Academia and other public-facing research communities are currently facing a number of pressing questions about how the practices of research and publication must change in the age of AI. One question that I find particularly pressing is the rapid rise of false citations. I see false citations all the time these days, in research that is submitted to me in my capacities as a journal editor, as a referee reading submissions to other journals and presses, and as a teacher.
I am going to argue that false citations are fraud. More specifically, I plan to argue that submitting research with false citations to a research journal or academic press constitutes fraud. My argument here is agnostic about consequences, sanctions, or legal standards, although I also care about those things too and I will be drawing on some legal concepts in the course of the argument below.
Instead, I want to establish the normative position that false citations are fraud. In other words, I want you to think that when someone shares research with you that contains false citations, they are committing fraud. I want you to teach your students that you will consider them to be committing fraud if they submit research with false citations for review. I want you to believe that it would be fraudulent to author a false citation, and therefore not to do it—regardless of whether you would suffer any sanction for being caught—because fraud is wrong.
This argument is specifically about false citations. Lots of other frustrating parts of the current publication landscape are not false citations, and my argument does not apply to them. And, lots of other things are fraudulent that are not false citations. My task is here is narrow, to explain why you should think that false citations are fraud.
To do so, I will first define false citations, and then I will define fraud. Then I will explain what a citation does, and use this to explain why submitting a false citation for review at a journal or press constitutes fraud, regardless of how it got there.
What is a false citation?
A false citation is a citation to a false reference. A false reference is a reference entry that does not exist. It “does not exist” in the sense that the reference entry does not point to a piece of research that bears that title, author, or associated metadata. A false reference fails the core function of a reference, which is to refer to something.
False citations are not the same as citations to documents that exist, but in which the author misconstrues or represents their content. Those are also an interesting and difficult problem, but they are not the same as false citations.
False references are also not the same as references with errors in them. Here is a reference with an error in it:
Thomas B. Pepinsky. 2014. “The Global Economic Crisis and the Politics of Non-Transitions.” Government and Opposition, Vol. 47, No. 2, pp. 135–161.
Here is a false reference:
Thomas B. Pepinsky. 2014. “Authoritarian Coalitions.” American Political Science Review, Vol. 109, No. 1, pp. 437–449.
The former refers to a real research output, but has a mistake in it (the year is wrong). The latter does not refer to a real research output at all. It is false because it does not refer to anything.
You are probably thinking of an edge case. Maybe it will be hard to tell in some specific instances if a citation points to a false reference or just a flawed reference. But I maintain that, regardless of whether we can easily distinguish between a flawed reference and a false reference, there is a difference. The false reference is my concern.
To repeat: a false citation is a citation to a false reference.
What is fraud?
Fraud has many meanings. Here is one source that gives a nice summary of legal conceptions of fraud in the Anglo-American context:
In civil litigation, a fraud claim may include an intentional misrepresentation, a negligent misrepresentation, or, in some circumstances, a false promise.
- Intentional misrepresentation occurs when a person makes a false statement of fact knowing that it is false or recklessly disregarding whether it is true, intending that another person rely on the statement. The person to whom the statement was made must reasonably rely on it, and that reliance must cause harm.
- Negligent misrepresentation differs because the person making the false statement may honestly believe it to be true. The problem is that the person lacked reasonable grounds for that belief. Negligent misrepresentation is a separate tort and does not require an intent to defraud, although it generally requires an intent to induce reliance, reasonable reliance, and results in harm. A broken promise is not always fraud.
- A false promise may support a fraud claim when, among other requirements, the person making the promise did not intend performance during consideration, and intended that the other person rely on the promise. The other person must have reasonably relied on the promise, the promise must not have been performed, and the reliance must have caused harm. For example, California recognizes false promise as a form of fraud and describes its elements in California Civil Jury Instructions (CACI) No. 1902.
The relevant bits are “intentional misrepresentation” and “negligent misrepresentation.” In the American legal context, you can be held liable for fraud if you recklessly disregard whether or not a statement is true, or if you honestly believe something is true but lack reasonable grounds to hold that belief. (EDIT: To clarify, it is not literally called “fraud” if the issue is negligent misrepresentation.)
Fraud must also cause someone else to rely on that statement, and result in a harm.
For my purposes, I am not interested in whether it is possible to prove that a statement is fraud. I am narrowly focusing on the conceptual argument, that any action or statement of either of those forms is fraud.
What Do Citations Do?
I focus on false citations as the instance of fraud, instead of false references, because the citation is where the misrepresentation happens, not in the reference itself.
Think of the function of a citation. At root, a citation is a marker that says, I claim that there is information in some other document that is relevant for my argument at this point. The citation maps to a reference entry, which contains descriptive information about that other document that allows the reader to locate it. Importantly, a citation entails two separate claims: an authorial claim and an existential claim. To illustrate these two kinds of claims, consider the example,
- Pepinsky (2010) argues that A, B, and C.
This citation’s authorial claim is that Pepinsky (2010) argues that A, B, and C. It can be deemed to be true or false check checking if, in fact, Pepinsky (2010) argues that. If Pepinsky (2010) argues ~A instead of A, or only argues D and makes no reference to A, B, or C, the authorial claim is false. Many objections to citations are objections to their authorial claims.
That citation’s existential claim is that Pepinsky (2010) exists. It amounts to the claim that there exists another document, uniquely identified within this document by the marker “Pepinsky (2010)”, with the characteristics stipulated in the references section. A citation’s existential claim can be true even if its authorial claim is false. If a citation’s existential claim is false, then its authorial claim is irrelevant.
False citations are fraud when they are submitted for review
I claim that every false citation reflects either intentional misrepresentation or negligent misrepresentation, each of which constitutes fraud, and this covers any possible mechanism that could generate a false citation in a piece of research.
How can a false citation arise? What mechanisms could generate a citation to a reference that does not exist?
One possible mechanism is that an author (or authors) has made up a false reference, and then cited that reference. That is plainly fraudulent: it is a false claim by an author who knows that it is false.
Another possible mechanism is that the author has delegated the composition of text to someone or something else. Consider a case where an author has delegated text composition to a research assistant who is not an author, and that assistant produces text that contains a false citation to a false reference. In that case, the author may truthfully claim that they did not generate that text. But if they have affixed their name to that text as an author, then they claim reponsibility for that text.
The text that contains the citation constitutes an existential claim. Not verifying that that claim is true, while also claiming responsibility for it, is reckless disregard for the truth of that claim at worst, and negligent misrepresentation at best. The problem is not that there is a false reference in the bibliography, it is that the author written a citation that claims that that reference exists. That is what is fraudulent.
A related mechanism is the same as the previous one, but instead of a research assistant, the source of the text is AI. The analysis is exactly the same as in the previous case, and the conclusions are too.
A final possible mechanism is that the author possesses some sort of technology that manages references and citations. These are databases that store reference information, with a front-end interface that generates and formats references and citations within a given document. One attractive feature of citation management software is that computer programs do not make errors. Given a valid input, the software produces a valid output. The only way to produce an invalid output is to supply the database an invalid input.
But what if the author possesses a bad technology that garbles the output, generating text that differs from the input? Perhaps the author attempts to create valid references using true inputs, but inputs them into the bad technology, and the result is invalid outputs that are false references, and citations that point to them.
Again, if the author has affixed their name to that text, then they claim reponsibility for what it contains. That false citation is negligent misrepresentation, because there are no reasonable grounds to believe that bad technology produces valid outputs; or intentional misrepresentation, because it is reckless to use a technology without ensuring that it produces valid outputs.
Let’s conclude with a plausible real world scenario. We read the following in manuscript:
Pepinsky (2014) argues that a coalition-based perspective on authoritarianism is helpful for understanding how authoritarian regime behave.
References
Thomas B. Pepinsky. 2014. “Authoritarian Coalitions.” American Political Science Review, Vol. 109, No. 1, pp. 437–449.
On one hand, that reference does not exist, so it is a false reference. But many people might have the intuition that this citation could be a mistake or a flub, not fraud. Why? Because Pepinsky does argue things like that,* and an author might sincerely be attempting to credit him with that argument, which would be a valid authorial claim. If your job exposes you to false citations, one argument that you will routinely hear is that a well-meaning people accidentally try to do the right thing and fail, and that cannot be fraud, it must be something else: a bad research assistant, a bad AI, or bad reference management.
But, as I have shown, every citation makes both an authorial and an existential claim. This false citation is fraud because the existential claim is false, and the author makes that claim, not the RA or the technology. The authorial claim is wholly irrelevant.
Who Cares?
As noted above, fraud also must cause someone else to rely on that statement, and result in a harm. This is easy to demonstrate in the case of a false citation in a piece of research submitted to a journal, press, colleague, or teacher. The purpose of the citation is to stipulate to the reader that a document contains a relevant statement (reliance). It may lead the reader to make decisions or take actions that they would otherwise not make, or to hold beliefs about the veracity of claims in the manuscript that are false (harm).
If a research output were held in private, or shared with an explicit proviso that the citations are not valid, then either reliance or harm no longer applies, and the false citation is not fraud. I am not aware of any research outputs that are privately held or explicitly tagged as false.
In summary, any plausible mechanism that generates a false citation in a research output is fraud. Whether the author deliberately deceived, recklessly deceived, negligently deceived, or truly but recklessly believed, each of these constitutes fraud.
NOTE
* Here’s a valid citation if you want one: Thomas B. Pepinsky. 2014. “Authoritarianism: Institutions and Logics.” APSA Comparative Politics Newsletter 24, no. 1 (Spring), 1–4.
