Harm reduction complications

I'm not going to teach you anything by stating that social media's a fucking mess. This time, I want to hone in on the highly specific type of mess that results when covid-cautious people and mask bloc activists run into veteran sexual health advocates, most often cis white gays. The mess always centers on a specific sentence: "Shame does not work". This doesn't just happen online, of course. I've run into it IRL more than once.

I want to dive deep beyond the angry rants, because something eerily similar is happening in conversations about LLM usage. As my jumping-off point, I'll use this mostly great video shared by the always-excellent Pizza Pachulia: "How to (Anti) AI Better", by Dr Fatima. Do take the time to watch it in full if you can. It will likely annoy you. It makes some solid points. It also has a number of issues - which is why there's an entire wall of text below.

What's a harm reduction anyway

This will as always not be a formal, strong definition: I'm a queer anarchist code monkey with a blog and extensive mask bloc'ing experience, not an academic. My understanding of the term is rooted in the advocacy I do, the conversations I have had, and the theory I've magpie-assembled from fragments gleaned across social media, podcasts and articles.

Harm reduction is a non-punishing framework that is notably deployed on the topic of drug (ab)use. It is rooted in scientific and empirical findings that, indeed, shame does not work to reduce risk-taking behavior. The framework starts from the proven assumption that people are more likely to tell you about the risky things they have done if they do not fear your reaction.

When it comes to drug use and notably addiction, it acknowledges that just telling people to stop is typically counter-productive: it instead focuses on preventing harm that can be caused to or by the drug user. It encompasses actions such as advocating for decriminalization of drug use, providing free clean supplies and safe usage sites, and generally avoiding judgment.

You will also encounter harm reduction concepts in sex worker advocacy. We have extensive evidence that bans applied to either the sale or the purchase of sex make things significantly worse, including for victims of sex trafficking. Discussing this in detail is out of scope for this article, but if you think the Nordic model is a good idea, please go do some research, the data's horrifying.

Harm reduction also comes up in sexual health advocacy, notably from organizations fighting against AIDS and supporting people who are HIV+. Historically, this has meant nasty debates about "irresponsible" sexual practices. Online and in the present, it is quite often (generally cis and white) gay men promoting PreP, reminding everyone that Undetectable = Untransmittable, and so on. The angle is, once more, to avoid judgment, to support people in whatever practices they decide to partake in, to provide supplies and information. This particular group is the one that's been driving me and my fellow queer anti-covid activists batshit for years and is therefore at the root of a lot of my thinking on the topic: we'll get into that.

The video by Dr. Fatima that I linked above recommends deploying the harm reduction framework for LLM users as well. It explains that it is better if people tell you about their use of LLM products instead of hiding it. It also argues that having those conversations is important because it allows you to figure out what the underlying problem is, rather than shutting down avenues of contact. This is a line of thinking that is being deployed more and more often as we get deeper into this brave new world of "AI" software eating the world: Anil Dash published a post with a similar angle just as I was revising my first draft of this piece.

This is however not the only way that the harm reduction language is deployed in a LLM context, and that's where the trouble begins. But we'll get to that after digging just a bit deeper into the benefits.

Okay, so it's good?

Yes! Harm reduction is a very good framework, it works, it saves lives and it's extremely important. It absolutely applies to covid precautions and LLM use... to some degree, with caveats.

Avoiding judgement is a vital tool in interpersonal outreach and supporting people, specifically. This is primarily because people who are ashamed will, as mentioned, avoid telling you things. This, in turn, makes conversations impossible. If you can't have a conversation, you can't offer resources, you can't help, you can't diagnose problems.

Let's list some more concrete elements.

When it comes to covid precautions, you want to avoid judging people because there are numerous barriers to taking said precautions. People might be lacking information you could provide. They might be experiencing significant social pressure not to mask. They might have asthma and feel like they are choking with most respirators on the market. They might, like all of us, be carrying a bucketload of unprocessed trauma about the whole mass death and abandonment thing that is genuinely difficult to look back at, making it easier to pretend it's all over. Respirators and air purifiers might be too expensive: this is why mask blocs exist, after all.

You also want to avoid judging people because open conversations about health, contagiosity, risk-taking and testing are how we avoid further contaminations. In this regard, the teachings from HIV and STI experts are essential.

When it comes to LLM usage, you want to avoid judging people because there are many good reasons one may reach for a bad tool. Social pressure again. Endless administrative bullshit. An already-shameful situation you really need help with but don't dare tell people about. Communication difficulties making a chatbot seem easier to talk to. Lackluster documentation. So much fear, so much stress, in an era where support networks have been shattered and rebuilding community seems harder than ever. It's vital to leave the door open for those who want to quit.

This is where the comparisons with sexual health and drug use have their limits: there is a difference between someone who uses drugs or has unsafe sexual practices to compensate for a terrible situation that needs fixing, and someone who does so purely recreationally. That difference is not always easy to figure out and this is obviously yet another false binary. The key point is that people dropping acid for fun or having consensual, cheerful unprotected group sex is not actually a problem we need to fix; but asylum seekers resorting to ChatGPT because no one will help them to write yet another letter demanded by abusive bureaucracy is obviously a problem we need to fix.

Any competent harm reduction advocate is surely aware of that. Right?

Imagine being Act Up and against masks

The problems start when "Shame does not work" gets deployed as a form of tone policing. This is partially brought on by the quirks of social media, where the line between venting and advocacy get really thin (case in point: my Bluesky account). But I think we should stop pretending that it is not also brought on by some harm reduction advocates using the framework as a convenient way to excuse their own harmful behavior.

Which brings us back to the self-declared Bad Gays taking potshots at covid cautious posters.

The thing with airborne pathogens at large and covid in particular is that a person's careless attitude goes beyond just them. I've written about this at length and won't renegotiate it here. And so what even counts as shame gets complicated: you should not be shamed for a lack of precautions, but explaining how your lack of precautions puts others at risk is easily construed as shaming. It is more than fair for me to express anger at an inaccessible world, but that very anger can be read as shaming those that won't make an effort to fix things.

I will not pretend that the posts that spawn this discourse are always crafted with care. I am deeply frustrated by a lot of the rhetoric and tone that gets deployed in favor of covid precautions, because I do believe it makes my advocacy work harder. I only wish to stress that there is a grey zone of interpretation, where informing people about the harm that their actions can cause - that their choices can cause - is quickly denounced as bad, because "Shame does not work".

Meanwhile the same fuckos saying this are cheerfully posting unmasked selfies from HIV conferences. I have questions. I'm not sure I want answers. I can't help but wonder how much of this is colored by racism, ableism and misogyny and transphobia, considering the typical demographics of mask blocs.

This is doubly heartbreaking. For one, those are people I would in theory like to be in community with, if they weren't such jerks about it. But their attitude also hinders the work: those of us on the covid beat would benefit greatly from their experience and knowledge, especially when it comes to the cruel task of breaking bad news about their own health to someone.

Surrender factory

LLM usage is on first glance less dire in that regard. You're not rolling the dice on maybe killing grandma by using a chatbot, most of the time. The key element to me is that LLMs available to the general public are very new: they are a new curveball, a new complication, a new harmful behavior that feels like it should be easily prevented. All of it builds on existing problems, yes, but we were working on addressing those. We had answers. We had no need for any of this. Nobody was asking for it.

AI startups have a well-documented interest in making us think this is the new normal and it's too late to say no. (See transcript in this piece which also spoils all of Outer Wilds for some reason). And so the harm reduction approach gets extremely slippery because it can easily become a form of surrender to a status quo that is not actually settled.

Alex Hanna and Emily M. Bender - co-authors of The AI Con - get into it in-depth with their guest sava saheli singh in episode 82 of Mystery AI Hype Theater 3000 . There is an active, co-ordinated push towards using "harm reduction" to mean "we need to teach everyone to use those tools". As they discuss, those initiatives fail to define who is being harmed, how they're being harmed, and what can be done to reduce that harm. At no point do said initiatives consider the option of just not using the fucking thing. sava puts it well: under the pretense of keeping students safe, widespread deployment of LLM products pushes surveillance on students instead of listening to their actual safety demands, such as not getting their skulls bashed in while protesting for Palestine.

For another angle on a similarly bad simile, Maggie Fernandes has a detailed denunciation of comparisons to abstinence-only sex "education". The article notably stresses that abstinence "sex ed" is about control, fear, and disinformation, certainly not about empowering teenagers to critique and refuse sex they don't want to have. Critical LLM refusal is about empowerment. It is the very opposite of oppressive.

People also forget faster than you'd think: just ask them about 2020. We need to be very, very careful in how quickly consensus can be installed when there is this much money and this much effort behind it. We should not judge individuals for their LLM use. But we should absolutely judge individuals and organizations for their LLM advocacy, and we should in turn advocate for a full ban on consumer-facing LLM products such as those currently on the market. We should not direct people towards individual use of different LLM products, but towards solutions that are community-oriented and LLM-free.

Dr. Fatima's video fails to make this distinction as sternly as it should and ends up deeply frustrating for it. When discussing options for political action, it also makes the unfortunate choice to direct people towards organisations that are peddling an "existential AGI/ASI risk" angle on banning LLMs: those are the TESCREAL people and you should not give them a single molecule of oxygen.

You can still have standards

The harm reduction framework is vital to provide support to people. It does not always apply tidily to deciding what rules should be put in place for a community, for a group or an event. It applies even less tidily to deciding what policies and laws we should advocate for. This is not an especially difficult distinction to draw if you've been in the game for any length of time: be highly suspicious of anyone who seems remarkably keen on claiming expertise, but utterly fails to make said distinction.

Drafting policies and standards is difficult, mind you. As Maggie Fernandes linked above stresses and as Sonja Drimmer puts it, "a policy requires policing"; and by explicitly banning something, you acknowledge its relevance. The approach Sonja Drimmer suggests instead is to reward engagement with engagement, and disengagement with disengagement. This is easily interpretable as a form of punishment, but it does not have to be, so long as the door is always left open for change and requests for support.

In some contexts though bans and other hard requirements are the way to go. The history of smoking bans in particular is worth looking at - Olivia Guest has some worthwhile material on the topic - because smoking does not just harm the drug user, it harms everyone around them. And so it had to be banned in public places in the name of public health. Tobacco companies fought hard against this through everything from marketing cigarettes as cool (including to children!) to seeding the scientific consensus with junk papers full of conflicts of interest.

The key is to think of harm reduction as a tool of collective safety and community-building rather than something strictly individual, even as it applies to one-on-one interactions quite often. Making some harmful behaviors unacceptable in social spaces can be a great way to encourage positive change, after all: if someone wants to join, all they need to act on is do better while in the space.

Obviously implementation is non-trivial. How do you decide what the rules are? How do you communicate them, in advance and on the spot? How do you enforce them? How do you react to problems? How do you cleanly avoid painting someone as a bad person just because they cannot follow the conventions of the space? Codes of conduct for conferences and best practices for social media or forum moderation can be a place to start thinking about all this. Writings about police abolition are another direction worth digging into.

What does that look like in practice?

In a previous piece, I have mentioned pedagogy mode and rage mode. That idea can be expanded into some seemingly-obvious questions: think about what you want to achieve, who your audience is, what the context is.

If you're talking to someone you wish to be or have to be in community with, you will get strong benefits from applying as many harm reduction principles as possible. This is because one of your key goals is typically to preserve the relationship. In a workplace context, this is also known as "keeping your job". You will want to avoid judgment and to provide help and resources and tools. You will want to get to the bottom of the underlying problem, and see what can be done to address it. This can be emotionally draining: remember to surround yourself with people who are already on your side, people you can debrief and vent with. And remember that you do not have to pick every single fight - you want to foster community and relationships, and this includes finding allies and accomplices.

A practical example: No matter how infuriating it is, I strive to maintain open channels of communication with my coworkers who are using LLM products. I am vocal in my opposition, I am actively demanding that some topics be addressed at the company level. Part of why I am vocal is so those who share my views can see that they're not alone, even if they do not feel safe speaking up.

If you're crafting rules for a space, a community or an event, then you can be a lot more strict. What behavior is acceptable? What is welcome and unwelcome in a given space? A good line can be "We don't do that here". At the scale of society, there need to be spaces for everything - but your space does not have to be all things to all people. Access conflicts are a thing. You likely cannot offer a space that lets people use drugs and lets people bring their young children at the same time. That's okay.

A practical example: We mandate masks at the trans support meet-up, with leeway for those who literally cannot wear one. This is non-negotiable. It is not judging individuals to have this rule for this one event, especially when near-everything else does not give a fuck.

If you're advocating for changes to laws and policies, you need to be both ruthless in your convictions and keenly aware of what the wider game board looks like. Harm reduction for drugs at the level of policy is about decriminalizing users and securing funding for support resources, not letting providers get away with murder and certainly not about actively promoting drug use. Be extremely careful about why and how someone opposes or supports specific demands. Beware ends justifying the means and unexpected circumstantial alliances: you know what they say about the road to hell and good intentions.

A practical example: You'll see people push for social media bans for children under the guise of "preventing harm". This is factually incorrect, actually pushed by the age-verification software lobby, and enabling mass surveillance. Don't take my word for it: instead go read people like Heather Burns, who have been slowly losing their minds in those conversations for many years.

In short, harm reduction is a land of contrasts

I wish I could just blame social media.

Those questions I encourage you to ask above - what am I trying to achieve, who is my audience, what is the context - get blurred automatically on social networks. I am venting, but also doing advocacy. I am doing advocacy, unless I'm actually shitposting. The audience is my followers unless a post escapes containment. The context is that I saw a post that annoyed me but, due to context collapse, people will assume I saw a different post and am being mean about it.

I praise pancakes. I am assumed to hate waffles.

But I've encountered plenty of bad-faith deployment of harm reduction in real life too, especially from queer people and - I stress once more - especially from white, cisgender gay men. I think they are in a uniquely tricky position of being at the exact distance from power that makes them think they have self-interrogated enough to avoid reproducing violence, while still being plenty well-equipped to reproduce violence.

That said, doing the work literally anywhere else than on social media feels easier. Write long blog posts instead (hi). Have conversations, whether in private or in public, online or IRL. Fill a fridge, move some boxes. Gather info and make a zine. Join an org, make sure it doesn't suck. Figure out what harm is happening and how to reduce it.

Create the org if it doesn't exist. Make sure it doesn't suck.