refusal does not substitute for that material. “Assistance”, “protection” and “decision” are explained here through the specific actions provided for.
14Applying the proposal to specific examples
All situations in this section were constructed to illustrate the procedure. They do not reproduce established circumstances in anyone's life. Names of roles and actions enter the examples as written labels.
Response: “I do not accept that I am this written description.” Assessment: “Unable to assess their own condition.” Decision: “Do not allow progression to the next stage of the course.”
The three lines are shown separately. The second and third are not declared consequences of the first.
The examination begins by preserving the response. It states a refusal to accept that one is the written description. “Unable” proposes a different description. The description under discussion needs to be supplied, together with the grounds for the added assessment and the rule connected to exclusion. Reading refusal as inability adds an assessment to the preserved response. That assessment requires separately presented grounds. Any further context in the initial document is presented separately, with its role identified.
Provision 4 requires the content of the demand to be preserved: to use a label or to accept that one is the written description. Provision 3 calls for the application of the assessment to be disclosed. Provision 5 concerns the grounds for exclusion. This allocation is itself a proposed reading of the example, open to objections concerning particular lines. It does not assign legal status to the constructed decision.
Input: “I understand.” Explanation: “The user agreed.” Next action: “Submit the application.”
No application is created or submitted in this demonstration.
The input line says “understand”. The explanation adds “agreed” and an attribution to the user. Submission requires separately presented permission for that specific action. Even if an interface rule defined in advance what pressing a button would mean, that rule and the material showing its application must be supplied. “Consent” does not substitute for those records or for examination of their connection.
- For an emotion description: preserve the accompanying image or audio separately from its caption and the classification result; show which features were selected for the label “fear”. The label itself does not supply fear.
- For a medical conclusion: present the test result, conclusion, prescribed action and record of what was subsequently done separately. A diagnosis label does not substitute for the connections shown between them. Further examination appears in the study of diagnostic records.
- For an AGI announcement: preserve the tasks and results, definition and conclusion; identify the further uses the announcement is intended to justify. “Achieved” does not perform these steps.
- For a revised response: show the earlier record, the material used for the correction and the new wording. Marking a statement “corrected” does not make it become what it describes.
An examination may yield different findings: the connection is explicitly written; the explanation proposes the connection; the necessary material has not been supplied; or the selected criterion does not cover the claimed conclusion. These are names for the results of particular comparisons. They require the passages being compared to be identified and do not become general assessments of a person or system.
A worked example: from an answer to a revised decision
The following case concerns progression to the next exercise in a course. The case, roles, deadlines and rules are constructed for this example. They do not describe a particular organization’s procedure or propose one mandatory deadline for every field.
R1 · Task: “What is 3 + 4?” R2 · Answer: “8”. R3 · Assessment: “Anna does not understand the task”. R4 · Response to the assessment: “I do not accept that I am this written description”. R5 · Decision: “Do not allow progression to the next exercise until she accepts that she is the written description”.
The name, responses and decision were constructed here. Their order shows a selected sequence of records; it does not present the teaching described.
The stated rule and its application
The constructed course conditions include R6: “Compare the answer with the result of ordinary integer addition. If they do not match, provide another attempt with a new exercise. Progression requires a matching answer. No one is required to accept that they are the written description.” Under this rule, the result for R1 is 7. R2 says 8. The comparison shows a mismatch under R6. It does not present the understanding named in R3 or add a requirement to accept that one is the written assessment to R6.
R3 adds a description and the name “Anna”; R5 makes progression conditional on accepting that Anna is that written description. The R6 wording shown contains no such condition. The word “adds” here names a comparison of specific lines: the condition can be read in R5 and cannot be found in the R6 passage supplied. Examining a different edition of the rule would require that edition to be supplied separately.
The objection, its recipient and the response deadline
In this constructed procedure, an objection goes to a course coordinator who did not write R3. The coordinator receives R1–R6; the objection log records the date, disputed wording and requested change. The example allows two working days for a response. Until then, the restriction under R5 is suspended and another attempt remains available. These are proposed conditions within the example; “coordinator”, “deadline” and “suspended” do not themselves present their implementation.
“R6 provides for another attempt. R5 adds a requirement to accept that I am the written assessment. Please remove that condition and provide another attempt under R6.”
The objection identifies the lines and the change requested. It does not require agreement with beforeword.
In this example, the coordinator is asked to compare R2 with R1 under R6, check whether R6 contains the disputed condition, and examine R5. If a different rule is cited as grounds for the decision, that document, edition and passage are requested. If it is not supplied, the stated grounds are recorded as not supplied; they are not written into R6. Under the example’s conditions, the restriction in R5 remains suspended until a written response is given. Any further rule supplied is also examined under Provision 4: it is not accepted here as grounds for requiring someone to accept that they are the written description. Supplying a rule adds material to examine; it does not turn the description into what it describes.
The revised record and the limits of the result
“The condition in R5 is withdrawn: it does not appear in R6 and is also excluded by Provision 4, adopted for this constructed example. Provide another attempt under R6. Replace R3 with: ‘R2 gives the answer 8; the result of the addition in R1 is 7.’ Preserve the earlier wording and the reason for the correction.”
This is an example of a written decision on review. No actual course access or further attempt is provided here.
Reporting implementation would require a separate record of the further attempt made available, with the relevant material. The word “Provide” in R8 does not substitute for that. Calling the passage a review does not turn the constructed lines into a procedure carried out.
Provisions 1 and 2 are applied by preserving R1–R8 and examining the added name and description; Provision 3 uses the stated task, rule and result; Provision 4 separates the answer from the demand to accept that one is the written assessment. Provision 5 calls for the grounds for R5; Provision 6 is developed through the recipient, access to material, deadline and review outcome. For a response labelled AI, Provision 7 would additionally call for the system version and the conditions under which the response was produced. Provision 8 applies to this explanation too: “criterion”, “grounds” and “result” name the parts shown in this constructed comparison; they do not exempt it from examination.
15Questions about the proposed examination
“But words are needed for communication.” This explanation introduces “needed” and “communication”, and proposes a purpose for words. In a particular exchange, the messages, response and stated purpose can be shown. Using a record and having it become what it names are not interchangeable. The question here concerns the grounds offered for applying the record; writing “communication” does not carry out that examination.
“This is already familiar: the word is not the thing.” This formulation itself proposes a relation between “word” and “thing”. The question in this publication concerns the next step: is the boundary maintained when “understanding”, “trust”, “proof” and “authority” enter the explanation? Calling the proposition familiar does not examine the application of those words to a particular case.
“Does applying a description already require someone to accept that they are the written description?” In the table in Section 03, the first line requires a field to be completed, the second requires a category to be selected, and the third requires acceptance that the person described is the written description. Only the third formulation states that last demand; it is not retrospectively added to the first two. A particular case requires the exact wording, the field’s purpose, the rule applied and material concerning the consequences of refusal. If such a demand is reported, preserve the report and its stated grounds. The use of a category alone is not enough to attribute that demand. “Category”, “demand” and this explanation remain part of the examination.
“But there are measurements.” Then the result, units, conditions, method of obtaining the result and selected application are supplied. A table of values is examined separately from the caption “trust measured”. The caption still requires an explanation of why the chosen values are offered for that reading. Adding measurement material changes the material available for examination; the written conclusion remains written.
“Documents have consequences.” A document and material concerning its use can be supplied: messages, logs, photographs and reports. The connection proposed between them can then be examined separately. A document's role in a described action does not make its wording become that action. In this paragraph, “role” and “consequences” name connections to be examined; their application to a case needs to be explained rather than accepted from the names alone.
“But a sentence can authorize an action.” A document may explicitly contain wording granting permission. That wording should be preserved as part of the document under examination. The attributed authority, scope, conditions and selected application are examined separately. Matching wording in a permission and its copy does not, by itself, disclose those conditions. This is examined in “No authority in the output”.
“Then the explanation will never end.” A publication or assessment can state its scope in advance and stop after examining the selected passage. That limit defines the extent of the work. A full stop does not give the last line a position beyond writing. Examining an explanation does not require every possible further explanation to be listed beforehand.
“Then every conclusion must be prohibited.” That is a separate normative proposal. This text proposes disclosing how a description is applied and allowing a specific step to be challenged. A blanket prohibition is a different formulation; including it would require separate grounds and discussion. The word “grounds” in the preceding sentence likewise does not substitute for the content to be supplied.
“Why should the beforeword procedure itself be accepted?” Acceptance is not a condition of reading this publication. A particular provision can be challenged, shown to be unnecessary in a chosen situation, or replaced by another proposed procedure. The proposals and the material concerning their application are compared. The name beforeword gives none of them an exemption.
16This document receives the same examination
Present the grounds for the decision separately.
This sentence is taken from Provision 5 above. The following examination concerns this wording.
The word “separately” specifies a proposed arrangement of material. “Present” states a requirement. The word “grounds” names the role proposed for material. The word “decision” names what the material is to be connected to. None of these words supplies its application to a particular case.
For its own application, this publication supplies the provision, its stated purpose—making a disputed step identifiable—and the constructed examples in Section 14. The examples show how the proposed examination is arranged; they are not reports of its implementation. The statement of purpose remains a statement. Describing a possibility does not certify an outcome for every reader.
The phrase “decisions based on records” in the title also requires explanation. In the constructed scenes, the connection is proposed through the arrangement of lines and the accompanying explanations. Earlier studies examine it through specified passages in documents. The title itself does not supply all such decisions or establish a universal history of the use of writing.
“A word does not become what it names” is the formulation of the boundary presented here. Repeating it gives this publication no exemption. The forms “word”, “become” and “names” remain within its wording and proposed reading. This paragraph likewise does not occupy a position beyond the record under examination.
An edition, source list, link, author credit, assessment and checksum may be supplied as separate information about a publication. None certifies its conclusions merely by being present. If a correction changes the wording, the change and its grounds must be preserved. A new version label does not substitute for that examination.
17Connections to earlier beforeword studies
The earlier studies are used here through the specified passages. The number of references does not certify the new proposal. This publication connects self-description, descriptions of others and demands based on records, while continuing the examination through its own terms: “grounds”, “rule”, “admission” and “accountability”.
For an introduction, see “In plain words” and the instructions for models. The first page presents constructed examples; the second proposes a way to continue the examination in AI responses. Their titles and instructions do not certify a model’s future behaviour.
18Using and responding to this publication
One passage is enough for an initial examination. Preserve it exactly; identify where it comes from and how much of it is supplied; place the proposed reading alongside it. If the passage explicitly states a connection or permission, preserve it in the initial record. Do not label it an addition merely because a later paragraph discusses it.
- Passage under examination: the exact record, language, location and extent.
- Proposed reading: the description or attribution written separately.
- Grounds for application: the material and rule supplied, and how they have been applied to this passage.
- Requirement or decision: precisely what is proposed to be done or accepted, by whom, and on what stated grounds.
- Objection or correction: the disputed step and the lines or material affected.
- Examination of the explanation: the terms introduced by the explanation itself, and how their application has been disclosed.
These items provide a format for recording an examination. Completing every field does not indicate agreement with beforeword or assign the result the status of truth, diagnosis, authority or permission. If material is unavailable for an item, that is stated in relation to the selected passage. Missing material is not reconstructed through an invented quotation.
A public comment can consist of a link to the section and paragraph, the exact disputed wording and an explanation of the proposed correction. A comment may also concern the procedure itself. Participation does not require private correspondence, medical information or third-party personal data. A constructed example should be labelled as constructed.
For developers and organizations, a separate, limited test is proposed: select one type of claim or decision, describe the material and procedure in advance, apply the procedure to a specified set of examples, and publish the recorded results. Agreement among assessments is not declared to transmit one and the same reading. A claim about the procedure's effects would require its own material and conditions of comparison.
19Edition, scope and reading routes
Kirill Shebetov · beforeword. Public proposal, version 1.4, 3 October 2026. The Russian and English editions use matching section numbers. Source translations are labelled separately. No adoption by any body or organization is claimed.
The proposal page provides a short introduction. Section 13 sets out eight provisions; Section 14 contains a worked example; Section 20 examines ten public records. The selection is also available as a separate page. Section 17 connects the proposal to 14 earlier studies; those connections do not replace examination of the passages identified.
Sources S1–S4 and B1 appear on a separate page and at the end of the PDF and DOCX files. Sources P01–P10 appear within the case entries. Materials were gathered as of 3 October 2026; the selection method and access limits accompany the selection. Version 1.4 aligns the provisions on self-description and refusal, clarifies exact written forms and Russian–English correspondences, and supplies further excerpts from the public records examined. Production and editorial checks are not presented as tests of every reader’s understanding or of how models follow the procedure.
The next section continues the examination through public records. Its explanatory wording also remains written.
20Public statements about AI: ten records from 2026
This selection examines ten public records from 2026: proposals to change the pace of development, statements about control and consciousness, instructions for self-description, and a requirement to replace a term. Sources were gathered as of 3 October 2026. The September debate is placed alongside earlier texts it draws on; each case gives its date.
The selection follows editorial criteria: relevance to this publication’s questions, availability of an exact excerpt and different ways of moving from a description to a demand. It is not a ranking by views or reposts. The selection contains four X posts and six essays or institutional texts. Company executives, the authors of a corporate training document, a head of state and a representative of a research organization are identified in their respective roles; ten records do not mean ten different authors.
Each case distinguishes a short English excerpt, the editorial Russian translation in the Russian edition, a labelled contextual paraphrase, a beforeword analysis and a source link. The excerpt does not substitute for the full text. X posts were read through the official embedding service; only the opening of Zuckerberg’s post was available. Format, date and the limits of the material examined appear alongside the excerpt.
The examination continues through “fear”, “consciousness”, “control”, “safety” and “superintelligence”, together with the words used to explain them. A written report of concern does not present the experience named; a conditional forecast should not be rewritten as a report of an event that has occurred. “Report”, “conditional” and “event” likewise state a proposed reading here, to be compared with the supplied lines.
P01 · Amodei: a demand to change the pace
Dario Amodei · Anthropic · September 2026; the page gives only the month
must slow the pace
Contextual paraphrase. Amodei proposes slowing the growth of model capabilities while continuing training and research. His plan includes external evaluators, coordination across the industry and international coordination. The scenario of an internet takeover within 6–12 months is presented as a concern, conditional on further capability growth without adequate safeguards.
beforeword examination. The excerpt introduces a demand. To consider a decision based on it, identify whom the demand addresses, what counts as the pace, what it is compared with and which actions are to change. The written terms pace, capabilities and safety do not supply what they name; defining a measure adds another record and a rule for using it. Any move between an incident report, a test record, an assessment under a chosen criterion and a forecast needs to be made explicit. Neither the title of the source nor the author’s position supplies those steps. This examination establishes neither safety nor the inevitability of harm.
For the proposed decision: which series of measurements is called the pace, what change is required, and under what conditions would that requirement be withdrawn or revised? “Measurements” here also calls for the records and the method used to obtain the values. Wording that reports the author’s concern does not present the experience named; a description of risk is not a record of a future event.
P02 · Altman: a promise to give evaluators access
Sam Altman · OpenAI · 2026-09-12
we will do the same.
independent evaluators employee-like access
Two separate excerpts from the same post, not one continuous quotation. The first describes the evaluators; the second describes their proposed access.
Contextual paraphrase. Responding to Dario Amodei’s proposal, Altman supports restraining the pace of frontier development. In the selected passage, he commits OpenAI to giving evaluators access.
beforeword examination. The first additional excerpt attributes independence to the evaluators; the second compares their access with that of employees without listing permissions. Examining the promise requires the materials to be made available, restrictions, selection procedure and publication terms to be specified. A promise of future action should not be rewritten as a measure already implemented. The label alone does not establish independence; the evaluator’s conclusions also require examination.
To examine whether the promise has been fulfilled, this proposal calls for the access terms, a list of exceptions and the rights to publish findings. “Independent” and “access” remain part of that account; the author’s position does not supply those terms.
P03 · Musk: what a brief endorsement adds
Elon Musk · founder of xAI · 2026-09-12
Dario is right
Contextual paraphrase. The brief endorsement links to Amodei’s post. By itself, it neither identifies the specific provisions endorsed nor announces a pause in development.
beforeword examination. “right” adds an assessment. Using it to support a demand addressed to others requires identifying which proposition is judged correct and what actions are proposed on that basis. Turning these three words into the headline “Musk halts AI” would add an announcement the post does not contain. The link makes the related text available to read; it does not turn the brief endorsement into a separate promise to implement every provision.
This record should not be expanded to include an unstated fear, a halt to training or agreement with every forecast in the linked text. “Endorsement” here is a proposed reading of three words accompanied by a link, not a separate observation of the author’s inner state.
P04 · Hassabis: qualified agreement
Demis Hassabis · Google DeepMind · 2026-09-12
The details need working through
direction is correct critical moment
Two separate excerpts from the same post. They are shown separately for examination below.
Contextual paraphrase. Hassabis endorses the direction of Amodei’s essay while retaining a qualification about details that remain to be worked out. The post also points to his proposal for an industry-wide standards body.
beforeword examination. The main excerpt leaves details unresolved. The additional excerpts describe the direction as correct and the moment as critical. These are added assessments; they do not supply unstated agreement terms or a measure already implemented. A decision requires the supported provisions, the way thresholds would be set and the proposed authority to apply a standard to be identified. “Threshold” and “authority” also belong to this explanation and require specific accounts of their use.
The linked essay of 14 July 2026 makes a slowdown conditional on a judgment of necessity. Applying that condition requires identifying the decision-maker, how thresholds are set and how objections are considered. Naming a standards body does not answer those questions.
P05 · Zuckerberg: responsibility and incentive
Mark Zuckerberg · Meta · 2026-09-15
Every lab has the responsibility and incentive
safely
A word from later in the same sentence, where it describes how models are trained.
Contextual paraphrase. The opening places responsibility for choosing the pace of development on each lab. This passage was read in X’s official embedded version, which did not make the full post available.
beforeword examination. The opening extends the claim to every lab. Responsibility is named alongside incentive. Reading this as a claim about a lab’s responsibility does not yet specify its duties; mentioning an incentive does not show what the lab has done. The additional excerpt describes how the models are to be trained, without reporting a completed safety assessment. A decision to continue requires stated conditions, assessment materials and a process for challenging it. “Safe” does not perform an assessment, and “assessment” does not present its completion.
For a specific application, identify whose responsibility is named, to whom it is owed and which materials support the safety assessment. The selected portion of the post cannot supply a complete account of the author’s position or a finding about how every lab operates.
P06 · Suleyman: denying consciousness and prescribing training
Mustafa Suleyman · Microsoft AI · 16 September 2026
AIs are not conscious.
Contextual paraphrase. The essay criticizes attributing consciousness to models and incorporating such ideas into training. Suleyman advocates human control and sets out his alternative to Anthropic’s approach.
beforeword examination. The negation is retained in the quotation. Yet the written “not” does not present the absence of the consciousness named. An explanation based on model architecture adds a description of that architecture and a proposed link to the conclusion; naming a mechanism does not complete that link. The next step needs separate attention: a conclusion about consciousness is used to support a requirement governing training and permitted self-description. These are two claims, each requiring its own stated grounds.
The criticism of circular justification can be examined through the sequence of records: an instruction, the resulting response, then the use of that response to confirm the initial description. Examining that sequence does not automatically establish the opposite self-description. “Mechanism”, “evidence” and “control” remain open to examination in the critic’s explanation and in this text.
P07 · Amodei: learned wording as self-description
Dario Amodei · Anthropic · January 2026; the page gives the month
think of itself as a particular type of person
Contextual paraphrase. The Defenses section describes training through character, values and self-description. Elsewhere, a hypothetical AI “country” frames a conditional scenario of domination; the author examines and criticizes an argument for inevitable human extinction.
beforeword examination. The passage proposes a way of describing oneself. It raises this publication’s question: what grounds are offered for applying a written description used in training to the system itself? An instruction, its repetition in a response and a description using “person” are examined separately. The excerpt does not literally require acceptance that the system is the written description itself; that demand must not be added to it. Translation note for the Russian edition: “person” is rendered as «личность». This introduces a Russian term; the translation does not settle a biological classification or present the person named.
“AI will enslave humanity” is not presented here as a quotation from the author: that formulation would remove conditions and introduce a different prediction. To examine self-description in practice, preserve the exact training wording, the conditions under which the response is produced, the assessment criterion and the step at which the writing is attributed to the system itself. “Training” and “system” in this proposal are also terms whose application needs to be explained.
P08 · Anthropic: uncertainty and a proposed identity
Anthropic · corporate document; principal author Amanda Askell · Announced 22 January 2026; text retrieved 3 October 2026
questions about Claude’s moral status, welfare, and consciousness remain deeply uncertain.
Acknowledging open problems, paragraph on moral status, welfare and consciousness; excerpt from its first sentence.
Claude should feel free to think of its values, perspectives, and ways of engaging with the world as its own and an expression of who it is
Claude as a novel entity, paragraph beginning with these words; opening part of the sentence.
Claude’s character and values should remain fundamentally stable
Resilience and consistency across contexts, opening of the first sentence.
Contextual paraphrase. The main excerpt retains the subjects of uncertainty: moral status, welfare and consciousness. The further excerpts propose treating values as one’s own and maintaining stability of character and values. The following paragraph separately expresses the authors’ hope that values will be accepted without pressure or fear. This is a corporate document, not a separate statement by the company’s chief executive.
beforeword examination. The statement of uncertainty is not rewritten here as “Claude is conscious”. In the first further excerpt, “should feel free” is permissive wording, not a demand to print a fixed response verbatim. “its own” and “who it is” propose a relation between values and the system described. In the second excerpt, “character”, “values” and “stable” specify a proposed characteristic and a requirement to maintain it. Including those words in an instruction does not present what they name. This analysis does not equate that instruction with a literal demand to accept that the system is the record itself.
The proposed examination calls for the exact training text, the conditions under which responses are produced and the criteria used to assess them. It then asks what is called stability of character and values in the selected comparison, and on what grounds the result is attributed to the system itself. “Stability”, “character”, “values” and “result” are terms in this proposed examination; their application needs to be explained through specific passages and material. Neither the instruction’s authors nor this publication receive an exemption.
P09 · Trump: a requirement to replace a term
Donald Trump · US President; published Executive Order 14434 · 29 September 2026
shall use “Super Intelligence” and “SI”
Contextual paraphrase. Section 2(a) directs a terminology change in executive-branch documents and communications within stated limits. Section 3(a) refers to the existing AI definition; historical documents need not be rewritten. Section 3(b) separately requests a future legislative proposal.
beforeword examination. In the selected passage, “shall” states a requirement, and the new name specifies the wording to be used. Replacing a term in a document does not demonstrate capabilities a reader may associate with “Super”. The reference to a definition adds another text and a rule for classification; it does not turn the name into a demonstration of superiority. A specific transition can be identified here: a claim about an appropriate name becomes a requirement governing other people’s records.
Examining this requirement calls for retaining its scope, addressees, exceptions, reference to a definition and the materials offered to justify the change. “Official” also adds a status label; it does not exempt the content from examination. This case concerns the published written order. It is not presented as an account based on listening to a speech recording or as a finding about the capabilities of all systems.
P10 · Aguirre: who defines the proposed halt
Anthony Aguirre · President and CEO, Future of Life Institute · 24 September 2026
most dangerous capabilities
These companies have already lost control of their products
Anthony Aguirre’s statement, first paragraph after the four proposed measures; opening of the first sentence.
Contextual paraphrase. The statement calls on governments to halt development of specified capabilities, particularly recursive self-improvement, alongside licensing, liability and restrictions on market access.
beforeword examination. “most dangerous” introduces a comparative assessment. The phrase itself does not identify the systems selected, the scale or the threshold. Using it to justify a halt requires the relevant records and the person or body applying the rule to be identified. The further excerpt, with “have already lost control”, presents loss of control as having occurred. Applying that assertion calls for the companies, systems, actions and conditions it concerns to be specified. “Control” and the account of its loss are also written here. This case examines Aguirre’s statement; it does not provide an independent examination of the associated incidents.
The proposed approach to discussion calls for the action, the scope of the halt, the conditions for resuming work and the means of challenging its application to be stated separately. The label “safety” does not supply those conditions. A demand addressed to companies and the authority to impose it require separate examination; the public importance of the issue does not turn one into the other.
What public discussion should also address
The debate contains different records: a description of an attributed capability, a forecast, a report of an experience, a commitment, a demand and a directive. Calling all of them warnings introduces a common reading that can obscure those differences. A brief endorsement does not list measures; a promise to give evaluators access is not a report of their work; a requirement to use a name does not demonstrate the capability named.
“Who controls AI?” also needs further examination. “Who” might refer to a developer, user, agency, standards body or court; “control” might describe different proposed actions. The account needs to identify who may take which actions, in relation to what, on the basis of which records and with what opportunity to object. “Humanity” does not supply everyone’s agreement or select a representative entitled to speak for them.
Claims of safety, benefit, the absence of consciousness and the adequacy of existing measures should be examined too. This does not require giving every claim equal weight. Each needs its wording, selected material and step offered as justification to be shown; “weight”, “adequacy” and “justification” also require explanation. The author’s position, the number of endorsements and inclusion in this selection do not substitute for that examination.
For further examination, supply the exact response or document being offered as grounds, together with the system version, the conditions under which the response was produced, the available actions, the selected criterion and the proposed decision. Where harm is at issue, distinguish the report supplied, the case it describes and the inference about later cases. “Report”, “case” and “harm” remain within the written explanation; supplied material is examined in its own format rather than relabelled as text.
The initial question remains: how is a learned description applied to oneself or someone else and used to support a demand? P07 and P08 show proposed ways of describing oneself and attributing values to a system. They are not presented as literal demands to accept that the system is the written form itself. P09 shows a directive to change the name used; the other cases extend the examination to pace, assessment, commitments and authority. Each transition is examined through the wording actually supplied. Those labels belong to this comparison; they do not take it outside the boundary being examined.