3.1. Harm-based account

Let us begin with one account that has gained much traction in recent years, not only in academic discourse but also gray literature, which is that of dis/mis/malinformation. For instance, it is adopted by the American’s Cyber Defense Agency,Footnote ⁵ Institut national de la santé et de la recherche médicale (Inserm),Footnote ⁶ the Canadian Centre for Cyber Security,Footnote ⁷ and the Council of Europe. Footnote ⁸

According to this widely adopted classification (Wardle & Derakhshan Reference Wardle and Derakhshan2017):

• Misinformation is when false information is shared, but no harm is meant.

• Disinformation is when false information is knowingly shared to cause harm.

• Malinformation is when genuine information is shared to cause harm (5).

Some have questioned the coherence of “false information.” They hold that information is true by definition and as such “false information” amounts to a contradiction. A notable proponent of that view is Luciano Floridi (Reference Floridi2013) according to whom information is data that is well formed, meaningful, and truthful (31). As such, he construes misinformation as “semantic content that is false” where “semantic content” is a shorthand for well-formed, meaningful data (2010, 50). Others have contended that “information” does not require truth by pointing out that computer scientists and cognitive scientists use the term in a way that does not imply truth (Scarantino & Piccinini Reference Scarantino and Piccinini2010).

It is not difficult to reconcile both by views by suggesting that “information” can have a loose and a strict sense, and only the strict sense requires truth and it is this strict sense that concern us here. For our purpose, then, one could simply replace “information” in the definitions above with “semantic content” – I sometimes will use “content” for short. This leads to the following:

• Misinformation is false semantic content that is shared but no harm is meant.

• Disinformation is false semantic content that is shared to cause harm.

• Malinformation is true information shared to cause harm.

It should be clear then that this classification takes place along the two axes of truth and intention-to-harm as shown in Table 1.

Here, we have three (four) key concepts, “truth”/“falsehood,” “intention,” and “harm.” Everyone in the literature agrees that truth is an important ingredient in discussion of dis/misinformation. Although I largely agree with this diagnosis, I think that in many important cases this criterion can be problematic, but discussing that is for another paper. This leaves us with “intention” and “harm.” I take issue with both.

Let us start with harm. What exactly do we mean by it? In their report, Wardle & Derakhshan mention the “harm” 21 times, but they do not attempt provide a definition or even a characterizations of it. They do say that harm can be caused to a person, a social group, an organization, or a country (2017, 20). They also provide some examples of where they think dis/misinformation has caused harm. For instance, they mention a disinformation campaign that was meant to damage Emmanuel Macron’s chances during the 2017 French Presidential election, including a fraudulent article purporting that he was funded by Saudi Arabia (21).

Here we can ask what was that harm intended, and to whom exactly? One might think that the answer is rather trivial, namely that it is intended to harm the image of Macron. But can we be confident of that? What if the source of disinformation was of right-wing leanings, and whose aim was to support Marine Le Pen by all means necessary, meaning that they spread disinformation with the original intent to benefit Le Pen even if that harms Macron? In this case the intent to harm Macron, if at all existent, is only secondary to that of benefiting Le Pen. Or what if the source of disinformation thought Macron unfit to act a president and as such spread disinformation with the intent to benefit the French republic?

Take other examples, such as someone who is spreading that Covid is a hoax due to her distrust of the government, or an antivaxxer who knowingly or unknowingly is spreading “false information” because she genuinely thinks that vaccines are hazardous to one’s health. Neither of those has any intention to harm, although they certainly have the capacity to harm. The problem with “harm” in the definitions of dis/misinformation, as I see it, is that it cannot be dissociated from the intent of the source – we are thinking of the intent to harm. Also, the source may not share our same values, so what we may take as harmful they may not. What makes things even worse is that, from an epistemic perspective, intentions are almost always opaque – call this the problem of intention opacity or “the problem of opacity” for short. We are seldom in a position to assert with confidence what someone’s intentions are.

But even if we are confident of someone’s intentions – say they explicitly state it, if we go back to the Macron’s example. Let’s say someone does spread information because she wants to benefit Le Pen, and she strongly believes that the survival of the French republic depends on Le Pen being elected. In this case, Macron is a just a place-holder for whoever is running against Le Pen. As such, the harm done to Macron is strictly accidental.

A plausible response here is to point out that the same actions may have more than one intent, some of which are internal to the action of disinformation itself, such as distorting Macron’s image, and other external ones, such as saving the republic, and that it is the internal aim that is relevant when discussing harm. Although I find it to be a strong response, I still think that it is not entirely convincing because it fails to appreciate that when considering intentions, what matters more is not whether intentions are internal or external to particular actions, rather it is which intention has precedence in the mind of the actor. So even if we grant that the intent to distort Macron’s image is internal to the action of disinformation, the intent to, say, benefit Le Pen is not an external one. It, in fact, takes precedence in the mind of the actor over the internal intention of the action and, in a way, conditions it. That is why I would still maintain that, the objection notwithstanding, distorting Macron’s image may be made with no harmful intent even if we know it will most likely have caused harm.

One may try to improve the harm-based account by restricting the kind of harm involved in dis/misinformation to the epistemic kind, as opposed to, say, the physical, social or otherwise. This is certainly better than simple harm-based accounts, and it seems to be the (epistemic) direction in which later (misleading and ignorance generation) accounts are going. However, it still has the same drawbacks as other accounts that emphasize intent. To elucidate, consider the case of a “noble lie,” which is falsehood deliberately propagated for a presumed greater good. A typical example is one propagated by the political elites to maintain social harmony. I find it difficult to accept that in such cases epistemic harm is in fact intended. Or take the more epistemic example of a noble lie, where an almost always reliable source of information, once having spread misinformation, decides to cover up with a lie to protect its credibility in the face of other very unreliable sources. In such, and similar cases, “disinformation” cannot even be said to be intended to cause epistemic harm, as those trusting this source are epistemically better off continuing to rely on it for information rather than others. Could the epistemic harm account be salvaged? I think it can be, if we distinguish between content and activity as well as introduce “capacity” into the definitions, all the while hyperspecifying the epistemic harm at stake in each situation, but then what we will end up with is very different from the current understanding of harm-based accounts, and will look more like my information-hazard-based account so I will not discuss it further.

From what has been said so far, including the problem of opacity, the disagreement over what counts as harm as well as the moral and legal stakes that hang on these, we are led to conclude that the “intent to harm” is inoperable and as such unsuitable in delineating dis/misinformation. The emphasis on operability cannot be overstated. That is because, as noted earlier, the recent explosion in the dis/misinformation literature is largely due to practical consideration. There is a general sense of urgency, and a need for action in the face of what some have even dubbed as an “information pandemic” or “infodemic” (Zarocostas Reference Zarocostas2020). As such, we not only want that our concepts be clear but also that they can be put to good use in addressing information problems.

3.2. Misleading-based account

Problems such as these have led many scholars to drop “to harm” entirely and replace it with something more specific, namely “to mislead” (e.g. Guess & Lyons Reference Guess and Lyons2020; Chadwick & Stanyer Reference Chadwick and Stanyer2022). A notable example is Floridi (Reference Floridi2013), who defines disinformation as “misinformation purposefully conveyed to mislead the receiver into believing that it is information” (260) – where information, for Floridi, is true. Another example is one of Fallis’s (Reference Fallis2014b) definitions of disinformation as

Table 1.

A classification of misinformation and related concepts along the axes of truth and intention to harm.

It should be noted that cashing out dis/misinformation in terms of the intention to mislead already constitutes an improvement over the harm-based accounts since, unlike harm, it provides something more specified for us to evaluate, namely falsehood. It is important to keep in mind that “to mislead” in this context is generally taken to mean to cause false beliefs (see Harris Reference Harris2023). Yet, misleading-based accounts of dis/misinformation are also unduly limiting since in many instances disinformation does not simply aim to cause false beliefs but instead aims to block true beliefs. A textbook example is how a small number of scientists backed by the tobacco industry were able, despite overwhelming evidence, to sow the seeds of doubt about the link between the use of tobacco and cancer (Oreskes & Conway Reference Oreskes and Conway2011). Here, and in similar cases, the aim is not necessarily to cause false beliefs but to prevent people from forming true beliefs despite the evidence. This consideration has later led Fallis to amend his definition of disinformation to include that it is likely to create false beliefs or prevent true beliefs, but even this, as will be clear, is also too narrow (2015, 420).

However, and notwithstanding its improvements, the misleading-based account also fails to address the problem of opacity. Also, and not unlike other accounts that lump content and activity, it suffers from the possibility of cross-classifying misleading semantic content. Is my grandfather spreading misinformation or disinformation when he shares a deliberately misleading Tiktok video on the side effects of the Covid-19 vaccine? The answer is not straightforward. This is because disinformation is sometimes taken to be distinct from misinformation and sometimes a subspecies of misinformation – depending on whether or not intentionality is included misinformation (see Fallis Reference Fallis2016).

All of this has led some to suggest that we drop intentionality altogether from our definitions and classifications of dis/misinformation (Corce & Piazza Reference Croce and Piazza2021). Others, such as Fallis (Reference Fallis2014a), suggest that we can do away with intentionality by replacing it with “function,” in that although such semantic content may not be intended to mislead, they nonetheless have a function of misleading. Here, Fallis is taking a page out of evolutionary biology, according to which function needs not involve teleology. But why is intentionality so important for most authors despite its problematic status?

The answer to me seems to be an ethical one. That is while false semantic content is spread readily, we want a way to distinguish between honest mistakes, and deliberate distortions, and intent plays a major role in that. I do not want to take away from this ethical consideration. But I take issue with the fact that it has been made integral to what I consider to be a largely epistemic but more importantly ontological/conceptual issue. As noted earlier, I do not intend for ontology to be understood in any deep sense, but I do want that our concepts and classifications be answerable to reality. Failure of current definitions and classifications to do so has led to the difficulties we have seen regarding the proper place of intentionality within our definitions of semantic contents.

Let’s start with a question. Why is it that when speak of “information,” that is true semantic content, we do not bother with intentionality? The answer to that is that semantic contents to the extent that they are contents, be they of propositions or representations – I am not interested in taking a position on this, are not the kinds of things that include intentionality. But when we speak of information, not as content but as a communicative act we then include intentionality. This distinction between content and activity is something I already alluded to earlier when I attempted to specify the current information problem as the undermining of the epistemic aim of information as an activity. Here intentionality determines the aim of activity. But I also said that what I suggested as a modest epistemic aim of information, which I construed as to seek and accept truth and/or avoid and reject falsehood, is actually a bare minimum condition for successful information as a communicative activity. In a nutshell, “information” can denote two ontologically different things, one is content which does not include intentionality and the other is activity which includes intentionality.

Bringing what we have learned so far to the problem of defining and classifying dis and misinformation, we can see where almost everyone has gone wrong. That is because they have mixed content with activity which has resulted in such “disfigured concepts.” Let us look at “misinformation.” It is currently used, I said, to denote false semantic content that is not intended to harm/mislead. But, the current definitions aside, what does our pre-theoretic understanding of “misinformation” tell us it means? The answer is that it depends on how it is used in a sentence.

When I say that “Sam is spreading misinformation,” I mean something along the lines of that there is a thing that Sam is spreading, an object of Sam’s activity, which is misinformation as content. To be precise the object is not really the content itself but the bearer of such content, whether that is taken to be propositions or representations – but let us not worry too much about that here. And content qua content, I said, is independent of intentions. But if I say that “Sam is engaging in misinformation,” I do not take “misinformation” to be an object, for we do not engage in an object, but an activity. In this case, misinformation is the activity of spreading false semantic content with no intention to mislead – we are sticking with “mislead” for the moment. But what about disinformation? It seems that similar patterns of reasoning can be applied to it as well, in that as far as language is concerned, we can also have disinformation as content and disinformation as an activity, with the latter but not the former involving intentionality (to mislead).

Such observations simple though they may be, actually lead to interesting results, namely, that in so far as content is concerned both misinformation and disinformation mean the same thing: false semantic content, period – this will further be refined later. In this case, it is recommended that we drop one of these as it really is redundant. Now choosing which to drop and which to keep I think is a largely conventional issue, where a convention is understood as a solution to a coordination problem. So I will suggest that we keep “misinformation” and drop “disinformation” when we mean false semantic content. This is somewhat in line with some uses of misinformation where it is taken not to include intentionality. It follows from that that we are now able to say, as a first approximation, that someone can engage in disinformation (activity) by spreading misinformation (content). Here, we have malicious intent using false semantic content.

So what we end up with are three concepts, misinformation (as activity), disinformation (as activity), and misinformation (as content). Linking that to our previous discussion of the epistemic aim of information and information hazards, we are able to see that the undermining of epistemic aims of information can take place through activities and through contents. Granted, I said that for an activity to constitute an information hazard it will have to use contents. But it is important to tease out these two things when zeroing in on the problem because it allows us to address it with higher precision.

In fact, this distinction helps us to realize that the information problem can be broken down into two more specific but very much related problems, each perhaps requiring its own approach(es) and solution(s): the content problem of information(hazards) and the activity problem of information (hazards).

The former, that is, the content problem, concerns misinformation “in the wild,” that is regardless of who/how/why it is being communicated. It almost takes misinformation as existing as “facts” that need to be dealt with “out there.” This generally deals with them, in a post hoc fashion, that is, after their dissemination (but see Lewandowsky &Van Der Linden Reference Lewandowsky and Van Der Linden2021). The latter, that is, the activity problem, concerns the very communication activity which includes among other things, the relation between source and target, including epistemic and moral trust, expectations as well as other social psychological variables that go into communication.

The content problem, I would say, is the one that most scholars were/are concerned with, especially when they speak of “degradation,” “fragmentation,” or “pollution” of the “information environment” or similar spatial metaphors (e.g. Ahren et al.Reference Ahern, Connolly-Ahern and Hoewe2016; Kraft et al. Reference Kraft, Krupnikov, Milita, Barry Ryan and Soroka2020; Toff & Kalogeropoulos Reference Toff and Kalogeropoulos2020). This, however, they do unwittingly as they lack the conceptual distinction between content and activity, which ultimately created for them the placement problem.

I, on the other hand, recognize the conceptual distinction, and as such distinguish between the content and activity problem of information. And while, I acknowledge that solving the information problem will require attending to both of them. My “conceptual therapy” contributes to solving the content problem, if only, by exposing the placement problem that has plagued current accounts as a pseudo-problem. This, I believe, constitutes progress.

To summarize, our investigation has led us to conclude that the problem of information hazards can be broken down into a content problem and an activity problem, each of which, can lead to the undermining of the epistemic aim of information. We also ended up with three different concepts. The first is misinformation as content, which is intent-independent, the second is misinformation as activity and includes intent, and the third is disinformation as activity.Footnote ⁹ I also said that cashing out intent in terms of misleading, although an improvement on harm-based accounts, is nonetheless unduly limiting. In what follows I consider one of the most recent and more sophisticated accounts: Mona Simion’s (Reference Simion2023), which replaces “misleading” with “generating ignorance.”

3.3. Ignorance-based account

Simion, who is mainly concerned with disinformation, defines it as ignorance-generating content. More precisely, she defines it as follows:

Where to generation/increase ignorance for Simion is understood as:

I have noted earlier how misleading-based account, although an improvement on harm-based accounts, are still incomplete, especially when to mislead is understood as to cause false beliefs. That’s because I explained that one can disinform without causing false beliefs but, for example, by getting people to suspend judgment about a true claim when there’s enough reasons not to. That’s why Simion’s account centers not on false beliefs but on the generation and/or increase of ignorance. This, to me, on the whole, is a welcome addition, but as I will explain later, its extent is not fully appreciated even by the author herself.

Now although Simion takes herself to be mainly interested in disinformation, she does discuss misinformation as she argues against what she calls “disinformation orthodoxy.” The latter she cashes out in terms of three assumptions that she discusses and ultimately rejects (3).

The first assumption is that disinformation is a species of information. This, I have discussed earlier, and while I largely agree with Simion’s conclusion, I said that it is not really a substantive disagreement since information can be used in a strict and loose sense and treating disinformation as a species of information can be understood as using the latter in a loose sense.

The second assumption is that disinformation is a species of misinformation. Here, Simion tries to make the argument that the mis prefix modifies as wrongly, badly, or not. While the dis prefix modifies as a depravation, exclusion and expulsion. She remarks that dis does not simply negate, but rather it “undoes.” She provides the convenient example of “misplacing” as wrongly placing and contrasts it with “displacing” as taking out of place. However, Simion is well aware that her linguistic distinction is porous, even shaky, when she points out that there are cases where dis modifies as not such as in “disagreeable” or “dishonest.” This leaves the question of whether disinformation is a species of misinformation unresolved – I will later argue that in certain cases disinformation is misinformation.

The problem with Simion, however, is that she, like everyone else in the field, fails to distinguish between content and activity. Consider how she combines both “content” and “communication” in her definition. This is also made clear at a more practical level, when she claims that disinformation, as opposed to misinformation, is not essentially false since she is able to disinform using true content and generating false implicatures (5). The key word her is “to disinform,” which denotes a verb, and verbs refer to actions/activities and not to contents.

The third assumption that she considers and rejects is that disinformation is essentially intentional/functional. Simion rejects this almost unanimously held assumption using first pragmatic considerations and second a thought experiment. I consider both in turn. The pragmatic consideration tells us that it is ill-advised to take disinformation as intentional, when, with the advent of black-box Artificial Intelligence (AI), a certain AI system may spread misleading content with neither intention nor function. In such cases, according to Simion, we have disinformation, but it includes neither intention nor function.

Information problems resulting from AI advancement notwithstanding, it seem to me that what worries Simion is also the content problem of information. Here seems to be an implicit assumption of hers, that the content problem of information hazard is largely a disinformation one – in the conventional sense, or that the target of the problem deserving of our intervention is disinformation as opposed to misinformation. This, however, I think is a mistake, not least because I do not accept the current confused definitions, but more importantly if our worry is the erosion of truth from the information environment, then it makes little difference whether this is taken place through dis or misinformation. Next, let us consider her thought experiment.

Simion invites us to consider a trusted journalist that is interested in highlighting even the slightest of scientific disagreements, thereby making certain scientific issues seem far less settled than they really are. Examples she gives are climate change and vaccine safety. Simion maintains that this is a classic case of disinformation, which constitutes a litmus test for accounts of disinformation. Let us remember here that this example is meant to support, or more precisely illustrate, that disinformation is not, or should not be construed as intentional. After all, the said trusted journalist has no intention to mislead. Also, according to Simion, the journalist cannot be said to be spreading false content/misinformation since he/she is communicating genuine, albeit fringe, scientific findings.

Notice that for Simion, as for others, the case is dichotomous. That is, it is either a case of misinformation or disinformation, and since it is shown that it is not the former – the journalist is not spreading false content, then it must be the latter, which makes it, according to Simion, a “classic case of disinformation spreading” (5). It should be clear, by now, however, that this, in fact, is a case of false dichotomy. As I have been arguing, the distinction between content and activity makes for a richer conceptual toolkit to construe the situation. So we need not concede that it is in fact a case of disinformation to begin with. In what follows I show how my account passes Simions’s litmus test.

What activity is the journalist engaging in? On my account, the journalist is not engaging in disinformation, since the latter involves the intention to generate ignorance – here I am on board with Simion’s replacement of misleading with ignorance generation as hers is more encompassing. What he/she is engaging in is the activity of misinformation. But how so, given that he/she is not using false content? My response is that the activity of misinformation can take place through false content or true content with false implicatures – this will be further refined. Indeed, what we care about with regard to content is the capacity of the content to generate ignorance. The reason why falsity is thought important in this context is that generally false content has a higher capacity to generate ignorance, if only because, taken at face value, false content itself generates ignorance.

Back to the journalist, what is he/she doing? Answer: misinforming the public by spreading true content with false implicatures. And this way we have answered Simion’s challenge. But Simion’s challenge does help us refine our account. After all, we started off by taking misinformation as content to strictly include false content, but now we realize that condition should be relaxed – I explain how that is done in due course.

After putting forward her account, Simion provides what she takes to be a comprehensive list of the means by which ignorance is generated/increased. This includes:

1. 1- Disinformation via content that has the capacity to generate false beliefs.

2. 2- Disinformation via content that strips knowledge via defeating justification.

3. 3- Disinformation via content that has the capacity to induce epistemic anxiety.

4. 4- Disinformation via content that defeat justification/doxastic confidence.

5. 5- Disinformation via content that carry false implicatures (7).

I think that Simion’s ignorance generating/increasing list is important even if I disagree with her classification. And although I think that she does better than perhaps all of the current accounts on offer, I think that her attempt at comprehensiveness fails partly for conceptual reasons that I have been discussing (i.e. failure to distinguish between activity and content), but also partly because she fails to consider, within the information hazard framework, means that are non-epistemic but which have epistemic results. What I am mainly concerned with here are psychological means, that is, cognitive and affective. Notice that Simion’s discussion involves beliefs, justification, knowledge, defeaters, etc., all of these are epistemic in nature, and for good reasons. Ignorance itself is an epistemic concept. But people are not ideally epistemic agents any more than they are ideally rational agents.

Although Simion account is somewhat protected from such criticism when she includes “normal conditions” in her definition of disinformation – which could be said to include psychological aspects, I think this is problematic since normal conditions in communication or/and in content appraisal often involve causally relevant emotional/affective components and ignoring this point or brushing it aside undermines Simion’s avowed attempt at comprehensiveness. Indeed, that psychological, including emotional, elements are usually involved or at least can be easily brought to bear on cognition is widely recognized, especially in fields that study judgment and decision-making (JDM), such as social psychology (see Keren & Wu Reference Keren and Wu2015). Specific examples include the role of attention in JDM (Mrkva et al. Reference Mrkva, Ramos and Van Boven2020), or the role of emotions in JDM (Västfjäll & Slovic Reference Västfjäll and Slovic2013). Such considerations play important roles in the means of ignorance generation. For example, inducing someone into a state of anger or fear may render them more likely to accept ignorance generating/increasing content that is mood-congruent (see Blanchette & Richards Reference Blanchette and Richards2010). All of that is to point out that despite its attempt at comprehensiveness Simion’s list constitutes only a specific, precisely epistemic, subset of ignorance-generating means, and that a more comprehensive list would have to include psychological kinds of means.