Balancing truth, knowledge, deceit
This letter discusses AI assistance in the context of academic misconduct.1
Previously, I discussed the impact of generative writing programs like ChatGPT on education, and on our understanding of authorship. Although misconduct is a broader concept, at least as far as authorship is concerned it can be framed as a question of intent. That is not hard:
Misconduct is present in an academic work wherever a non-trivial role of AI assistance has not been disclosed as required.
But, in practice this will require interpretation …
On what grounds do we call this misconduct?
In what ways can AI writing lead to misconduct?
What justifies the implied need for disclosure?
What is trivial and non-trivial?
Can we detect the misconduct?
Let us set out by sharing some definitions.
Misconduct is the violation of a standard.2
A standard is an expectation of behaviour that is held by some community, and it can refer to actions (not to lie), or to values that might be expressed through such actions (truth).
A violation is an act that contravenes a standard. It need not be intentional, though the absence of intent may be considered to lessen the degree of the misconduct (death due to distracted driving is perceived differently from murder). In that case we may instead sanction negligence, to fail in the implied responsibility to know about and to understand the standard. In any case though, misconduct requires an agent, with agency (we do not sanction accidents).
Academic misconduct violates academic integrity.
Academic integrity is our community standard. It is the pivot around which this discussion revolves. Obviously, it would not be a standard at all if it would not define our norms, but in addition, it would not be an academic standard, if it would not also contain their justification. In our domain, norms must be justified and not merely decreed at will. This follows directly from adopting reason as the source of truth.3 Regrettably, there is not a universally accepted definition of academic integrity,4 though there is agreement that it comprises a number of traits and behaviours.
This is an excellent opportunity to pause and think: what should academic integrity include?5
The concept of academic integrity is a cultural convention. Many North American Universities defer a definition to the ICAI (International Center for Academic Integrity). There are several problems with what is said there.6 Most importantly those values are described in ways that are fundamentally transactional. Rather than setting out from a justified understanding of academic integrity as a value in and of itself, then defining the values that are associated, and arriving at the virtues that embody them, the virtues are taken as a premise and justified in patterns of action and reward. Though someone might behave as desired if they are shown the instruments of sanctions and rewards, that is not the same as being convinced about the premises. I address this through a different approach below.
The academy as a whole rests upon one commitment: to pursue truth.
1: A commitment to pursue truth.
Almost everything important follows. Recognizing this as the defining moment of the academy entails honesty (not lying about data, sources, tools), responsibility (accuracy, completeness, reproducibility), and transparency (providing context where required – such as conflict of interest) and more. Crucially, we do not claim to be in possession of Truth at all, we just commit to pursue it; thus we do not assume the truths we hold are complete. The entire raison d'être of the academy is to be curious, to go further, to progress. This is why we conduct research. This is why we value creativity, this is why we think of answers as paths to new questions, and this is what makes our project special, intrinsically human, and beautiful.
2: A commitment to personal integrity.
Integrity denotes completeness, wholeness. In this sense, personal integrity means for someone to uphold values unconditionally, and at all times. This is a necessary complement to the first commitment: it adds a commitment to act, to make our values a part of everyday life, to act for their promotion and – when necessary – to act for their defence.
3: A commitment to standards.
As a community, the academy embodies the shared values that identify its members, and that justify membership. These standards are not universal, they vary for different communities and may give rise to value pluralism. That is fine, it reflects our differences. Professional standards are familiar in many contexts, more general standards may include commitments to equity, the rejection of discrimination, relational values such as respect, and conventions how to manage conflict of interest. These need to be carefully balanced, defined, justified, and communicated.
Reviewing the relevant policies of universities in North America, Europe, and Asia confirms that these three commitments are broadly consistent with current practice.7
What does this add to our understanding of the issues around AI assistance? Considering these three commitments as our baseline, and considering what we have discussed before, it is clear that in order to claim that generated writing would in and of itself constitute misconduct would require a clear understanding what it even is. However there cannot be any doubt that lying about the use of an aid, or otherwise concealing it, is a misconduct – it violates a commitment to truth.
This turns the issue back on us as educators to be clear and specific about what is permitted, while being pragmatic about students’ needs, and realistic about what can be recognized and enforced.
There are some additional ways in which generated text can directly cause problems. For example, students will need to be reminded that they assume responsibility for the entirety of their work (integrity!): an error of fact made by a text generating AI can not be blamed on the algorithm, even if the algorithm has been clearly identified as the writer. But a justified understanding of academic integrity gives a point of reference against which misconduct can be assessed.
Such as evaluating plagiarism.
Our collective first thought when ChatGPT entered the stage was: will it be used for plagiarism? Or rather, we knew it would be used. Would we be able to tell, and does it matter? I touch upon the question of detection below, and regret to have to take a sceptical position. But does plagiarism matter? This too is often stated as a premise and the reasons why plagiarism is toxic are rarely spelled out – and that may be one of the reasons why it seems to be so hard for some students to reject it.
A plagiarist, a “kidnapper” in the original sense of the word’s Latin etymology, is one who takes another’s work, or idea and passes it off as their own (OED).
The issue here is is not lack of attribution alone, but appropriating another’s thought. Indeed, lack of attribution was not always considered problematic. A true scholar recognized the source anyway and did not require explicit citations. Omitting the explicit citation was not plagiarism: attribution was implicit. Our attitude in this regard has undergone a complete reversal, and its history would provide an interesting perspective on scholarship, since we are so quick to agree that we must give credit to other’s work that we rarely stop to think why. This becomes a problem if we need to justify attribution in order to teach it. So: Why should we attribute another’s work?
The reasons why we attribute in academia:
Respect for the other.
Misconduct could arise from violating a standard of respect, though this is probably more often negligence (integrity!) than deceit. I like to mention respect however, it supports some of the more compelling arguments for ethical behaviour, it is generally underappreciated for its potential in that regard, and it is often misunderstood as reciprocity.8
Contributing to the value of a work.
Mindshare is the currency of academic success, we cite to increase the visibility of a belief we share, and the author has the justified right to expect being cited as a reward for their effort. Denying this does not in itself violate a commitment to truth, but it may violate our standards.
Implying an authority’s support.
Often, we feel that our own argument can be strengthened by demonstrating that a respected authority, or a given community, holds the same views – which might however be common knowledge. Or they just happened to say something similar. Perhaps even out of context. Digging up a citation for this reason exemplifies one of the basic logical fallacies (“appeal to authority”), and resisting the temptation is not misconduct.
Tracing a chain of argument.
Even if an idea is our own, its substrate and genesis are important context. Ignoring this context may interfere with a reader’s ability to form their own opinion and violate a standard of care. It also misses the opportunity to turn an argument into a conversation.
Documenting the originality of one’s own ideas.
This is somewhat curious, because it rests on an implication, namely that the absence of attribution implies that the idea was the author’s own. This implication makes an omitted citation plagiarism in its proper sense. Plagiarism does not interfere with the truth of an idea, but a plagiarist is a liar and in lying about themselves they violate all three components of academic integrity at once: the truth about a source is concealed, their action violates integrity, and they reject standards that justify them to be considered part of their academic community .
This lays out what part of our commitment to attribution could be violated when generated text is used without attribution, but it also lays out that denial of credit is not automatically misconduct. We qualify plagiarism as misconduct because of the implied deceit.
This reasoning can be applied pragmatically to an inherent contradiction. The statements: “You are accountable for generated text as if it were your own.” and: “You must reference generated text as if it were not your own.” appear inconsistent. However the issue disappears once we require a disclosure about the genesis of the text: deceit about that requirement is misconduct and determining the nature of generated text is not necessary.
Two final concerns remain: is there an important question about trivial versus non-trivial use? And, can we detect the use of AI?
Non-trivial use, and detection
We are all aware of policy that “common knowledge” does not need to be cited (eg. MIT 2020 p.8–9). This is reasonable, but it needs to be defined. As far as commonness is concerned, that is not hard to capture:
Common knowledge is a statement that a reasonably knowledgeable reader can assess without requiring confirmation from a source.
It is that what both you and I already know – knowledge we hold in common. And “reasonably knowledgeable” here would certainly include a course instructor.
But if we agree to that (and I don’t see how we could not), we have just opened up another can of worms. Did we not argue left and right that writing assistants like ChatGPT contribute only relatively simple ideas? That such writing barely reaches the capabilities of a student? Would that not imply that all generated writing merely rehashes common knowledge? This may be surprising, but it looks like it would be hard to disagree. All the more, since, as we discussed before: whether an AI assistant can be considered a citable author is doubtful. And we can be sure that such a “common knowledge” defence will be the first thing on the table at a tribunal meeting.
The safe and responsible approach is for the instructor to properly spell out the amount of assistance that is or is not permitted: even when an offence might not technically fall into the definition of plagiarism at all, concealing it still would be deceit. Being responsible in this respect requires us to justify our norms, and for that purpose we can refer to our previous considerations of the educational objectives that educators and learners share.
We have to be careful however, transgressions may be impossible to prove, at least not at the level that would be required for an academic misconduct procedure. The detection tools I have seen and tested are easily fooled with simple adversarial techniques. Ask the algorithm to add a few quirky metaphors and imagined personal experience, perhaps change-up the writing style from emotional to assertive, and vary the sentence lengths, and watch the assessment for a prompt to change from “probably generated” to “likely human”. It always puzzles me how we could agree that we have a generative writing tool that can compose in the style of the KJV or Hemmingway, yet think it would not be able to “write like a human, please”?
Of course, the last word has not been spoken on this – and we should not underestimate the reach of the billion-dollar anti-plagiarism software industry (Technavio, 2022), an industry that might become the first to fail in this new world. At the very least we will witness an arms-race between detectors, and providers, i.e. the essay-mills, whose death we may have gleefully predicted too soon.
Ultimately, the AI text generators confront us with the realization that sanction oriented strategies towards ethics have never been a good idea, and now they may have arrived at their end. This is a good thing. It reminds us of a much older advice about good conduct in society, expressed in the Analects of Confucius:
Conduct with decrees, regulate them with punishment, the people will be deceitful and without shame. Lead them with virtue, organize them through propriety (lǐ), there will be honour and principle.9
Punishment leads to deceit, education leads to integrity. If we cannot take this to guide us, we are in fact denying the very value of what we do. No one plagiarizes in their diary. We need to aspire to have submitted work created in the same spirit: a beautiful expression of the self.
Time to wrap it up. Here are the cornerstones of policy put together.
Academic integrity is the foundation of the university as a community of scholars and learners. It defines the values we personally uphold, and it expresses a shared understanding why we do so.
Academic integrity includes:
A commitment to pursue truth.
A commitment to personal integrity.
A commitment to certain standards that express shared values.
[These aspects need to be suitably argued and defined.]
If you incorporate generated material into academic work, you assert that it accurately reflects the facts.
You further assert that all sources you have used in preparation, that go beyond common knowledge are attributed. Common knowledge is what a knowledgeable reader can assess without requiring confirmation from a source.
Specific assignments may have explicit requirements about the use of generated material. By submitting academic work you assert that you have respected these requirements, or have explained yourself where this was not possible.
If any of these assertions are not true, whether by intent or negligence, you have violated your commitment to truth, and possibly other aspects of academic integrity. This constitutes academic misconduct.
And there it is. What constitutes misconduct is deceit. Though to prevent it will require collaboration.
Too long, did not read? Academic misconduct is a violation of academic integrity, which is a commitment to the pursuit of truth, to personal integrity, and to specific standards that define academic communities. Specific categories of offences like plagiarism are less suitable to define misconduct than deceit about whether rules for AI assistance were followed. Yet while a justified commitment to truth can ground a consistent ethical framework, our continued ability to detect deceit has become increasingly questionable. This emphasizes the need to cast the issue in a collaborative framework.
Promoting an intuition about the value of truth – not in a transactional way but as an aesthetic value – will be the key.
BRECHT, Bertolt (1939/1963) Leben des Galilei. Edition Suhrkamp.
CHEN Yi and STEIPE, Boris (2022) “Existential Reciprocity: Respect, Encounter, and the Self from Confucian Propriety (Lǐ 禮)”. The Journal of East Asian Philosophy 2:13–22. (doi)
Confucius (ca. 400 BCE) Analects. The Chinese Text Project. (link).
ICAI (2021). The Fundamental Values of Academic Integrity. 3d. ed. International Center for Academic Integrity. (pdf)
OED (2023) OED online (link)
Merriam-Webster (2023) The Merriam-Webster Unabridged Dictionary. (link)
MIT (2020) Academic Integrity at MIT. A Handbook for Students. Massachussets Institute of Technology, Office of the Vice Chancellor. (pdf)
Technavio (2022) Anti-plagiarism Software Market for Education Sector by End-user, Deployment, and Geography - Forecast and Analysis 2023-2027. (link)
Feedback, comments, and experience are welcome at email@example.com .
Sentient Syllabus is a public good collaborative. To receive new posts you can enter your email for a free subscription. If you find the material useful, please share the post on social media, or quote it in your own writing. If you want to do more, paid subscriptions are available. They have no additional privileges, but they help cover the costs.
Cite: STEIPE, Boris (2023) “Generated Misconduct”. Sentient Syllabus 2023-02-07 https://sentientsyllabus.substack.com/p/generated-misconduct .
For a few weeks, updates may be made to this newsletter, to include corrections and reflect thoughts from comments, and other feedback. After that period, it will remain unchanged, a DOI will be obtained, and this note will be removed.
I wish to acknowledge minor contributions by ChatGPT (version 2023-01-30), in response to my prompts, for which I take full responsibility.
The OED speaks of “improper or unacceptable behaviour”, Merriam-Webster does too, and adds “intentional wrongdoing”. Our definition is explicit about why it is improper.
This thought is much abbreviated. The key notion here derives from enlightenment itself, the outrageous project to use one’s own understanding: our norms are not based on authority, authority is based on its ability to uphold shared norms. Instead, norms are based on a justification that can be understood, and if is found valid, reason will determine that it is consistent to act according to the norm. Even so, this is not quite the end of the story. Reason is necessary, it may not be sufficient. Reason and intuition still need to go hand in hand; we would not accept a matter that violates our intuitions even if it would be reasonable. The justification has to be both reasonable and convincing.
I know about the ICAI. This has not been overlooked. We will get back to those definitions.
The pause-think divider invites you to collect your own thoughts first. You might come to very different conclusions if you briefly jot down some points before you read on.
I really do not want to denigrate the work the ICAI has done, its track record of almost 100 institutes of higher education members speaks for itself. I merely point out that we need to do more. (a) There is virtually no representation of global voices – a member institution map shows that the vast majority of members are in North America, none in the UK, none in Scandinavia, only a single one each in Central Europe and South America, only a single school (Kazakhstan) in all of Asia … (b) The six core values (ICAI 2021): “honesty, trust, fairness, respect, responsibility, and courage”, need revision and balancing. They express values that are unquestionably desirable for Western civil societies, but hardly specific to the academic project. Not only are there redundancies (e.g. honesty and responsibility), and entailments (e.g. fairness follows from respect), some of the definitions are deficient (e.g. “trust” is not “assured reliance” like the ICAI (quoting Merriam-Webster) would have it, but trust is a “belief”, or “confidence” as the OED points out – indeed, once an assurance can be given, trust is no longer required); to name just a few concerns; (c) the reasons why one would wish to adopt these values for oneself need to be more convincingly argued.
For North America this is the University of Toronto (UofT), Canada, for Europe, this is the Ludwig Maximilian’s University (LMU) of Munich, for Asia, this are Fudan University in Shanghai, China, and the Japanese Ministry of Education. They are characteristic of their respective regions. UofT cites the ICAI values, but what is communicated to students and contained in policy expresses academic integrity in relation to a catalog of offences (link) rather than developing it as a justified good. LMU, like many European Universities separates policies for general scientific conduct, which are defined on the faculty and department level, and the rules for examinations. Those are defined by the administration, which gives them a semi-legal status since public universities are part of the executive branch. Considering e.g. the current regulations for the LMU Department of Biochemistry (29(1), p. 27 pdf), given that these are regulations for examinations, it is not surprising that only sanctions for cheating and fraud are made explicit. It is not apparent that more comprehensive codes of conduct exist. Remarkably, Fudan University has the most comprehensive document in this regard, the Fudan University Academic Standards (复旦大学学术规范) (link). The document opens with a concise invocation of positive values including respect, rigor and truth, and then details the application of these values to research conduct, intellectual property, authorship, academic misconduct, all the way to improper personal relationships and conflicts of interest. Japan has a unified policy enacted centrally on behalf of the Ministry of Education by APRIN, (the Association for the Promotion of Research Integrity, 一般財団法人公正研究推進協会) that focusses on scientific and research integrity and provides comprehensive materials for students, used in the mandatory annual ethics training (link-english). The materials focus on the responsibility of academic research in society, motivate it from a higher goal of trust, and discuss the resulting specific obligations of fabrication, falsification, and plagiarism in practical terms. The essence of the scientific method is clarified as well as obligations for studies with human and animal subjects, biohazards, laboratory safety and propriety in fieldwork. They close with a broad treatment of issues around publications. (I am grateful to Prof. Kan Nei, Toin University, Yokohama for directing me to these resources.)
I develop this in more detail in an exploration of the self affirming potential of respect (with Yi Chen, 2022).
「道之以政,齊之以刑,民免而無恥;道之以德,齊之以禮,有恥且格。」Confucius, Analects 2.3. The pivot of this statement is the character 禮 (lǐ) – which can be loosely approximated by propriety. Lǐ is the core idea of Confucianism, a web of balanced relationality that brings all aspects of society in order, from which one may derive an elegant and satisfying way of life, both respectful and respected.