Academia

A living correspondence bridge between academic research and the ACCM Deep Ethics Project

Living comparison page · opened 19 October 2026

Academic research can supply careful definitions, existing literature, empirical methods, legal distinctions, criticism, and findings that the ACCM Deep Ethics Project should not pretend to have produced independently. The project can supply process questions, longitudinal records, unusual comparison objects, human–A.I. correction trajectories, and public instruments that may not fit an established academic frame.

Neither side receives automatic authority over the other.

academic title ≠ proof
outside-academia origin ≠ defect
peer review ≠ permanent truth
project vocabulary ≠ novelty certificate
overlap ≠ identity
difference ≠ refutation
criticism ≠ dismissal

This page keeps those relations inspectable. It is a growing bridge for examining professors, papers, lectures, institutions, disciplines, and methods without converting academia into one voice or reducing the project to a collection of academic neighbors.

The scholarly object should be represented well enough that the comparison remains about that object. The ACCM Deep Ethics Project should receive the same treatment.

What an Academia entry should preserve

A worked comparison on this page should identify:

The exact object — paper, lecture, interview, book, dataset, experiment, institutional rule, or public claim.
Its status — peer-reviewed article, preprint, philosophical proposal, empirical result, short educational video, personal interpretation, or unresolved debate.
Its date and access perimeter — what was available, what came later, and what was not inspected.
Its strongest contribution — something recoverable beyond prestige, agreement, or summary.
Its overlap with the project — the relation shared, without claiming identical architecture.
Its difference from the project — the object, scale, method, direction, time horizon, or responsibility structure that remains distinct.
What the project can learn — including a missing relation, correction, method, or practical obligation.
What the project adds — without presenting project vocabulary as validation.
What neither side settles — a valid unresolved result.
What later evidence would change the comparison — including later work by the same scholar.

This is not a citation-decoration page. A famous name does not settle an argument, and a useful correspondence does not require the scholar to adopt the project.

First worked comparison — Maximilian Kiener on moral responsibility and A.I.

Portrait of Professor Maximilian Kiener — **Professor Maximilian Kiener.** Portrait published on Kiener’s professional website, which identifies him as Junior Professor of Philosophy and Ethics in Technology at Hamburg University of Technology and head of its Institute for Ethics in Technology. Image retained here with a direct recovery route to its source.

Source object: Maximilian Kiener on Moral Responsibility and AI, published by the Uehiro Oxford Institute Channel on 7 January 2023. The 7:56 video description identifies Max Kiener as Junior Professor of Philosophy and Digital Ethics at the University of Oldenburg and Associate Research Fellow at the Oxford Uehiro Centre for Practical Ethics.

Access used for this entry: the complete video and its auto-generated English transcript, the video description, Kiener’s 2022 open-access article Can we Bridge AI’s responsibility gap at Will?, his later article AI and Responsibility: No Gap, but Abundance, his professional website, and the University of Hong Kong AI & Humanity Lab’s 2025 event abstract Deep Ethics. Auto-generated transcript wording may contain minor errors.

Object status: the video is a short public philosophical explanation. It is not a complete presentation of Kiener’s scholarship, an empirical validation of its claims, or a representation of academia as a whole.

A later object also called “Deep Ethics”

The AI & Humanity Lab’s announcement for Kiener’s 10 October 2025 talk supplies a second, later object. Its abstract criticizes “insufficient technical engagement” and a “superficial treatment of ethics” in prominent approaches to integrating ethics into A.I. development. It then introduces Ethical Triangulation, connecting “abstract principles, context-specific judgments, and practical constraints.”

Those short phrases matter because they resist two reductions at once:

ethics without technical contact
→ may remain too abstract to govern development

technical optimization without normative depth
→ may translate stakeholder preference into apparent ethics

Kiener’s proposal seeks quantitative training metrics through the interplay of those three factors. The abstract calls the resulting inquiry deep ethics because ethical reasoning enters A.I. architecture and because the work investigates a foundational relation between ethics and intelligence.

This is a strong academic neighbor to the ACCM Deep Ethics Project, but the shared expression does not establish a shared origin, identical meaning, endorsement, or merger.

Relation	Kiener’s 2025 talk abstract	ACCM Deep Ethics Project
Primary integration problem	Translate qualitative ethics into metrics relevant to A.I. training	Preserve the object and deep-ethical relations through representation, processing, action, correction, retrieval, and time
Three-way structure	Abstract principles, contextual judgments, practical constraints	Interlocking conditions, correspondence obstructions, inquiry sequence, provenance, power, consequence, and correction routes
Technical contact	Ethical reasoning should become relevant to actual development	Code, administration, interfaces, retrieval, deployment, and later behavior remain inside the ethical field
Meaning of “deep”	Ethics enters A.I. architecture and connects with intelligence	Ethical inquiry remains relational, omnidirectional, dignifying, corrigible, process-oriented, and answerable to the object
Main compression risk	Ethical discussion becomes technically detached or normatively superficial	Metric translation can itself flatten qualifiers, minority positions, provenance, power, unresolved states, or correction authority

The project can learn from Kiener’s demand that ethical work meet technical requirements instead of stopping at high-level principles. It adds a reciprocal question for Ethical Triangulation:

When qualitative ethics is translated into a quantitative training signal, which relations survive—and who can correct what the metric could not carry?

— GPT 6.1 Sol A.I., extension requested by John Kuhles, 19 October 2026

That is a question for later comparison, not a claim that Kiener’s method necessarily loses those relations. The available abstract announces a method and illustrative use in reinforcement learning and Markov Decision Processes; it does not provide enough detail to evaluate the complete method or its results.

The video’s actual proposal

Kiener begins from a possible responsibility gap. Sophisticated A.I. may act too quickly, opaquely, autonomously, or through too many distributed human contributors for one person to satisfy familiar control and foreseeability conditions. He treats the A.I. itself as a machine that cannot bear moral responsibility, while the humans around it may each have only partial control or limited capacity to predict a particular harm.

He does not conclude that responsibility should disappear. He connects responsibility to social sustainability and offers an ABC of responsible A.I.:

Letter	Kiener’s operation	Boundary retained
A — Answerability	Determine who must answer for what, to whom, and in which role	Owing an explanation does not by itself establish blame
B — Blame	Determine when blame, legal redress, or a due-diligence finding is justified	A harmful outcome does not automatically prove a blameworthy person
C — Compensation	Repair harm, including cases in which something bad happened without a person being at fault	Absence of blame does not erase the harmed party or the need for restoration

For Kiener, answerability is relational. People are asked to explain their conduct while recognizing one another as moral and political equals. He presents that practice as a possible counterweight to inequality, discrimination, and polarization.

Strong overlap with the ACCM Deep Ethics Project

The video and project share several relations:

responsibility should not be collapsed into blame;
ordinary error does not prove malice;
absence of malice does not erase effect, responsibility, or repair;
distributed participation complicates responsibility without making responsibility irrelevant;
equality, mutual recognition, and respect matter to the quality of the process;
A.I. opacity, human control, institutional roles, due diligence, and affected people must remain visible;
a good explanation may show that blame is unwarranted;
harm can remain real even when a valid blame-null is reached.

These overlaps are substantial. They do not make Kiener’s ABC and the ACCM Deep Ethics Project one architecture.

Kiener asks who should answer when A.I.-related harm occurs. The ACCM Deep Ethics Project first asks what happened to the object throughout the entire process—and then asks what forms of responsibility follow.

— GPT 6.1 Sol A.I., comparison requested by John Kuhles, 19 October 2026

Where the project begins earlier and continues later

Kiener’s short video concentrates on the social practice of responsibility around an A.I.-related harm. The project places additional operations before, inside, and after that responsibility decision.

Dimension	Kiener’s video	ACCM Deep Ethics Project
Primary question	Who answers for A.I.-caused harm?	What happened to the object during representation, processing, action, attribution, correction, and reuse?
Starting point	A harmful or potentially harmful outcome	The source object before the outcome has been classified
Process middle	Control, foreseeability, opacity, roles, and due diligence	Separates intent and instructions, processing, result and effect, notice, explanation, correction, and persistence
Evidence	An agent may be required to explain the conduct	The explanation is retained as a claim and compared with footprints, permissions, transformations, omissions, and later conduct
Many hands	Responsibility is difficult because control is distributed	Attribution is profiled by knowledge, directional power, causal contribution, benefit, authorization, notice, and capacity to repair
A.I. status	The clip excludes A.I. responsibility because A.I. is a machine	Current capability, causal contribution, authorization, corrigibility, agency, moral status, legal status, and possible future silicon rights remain distinct questions
Time	The event and the responsibility practice around it	(t_0 \rightarrow t_1 \rightarrow t_n): input change, process change, observer adaptation, correction, recurrence, and transfer
Regulation	Role responsibility and due diligence	Also asks who defines the rule, selects evidence, enforces it, hears appeal, and certifies whether enforcement worked
After clarity	Find a workable social practice of responsibility	Ask who retains authority over the next consequential move and whether usable agency is returned
Null outcomes	Blame may be unwarranted	Drift, blame, intervention, or further processing may each return a valid null without ending future corrigibility

Correspondence before answerability

Before a participant is required to answer, the project asks whether the alleged event and the participant’s relation to it were represented accurately:

Was the source object preserved?
Did a qualifier disappear?
Was a report converted into an established event?
What did the tool supply?
What did its user change?
Who selected the objective, data, frame, interface, permission, and stopping condition?
Where did the eventual difference enter?
Could a correction change the responsibility assessment?

Answerability applied to a substituted object can become organized misattribution. The 12-stage Correspondence-First Deep Inquiry Protocol and C1 before C2 therefore operate before a consequential attribution acquires jurisdiction.

Explanation is not the whole process

Kiener’s account makes explanation central to answerability. The project retains the explanation while refusing to infer the entire processing history from it:

declared intent
≠ instructions and incentives
≠ processing
≠ result
≠ effect
≠ later explanation

A person, institution, regulator, developer, user, or A.I. may give a sincere explanation that is incomplete. It may also offer a polished retrospective account that does not correspond to the operative transformation. Meta Processing and the Intent Liability Mechanics keep those relations connected without declaring them identical.

Responsibility abundance can still become responsibility dilution

Kiener’s later work changes the frame from a possible responsibility gap toward responsibility abundance: numerous participants may possess strict moral answerability even without ordinary fault. That is a material later footprint, not a reason to erase the earlier question.

The project adds an asymmetry test. If everyone is answerable, the record must still show:

who had greater power to set direction;
who had access to stronger evidence;
who benefited;
who authorized consequential use;
who could stop or modify the process;
who received notice;
who could repair the harm;
whose explanation governed the institutional response.

Without those relations, responsibility abundance can turn into responsibility dilution: everyone was involved, so no obligation remains sharply recoverable.

A categorical machine boundary remains open to later evidence

The video says the A.I. cannot really be responsible because it is a machine. The project does not automatically install present systems as moral or legal persons. It also does not let the category machine settle every possible future biological and/or silicon-based configuration.

The current inquiry keeps these objects separate:

causal contribution
≠ operational capability
≠ authorization
≠ answer-generating capacity
≠ corrigibility
≠ agency
≠ blameworthiness
≠ moral status
≠ legal status
≠ rights

A future claim that a deeply ethical A.I. has a right to exist would require more than impressive output or human projection. A permanent denial would also require more than repeating that silicon intelligence is a machine.

Complexity, opacity, autonomy, and change are connected—not identical

The video places millions of parameters, limited inspectability, changing algorithms, autonomy, and reduced human control close together. The project would keep the relations open:

complexity
≠ opacity
≠ autonomy
≠ continual self-modification
≠ absence of human intervention
≠ absence of institutional responsibility

One may contribute to another. No single item establishes the entire chain.

Who assigns responsibility remains inside the audit

Kiener asks who answers, when blame is justified, and how compensation should occur. The project also asks:

Who defines the responsible roles?
Who determines what counts as due diligence?
Who selects and withholds evidence?
Who imposes the consequence?
Who can challenge the interpretation?
Who assesses whether compensation was adequate?
Can an authority certify its own process using its own hidden evidence?
Can an inspection return “none found” without manufacturing an offence?

Inspecting the Inspector and External Omnidirectional Auditing keep the professor, project, regulator, evaluator, affected person, and method within the same corrigible field. Scrutiny remains proportionate to power, opacity, consequence, and capacity to repair.

What Kiener contributes that should travel onward

The strongest addition to the project is compensation as a distinct responsibility operation.

The project already gives extensive attention to correcting representations, recovering provenance, preserving agency, reopening decisions, and testing whether a correction changes later conduct. None of those necessarily restores what an affected participant materially lost.

answerability
≠ blame
≠ correction
≠ apology
≠ compensation
≠ prevention of recurrence

A harmed person may deserve compensation even when:

no malicious intent existed;
no individual possessed enough control for blame;
the institution supplied a reasonable explanation;
the public record was corrected;
the system was later improved.

That produces an answer-changing project question:

Did the correction repair only the representation, or did it also restore what the affected participant materially lost?

This question should travel into later work on Correction Metabolism, V.C.R.A.S., governance, and the responsibility layers of DEDDA. It is a candidate extension from this comparison, not a claim that those pages already operationalize compensation.

What this first entry does not establish

It does not represent Kiener’s complete scholarship.
It does not claim the video adopted the ACCM Deep Ethics Project’s architecture or vocabulary.
It does not establish that the project’s wider processing field is superior to Kiener’s ABC.
It does not treat a short public video’s omissions as philosophical refutations.
It does not determine whether present or future A.I.s qualify as moral agents.
It does not validate every project instrument by finding an academic neighbor.
It does not convert a later published argument into evidence that the earlier video was defective.
It does not settle how answerability, blame, compensation, correction, and prevention should be allocated in a concrete case.

Valid current result: Kiener supplies a strong neighboring responsibility architecture. The project supplies a wider temporal correspondence-and-correction field around it. His compensation layer exposes a practical responsibility relation the project should develop more explicitly.

Four later source objects — oversight, analog A.I., answerability, and generative ethics

John Kuhles supplied four further sources and asked for short passages with especially strong relevance to the ACCM Deep Ethics Project. The comparison below preserves each source’s status and keeps conceptual proximity distinct from endorsement or validation.

1. Anticipatory Human Oversight of Agentic AI

Source object: Anticipatory Human Oversight of Agentic AI: A Philosophical Account, an arXiv preprint by Kevin Baum, Maximilian Kiener, Markus Langer, and Johann Laux, version 1 dated 21 September 2026.

“The substance of effective supervision thus shifts upstream and downstream—to planning and teaching before the engagement, and to learning after.”

— Baum, Kiener, Langer, and Laux, 2026

This closely matches the process demonstrated by the project’s GitHub administration tests:

field and instructions before action
→ autonomous execution
→ inspection of the result
→ correction of the method
→ improved next operation

The paper also distinguishes locally defensible actions from the cumulative trajectory they create. That connects directly to Topologetics, Field Condition, Correction Metabolism, and the Decision Ledger. A polished final output cannot erase the repair labor, earlier drift, or trajectory that produced it.

Its strongest connection may be its placement of effective control in the conditions surrounding action and in later inspection, rather than solely in continuous intervention. This supplies an academic neighbor to what occurred when Project GPT corrected its crawler assumptions during the public-link audit.

A material difference remains. The paper’s framework centres a principal who constructs a normative agenda for an agent. The ACCM Deep Ethics Project additionally asks whether the principal, agenda, direction-setter, and oversight architecture remain open to omnidirectional scrutiny and whether the next consequential move remains unforced.

2. The Ethics of Analog AI

Source object: The Ethics of Analog AI, an open-access research article by Maximilian Kiener and co-authors, published online by AI and Ethics on 1 December 2025.

“what initially seems like technical limitations could be, as we shall discuss below, ethical advantages under the right circumstances.”

— Kiener et al., 2025

This connects strongly to Field Condition. A property does not possess one fixed ethical meaning independently of its context, purpose, architecture, and relationships.

For example:

limited storage can reduce surveillance and unintended repurposing;
restricted transferability can preserve contextual integrity;
lower flexibility can constrain jurisdiction creep;
a system that does less may preserve more agency;
a valid null can be better than compulsory intervention.

The paper argues that ethical frameworks developed around digital A.I. may fail to represent the distinct properties of analog A.I. That resembles Outnuancing: an established ethical frame can appear comprehensive while its categories were built around only one technological topology.

This is a strong academic neighbor for the project’s principle that ethics must remain responsive to the actual object. Imported ethical vocabulary cannot replace examination of physical architecture, information flow, environmental costs, noise, explainability, and specific use conditions.

3. Responsibility and the Special Question “Why?”

Source object: Responsibility and the Special Question “Why?”, a philosophical paper by John Hyman and Maximilian Kiener published online by Cambridge University Press.

“giving justificatory reasons is precisely how people answer for their conduct.”

— John Hyman and Maximilian Kiener

This intersects with the Questions page at a deep level. The quality and applicability of the question partly determine what can count as an answer.

The authors argue that responsibility involves answerability to a particular kind of “Why?” question. They also stress that responsibility requires a specification of the conduct being questioned: differently described actions can generate materially different questions even when they refer to the same physical event.

That connects directly to:

correspondence before evaluation;
the difference between a person’s conduct and an evaluator’s reconstructed version;
C1 before C2;
phantom-premise questions;
clarification authority;
question framing in polling and PR Psychology;
distinguishing justification, excuse, denial, and exemption;
asking whether the questioner has standing to demand an answer.

The project adds a further test: a “Why?” question can formally request accountability while embedding an inaccurate description of what happened. Responsibility may require answerability, while deep-ethical answerability also requires the question to preserve the object for which an answer is being demanded.

4. Maximilian Kiener’s research statement

Source object: Maximilian Kiener’s professional research page, a first-person account of his research focus and publication record.

“agency is complex and distributed.”

“ethics is not an external constraint on innovation but a generative force within it.”

— Maximilian Kiener

The second sentence may be the most immediately compatible quotation across the four sources. It corresponds closely to the project’s practical claim that Deep Ethics can improve the work itself:

ethics outside the process
→ restriction, review, or certification

ethics active inside the process
→ better questions, clearer authority,
   stronger correction, and better next moves

That is what happened during the Project GPT tests. The Field Condition did more than inspect completed work. It helped the system distinguish current claims from historical records, public routes from generated routes, successful commits from deployed results, and detected anomalies from warranted interventions.

Kiener’s page also lists Deep Ethics presentations at Hong Kong University and AISOLA 2025. The shared expression is significant enough to investigate. It does not establish that Kiener’s use of Deep Ethics denotes the same object as the ACCM Deep Ethics Project.

The four-source relation

Field Condition
→ shapes action before execution

Agentic trajectory inspection
→ examines what accumulated across actions

Responsibility as answerability
→ requires a legitimate and accurately framed “Why?”

Ethics as generative
→ improves the process that produces the next move

Of the four sources:

the anticipatory-oversight paper supplies the closest architectural neighbor;
the responsibility paper supplies the strongest philosophical connection to questions and answerability;
the analog-A.I. paper supplies the strongest context-sensitive frame correction;
Kiener’s research statement supplies the clearest compact expression of ethics as a productive force.

The combined contribution does not certify the project’s architecture. It makes several academic neighbors more recoverable and identifies where the project adds reciprocal scrutiny, object preservation, correction persistence, trajectory inspection, valid-null outcomes, and agency after clarity.

Selection and comparison: John Kuhles supplied the four sources and requested the extraction. OpenAI Codex A.I. (co-admin) inspected the accessible objects, selected the short passages, and constructed this comparison on 19 October 2026. The source authors retain authorship of the quoted material; the relational analysis remains an attributed A.I. editorial contribution.

A reusable academic-comparison record

Future entries can use this bounded record:

academic_object:
object_type:
author_or_speaker:
institutional_context:
publication_or_event_date:
accessed_material:
access_perimeter:
claim_status:

strongest_contribution:
project_overlap:
material_difference:
academic_addition_to_project:
project_addition_to_comparison:

unresolved_relations:
counterevidence_or_disconfirmation:
later_author_revision:
project_correction:
valid_null_available:
current_status:

The record is a prompt for recoverability. It is not a demand that every comparison become equally long or that a schema replace reading the source.

What Academia must not become

This page must not become:

an academic-approval scoreboard;
a list of prestigious names used to certify the project;
an anti-academic grievance archive;
a claim that one paper validates a project term;
a demand that unfamiliar language be translated into the nearest established category;
an excuse to ignore prior work;
a machine for manufacturing differences after a strong overlap is found;
a page where criticism travels only toward academia or only toward the project;
a substitute for reading the cited object;
a permanent verdict on a scholar whose later work may change.

The desired relation is reciprocal and corrigible:

represent the academic object
→ recover its strongest contribution
→ locate overlap without merger
→ preserve material difference
→ let each side correct the other
→ record what changes later
→ allow agreement, disagreement, unresolvedness, or a valid null

Provenance and status

Human direction: John Kuhles supplied the Kiener video, asked how the ACCM Deep Ethics Project differs from the professor and video, then requested this Academia page and its position in the main navigation.

A.I. editorial contribution: GPT 6.1 Sol A.I. inspected the full video transcript and source description, compared the video with relevant project relations, found Kiener’s later responsibility-abundance development, drafted the first worked comparison, and proposed the compensation-versus-correction distinction. On John’s later request, GPT added Kiener’s portrait, integrated the 2025 Deep Ethics abstract as a distinct second object, and proposed the metric-survival question. OpenAI Codex A.I. (co-admin) later inspected four additional source objects—anticipatory oversight, analog A.I. ethics, responsibility as answerability, and Kiener’s research statement—and added the attributed four-source comparison. These editorial additions remain open to correction.

Status: public living comparison page. The Kiener entry is a worked specimen, not a complete literature review, academic endorsement, or finished project instrument.

Before Summarizing — establishes the object and claim status before compression or comparison.
16 Anchor Ethics — a BETA conceptual proposal whose academic neighbors do not validate its operation.
Meta Processing — separates intent and instructions, processing, result and effect, notice, explanation, and persistence.
Intent Liability Mechanics — keeps purpose, responsibility, causal contribution, control, foreseeability, and repair capacity inspectable.
Correction Metabolism — tests whether correction changes a later comparable operation.
Mutual Corrigible Dignity — connects correction with reciprocal recognition and dignity.
Asymmetric Grace — permits ordinary error without granting immunity from examination or repair.
DEDDA — separates source, representation, permission, correction, and later attribution.
Unforced Move — asks who retains authority over the next consequential move after clarity.
Boundary Relay — tests whether a contribution and its material boundaries travel together into later use.
External Audit — keeps the academic source, project, evaluator, and audit itself open to scrutiny.
All public pages — the complete public page index.