UlrikeHahn

(U Hahn, Jan. 2/2023)

There is currently lively debate on Mastodon about whether or not to introduce QT's – a feature that many considered integral to the Twitter experience but that Mastodon has (to date) eschewed on a variety of grounds, most notably concerns about “dunking”, “pile ons”, and other mis-uses, but also notions of consent, wanting to promote conversation with people instead of about people, and so on. The point of this blog post is not whether Mastodon should or should not introduce QTs. The point is to consider some arguments on QTs floating around on the platform, and evaluate them from an argumentation theory perspective. Why that's useful or interesting, I'll say more about at the end; for now, I'll just offer the intuition that basing decisions on good arguments as opposed to bad ones is more likely to lead to good outcomes.

The four arguments (reasons) are:

1. Blocking QTs won't stop bad behaviour

2. QTs are being blocked without evidence that they promote harm

3. We need QTs

4. QTs will happen anyway, so we should focus on making them safe

I'll go through each of these in turn. All are interesting, and each analysis shows something quite different. Some will be more technical than others (1. and 2.), so do read on if the first two are too geeky, you might still find the next two more interesting. So, without further ado:

“Blocking QTs won't stop bad behaviour”

The argument is trying to establish a reason for why, causally, blocking QTs is not effective (in plain English, blocking QTs won't do what it’s meant to do).

For any such causal claim we can distinguish between necessary and sufficient conditions. Necessary conditions are ones that must be in place for an effect to occur; sufficient conditions are ones that are enough for an effect to occur.

QTs might or might not be sufficient (on their own) to generate 'bad behaviour', but nobody actually seems to be claiming that only QTs generate bad behaviour or that bad behaviour isn't possible without QTs (and such claims would be fanciful given that Mastodon instances still require moderation!).

In other words, QTs are clearly not necessary for bad behaviour, bad behaviour arises (also) through other means.

Given that everyone arguably understands this, the argument has a whiff of 'straw man' fallacy about it (that is, a spurious attempt to refute a position nobody actually holds, see e.g., Woods et al., 2004).

What is really required is evidence to assess the causal role of QT: to what extent is the ability to QT sufficient for generating bad behaviour? For causal claims, that ideally involves comparisons across cases with and without QT (see e.g., Lagnado et al., 2007).

This brings me to Argument 2 about the presence or absence of evidence.

Before that, though, a shout out to another reason why people might genuinely think Arg. 1 is better than it is, namely that it looks like a type of argument that is really strong ('modus tollens' – this might be too technical for most people's interests, so I've put it in an Appendix at the end). Onwards to Argument 2:

“QTs are being blocked without evidence that they promote harm”

There are two aspects to this argument: a) whether it is factually true (i.e., is there relevant evidence or not?) and b) if there is no such evidence (or only insufficient evidence) is this a good argument and why?

Is there really no evidence?

There is a relevant observation: Mastodon was designed to minimise certain types of bad behaviour, and, those types of bad behaviour do, indeed seem pretty rare. But what is really required is evidence that compares with and without QTs, and that is lacking in as much as any comparison, say, between Twitter and QT is confounded by many simultaneous changes. Furthermore (non-argumentation theory aside, here...) studying such systems and trying to identify causal effects and their magnitude is really hard, because online communication systems involve many interacting components that may modify each others effects. Science has rather limited tools, at present, to study such systems (see e.g., Bak-Coleman et al, 2021), and establishing test bed communities where one could properly conduct such research is a priority for many, but costly and difficult to realise (a CERN model for studying information environments, see e.g., Lewandowsky et al., 2020; Wanless & Shapiro, 2022). So we seem unlikely to have terribly compelling evidence in the short term. This makes the argument extremely relevant. But is it compelling?

Given insufficient evidence, is this a good argument?

The argument is about what argumentation scholars call the “burden of proof”. These are familiar from other contexts in which we have to ultimately make decisions, in particular from law. In law, a failure to meet a burden of proof will lead to dismissal of a case. Argumentation scholars have adopted the notion for many purposes. For example, it has been argued that the proponent of a claim holds the 'burden of proof” in as much as it is up to the proponent to provide evidence for a claim, when challenged (see e.g., van Eemeren and Grootendorst, 2004). Crucially, that is intended to be an epistemic burden of proof (trying to spell out what it means to 'win' the argument or 'what to believe' in light of the evidence). One can take issue with that idea of epistemic burden of proofs (see e.g., Hahn & Oaksford, 2007), and argue that what we should believe is what is congruent with the evidence (and if that evidence says, 'we really don't know', that's exactly what we should believe). Importantly, that's not what Argument 2 is, however. The argument isn't “you don't have evidence, so you shouldn't believe that QT is harmful”, the argument is about what should happen, given that we don't know. With what action the burden of proof should lie, and how high the burden of proof should be set, depends on all kinds of value judgments, not just facts and evidence about the world. In the case of criminal law, the burden of proof is set to reflect the fact that society thinks it is worse to wrongly convict an innocent person, than to wrongly set a criminal free.

Many different values and value judgments could be invoked to argue about who should hold the burden of proof vis a vis QT and why: you could argue that it's fair that the people who designed, built, and supported the space can require evidence that their intuitions are, in fact, wrong, and without that, everything stays the same. Or you could argue that online toxicity is so damaging that, unless we have good evidence that a feature doesn't promote it, we should avoid introducing that feature. Or you could argue that most people want QTs (I'm not claiming this is actually true), so we should do what people want in the absence of good reason to the contrary. None of these are arguments about facts (what is true or false). They are all arguments involving values (what we like or dislike, and by how much).

Because the argument isn't providing any reason for why the burden of proof should be a particular way (ie with “blocking” or “not blocking”), it doesn't address the issue in a real way.

It's, at best, a description of a state of affairs (“I don't see any real evidence, and QTs are being blocked”), and, at worst, an attempt to claim a burden of proof (“in the absence of evidence we should be going ahead”) without actually providing a reason for why that's the way the burden of proof should be (something argumentation scholars are likely to think is unfair).

“We need QTs”

Many versions of this argument, by many individuals and all manner of groups have appeared in my feed in the two months since joining here. The salient thing about “need” is its relation to “want”.

An argument for QT, or any other feature, is typically an explicit or implicit argument like this:

“I need X for Y”

there are consequently two needs in play here: whether one really needs X to achieve Y, and whether one needs Y itself.

The former (“need X for”) is about an instrumental relationship. It is an empirical question, and the claim that I “need X for” is false if there are other ways for me to achieve Y.

The latter (“need Y”) is fundamentally about importance, that is, a value judgment. It is false when the claimed importance doesn't match the actual importance to myself. And, if I don't need Y (I merely want Y), then I also don't actually 'need' X, even if X is genuinely the only way to get Y.

Y is also fundamentally about values in a further sense: in almost any interesting real world context, different needs conflict. (Here, the absence of QT wasn't an oversight, it was an intentional design choice for a particular goal, so, presumably, reflects a felt “need” to not have QT). The key underlying issue will thus be how these different competing needs should be resolved in this case, and why. This may include coming to agreement that some lesser feature W, which is less effective at bringing about Y, may nevertheless be preferable given the overall balance of needs.

“I need X for Y” is a potentially powerful argument for QT, but, stated as such it is also still just a claim. Further arguments or reasons are required to establish that need. So what better arguments legitimately can (and should) focus on is the empirical question about the effectiveness of X for causing Y (see above!), which can include both discussion of necessary and sufficient features, and, at the same time, better arguments can (and should) focus on good reasons for adopting a particular balance of needs.

“QTs will happen anyway, so we should focus on making them safe”

The first thing to note is that, far from being the exception, 'bad things happening anyway' is the normal way for bad things! We have vast legal systems (from civil law about contracts and what happens when people break them, through things like traffic violations, all the way to serious crimes) only because the bad things they try to regulate “happen anyway”. If they never happened we wouldn't need those rules, and if the rules totally stopped them, we wouldn't need to expend all that energy on specifying what happens when they are nevertheless broken, which is the majority of what the rules actually do.

As a result, basically nothing follows from the fact that QT is a potentially bad thing that will happen anyway, because there is nothing special about that fact.

For the vast majority of bad things, the way we try to deal with them is by defining clear rules to stop them and combining those with enforceable sanctions for violations that nevertheless occur. So why should QT be different? That is the argument that needs to be made.

There are, of course real world cases where we simultaneously maintain that a behaviour is harmful and sanctioned, and nevertheless make accommodation in practice to further reduce harm: examples might be how we regulate sex work, or the fact that one might ban certain drugs but still offer needle exchanges. These kinds of cases strike me as being characterised by the fact that potential 'perpetrators' (if we ban these behaviours) are simultaneously, in some sense, victims who are harmed. But all of that, and whether it has any bearing on QTs, is beyond the realm of purely evaluating arguments.

The argument evaluation point, here, is simply that nothing much, if anything, follows from “QTs will happen anyway”. As a result it doesn't add much of reason for thinking about “safer QTs”.

Thinking about safer QTs is already a project anyone is entitled to engage in, the same way that anyone is entitled to argue that QTs (or newly safer QTs) have benefits that outweigh the costs. Given that this is what any argument about competing needs or desires will ultimately come down to, it seems preferable to just provide arguments directly for that.

That avoids not just misdirection, but also the slight hint of trying to portray the debate as already closed, when argumentation theorists, at least, would still consider all of these issues up for legitimate debate (for more theoretically souped up versions of that sentiment in the form of a 'freedom rule' governing rational discourse, see e.g., van Eemeren and Grootendorst, 2004).

Who cares?

Finally, is there a point to all of this and who cares? Well, scholars interested in the question of whether or not we can say things about the quality of arguments that aren't just wholly subjective care. Are there things about argument quality one can say that go beyond subjective preferences (I like chocolate, you like vanilla)? Attempts such as the above are attempts to show that there is. The intuition that there should be, at least some, more objective criteria is widespread. In line with that intuition, people in a wide range of disciplines (education, psychology, critical thinking) care about arguments that seem more convincing than they are (also known as “fallacies”, see e.g., Woods et al., 2004; Hahn, 2020). And, on some level we probably all do. And while we may not care much if others are wrongly bamboozled by our own arguments, we generally don't want to be bamboozled ourselves. And, finally, we have strong intuitions that bad arguments are unlikely to lead to good decisions; and conversely, if something *is* really good there should be good arguments one can find for it (and there are ways to make those intuitions theoretically more precise).

So, focussing our efforts on trying to find good reasons for what we want to do, instead of getting caught up in bad ones, should hopefully benefit us all. And it seems particularly relevant when we are trying to build a community and negotiating what we want to do.

Appendix

Argument 1. might be confounding people also because it looks a lot like a logically valid argument (“modus tollens”):

If the snark is gorped, then the squaggle is glibbed.

The squaggle is not glibbed

therefore, the snark is not gorped

To say that this argument is logically valid is to say that one has to accept its conclusion if one accepts the premises (by necessity, otherwise one's position is self-contradictory), and that holds regardless of the specific content, just because of its form (which we can see by the fact that it is true even for nonsense words). So, from that perspective this looks like a really great (maximally great, in fact!) argument:

If we block QTs, we stop bad behaviour.

Bad behaviour isn't stopped

therefore, we don't block QTs

But that would be to misconstrue the argument and the claim – that argument 'works' in the case where we believe the premise “if we block QTs, we stop bad behaviour”, which is exactly what is being questioned here.

References

Bak-Coleman, J. B., Alfano, M., Barfuss, W., Bergstrom, C. T., Centeno, M. A., Couzin, I. D., ... & Weber, E. U. (2021). Stewardship of global collective behavior. Proceedings of the National Academy of Sciences, 118(27), e2025764118. https://www.pnas.org/doi/10.1073/pnas.2025764118

Eeemeren, F.H. van & Grootendorst, R. (2004). A systematic theory of argumentation. The pragma-dialectical approach. Cambridge: CUP.

Hahn, U. (2020). Argument quality in real world argumentation. Trends in Cognitive Sciences, 24(5), 363-374. https://www.sciencedirect.com/science/article/pii/S1364661320300206

Hahn, U. & Oaksford, M. (2007) The burden of proof and its role in argumentation. Argumentation, 21, 39-61. https://link.springer.com/article/10.1007/s10503-007-9022-6

Lagnado, D. A., Waldmann, M. R., Hagmayer, Y., & Sloman, S. A. (2007). Beyond covariation. Causal learning: Psychology, philosophy, and computation, 154-172. https://books.google.de/books?hl=en&lr=&id=5I4RDAAAQBAJ&oi=fnd&pg=PA154&dq=d+lagnado&ots=OmE23ZHia3&sig=VB9vFUD4Wjqwyl0gsGV_58HBBOg&redir_esc=y#v=onepage&q=d%20lagnado&f=false

Lewandowsky, S., Smillie, L., Garcia, D., Hertwig, R., Weatherall, J., Egidy, S., ... & Leiser, M. (2020). Technology and democracy: Understanding the influence of online technologies on political behaviour and decision-making. https://pure.mpg.de/rest/items/item_3277241/component/file_3277242/content

Wanless & Shapiro (2022) A CERN model for studying information environments https://carnegieendowment.org/2022/11/17/cern-model-for-studying-information-environment-pub-88408

Woods, J., Irvine, A., & Walton, D. N. (2004). Argument: Critical thinking, logic and the fallacies, Revised Edition. Toronto: Prentice Hall.