I have read the book Politiser le bien-être by Camille Teste. A possible English translation of the title would be: “Making wellness political”.

I found the book interesting, so I wrote this blogpost to summarize it. Maybe this can help make the book's message available to a larger audience: the book is written in French, and not freely available online. Of course, these notes are just my personal work: I am not affiliated to the author, and they do not claim to be a substitute to the book.

The high-level structure of the book is in three parts:

• Define wellness as it is usually understood, and present its shortcomings.
• Analyze how and why activist movements reject wellness, even though wellness can be useful to such movements.
• Propose how to build a political form of wellness, which can address its limits, support political activism, and enable change.

Wellness and its shortcomings

Wellness can be understood as health in the far-reaching sense of the constitution of the World Health Organization: “Health is a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity.” It can also be understood as encompassing various sectors, inspired by the list of the Global Wellness Institute (a wellness trade group):

• Beauty salons, cosmetics, anti-aging products;
• Nutrition, dieting, dietary supplements;
• Exercise;
• Traditional medicine and alternative medicine¹;
• Preventive medicine;
• Mental health, self-help, meditation, etc.;
• Workplace wellness;
• Spas and thermal/mineral springs;
• The list also includes some sectors indirectly connected to wellness such as wellness tourism and wellness real estate.

These practices are a significant sector of economic activity. Indeed, they represent an economy whose worldwide value is estimated by GWI to be $4.9 trillion in 2019, with expectations of growth despite a dip during COVID. This amounts to a few percents of the world economy, which the World bank estimates to $133 trillion in 2019.

So, there is a large industry that aims at making people feel better. What's not to like about it? The book argues that wellness can be criticized in at least three main ways:

• The need for wellness nowadays is made more acute by the capitalistic system² which is powering the wellness industry:
  • One leading cause is wage labor, which pushes people to work long hours at demanding and sometimes physically harmful jobs, sacrificing their well-being.
  • More insidiously, we can suffer because of the loss of meaning at work, and the possible harmful effects of one's work on the planet.
  • Other social trends makes wellness more necessary to some groups. For instance, patriarchy has negative effects on the well-being of women specifically, e.g., mental load, social pressure (e.g., to conform to beauty standards), and of course aggression risks. So women are especially in need of what wellness can offer³. One can make similar points about racism. In other words, the wellness industry exists in a system that actively works against the well-being of some people.
• Well-being practices are normative and not inclusive:
  • They drive away many people who do not conform, e.g., physical practices (yoga, etc.) are not marketed towards people who are too fat according to dominant standards.
  • Also, they are expensive, so they mostly cater to the needs of the rich, even though they are typically not the ones who need it most.
• Some wellness practices are actually a Trojan horse for problematic doctrines:
  • Essentializing gender or biology;
  • Cultural appropriation of non-Western practices;
  • Quackery, e.g., people favoring ineffective alternative medicine over effective conventional treatments, and facing dire health consequences;
  • Taking advantage of psychologically vulnerable people, e.g., selling them expensive therapy, or enlisting them in cults.

So, the effects of the wellness industry are sef-defeating because they happen in a system that works against wellness; they are misallocated, often problematic, and mostly try to alleviate the symptoms rather than fixing the root causes of unhappiness. The book further points out the deeper problem that our perception of well-being is wrong:

• The wellness industry leads people to believe that well-being is something that they can buy⁴, reducing happiness to a financial question: you will be happy if you earn enough and consume enough. This makes us forget about alternatives such as downshifting or simple living: doing less, working less, simplifying our lives, being happy with less, etc.
• Well-being is presented as a purely individual pursuit, with approaches focusing on the self (think of exercise, dieting, meditation, yoga, etc.). This can give us a fake sense of control, i.e., it gives the impression that we can be happy through our individual actions. However, the wellness industry is in fact neglecting the collective factors of happiness: cultivating meaningful relationships, being connected, feeling helpful, etc. It is also distracting us from the political causes of our unhappiness. For instance, it put individual workers in charge of dealing with their problems (e.g., working conditions, work-life balance), even though these are better understood as a collective struggle (e.g., via labor unions).
• The individual approach to well-being comes with a form of moral imperative: you have a personal responsibility towards others to become the best version of yourself. Instead, we should see this more at a collective level (again: caring for each other), and at a political level (aiming for a system which would make it easier for everyone to be happy).

This warped perception of well-being is part of a individualistic trend where we are less and less willing to rely on each other⁵ and to believe in the value of collective action, or even in the fact that collective action is possible.

Wellness and activist movements

Having presented this picture of wellness, the book moves on to the question of how wellness is perceived by activist movements. (The book uses “the left” as shorthand to talk about activist groups that seek to effect change towards some worthy goals, but I find this a bit polarizing so will stick with more neutral terminology.) It starts by a concrete example: a wellness week-end organized by Extinction Rebellion for its members, which drew significant criticism from other activist groups:

• The spiritual overtones of such initiatives are seen with skepticism because they are perceived as incompatible with a rational, fact-based approach. This may make the movement less effective (e.g., if spiritual practices are substituted to evidence-based practices), and may also harm the credibility of the movement (e.g., on scientific issues like climate change).
• Further, they are reminiscent of organized religion, but left-wing movements (in particular marxism) typically see religion as a tool used by those in power to justify the current world order (e.g., via promises of a happier afterlife).
• Last, the use of spirituality by cults, and the closeness between wellness and cultish practices, may mislead outside observers and make them see activist groups as cults.

There are other apparent reasons why, arguably, activist movements should not focus on wellness:

• Wellness activities are less urgent and less effective than direct action, e.g., an activist group should focus on its goals first and foremost, and not the well-being of its members. Likewise, the public perception of individual activists is often that they should be ready to give everything to support their cause, instead of wallowing in self-care.
• Focusing on one's well-being is a privilege, and one that the underprivileged typically do not have. Hence, activist movements may be seen as hypocritical if they present to act towards social justice but end up focusing on wellness activities that are intended for their privileged members.
• In any case, pragmatically, in an individual sense, activism is often tiring and disheartening, so it is often not a good way to be happy.

The book proposes the following retorts to these arguments:

• The goal of progressist movements should be, ultimately, to promote some form of well-being. In this light, wellness practices are a way to stay in touch with the long-term goals of the movement. It is short-sighted and productivist to aim to effect “maximal change” without stopping and paying attention to the conditions in which this change takes place and which effect it has on the members of the movement⁶.
• We should not accept the “warrior monk” image of tireless activist leaders who give everything they have. It is probably influenced by toxic masculinity, given that many of these charismatic leaders are male. In fact, they would often be better inspired to do less, spend less time in the spotlight, and be aware of their own limits, emotions, and weaknesses. Further, activist movements should make it possible for their members and leaders to express vulnerability instead of pressuring them to project an image of flawlessness. In fact, the common perception that activists must be perfect (in French, “pureté militante”) is really a double standard that we should reject.
• Typically, activists are in need of wellness, more so than the general population. First, activism is often done in addition to a full-time job, so activists will be more busy than non-activists; further, activists are often overwhelmed because there is too much to do for the cause and not enough helping hands. Second, specifically in activist movements that campaign for the rights of minorities, the members often come from the minorities in question, so this is especially hard for them: they must shoulder both the disadvantages that society imposes to that minority, and the work that must be done in order to effect change.
• In the long term, movements are not effective if their members burn out. Further, if they are too demanding, then they may also scare away potential newcomers. For this reason as well, paying attention to the wellness of activists is a good long-term strategy for activist movements.

In fact, the book argues that it is necessary for activist movements to openly talk about wellness and care. Indeed, there is a risk that this emotional labor will be done anyway in an invisible fashion (and, disproportionately, by women), and that it will not be recognized. Further, the very act of taking care of oneself and of other community members can be a political act, namely, saying that your well-being matters: especially if you are from a minority that society does not care about.

Building a suitable form of wellness for activists

After defusing this criticism of wellness in an activist context, the book concludes by presenting its end goal: developing a notion of wellness which both addresses the problems pointed out in the first part, and can serve activist movements as explained in the second part. That is, wellness should be compatible with the end goal of activist movements (i.e., with the world that they want to make happen), and also be compatible with the path to get there (i.e., it should help political struggle, make activists stronger, and make activist movements more attractive).

The book offers some general guidelines for wellness practitioners (e.g., teachers):

• Wellness practitioners should strive at welcoming all people, including those who do not conform to the conventional idea of attractiveness. They should make this explicit and visible in their communication, and focus on exercises that all their participants can do, including those who have a disability. This means that, instead of leaving some behind, wellness should cultivate the art of caring for each other and including everyone (e.g., when going for a walk, going at the pace of the slowest person).
• Wellness should respect people's boundaries: practitioners should only touch participants with affirmative and easily revocable consent, and making it easy to opt-out or leave instead of being trapped by group dynamics. Of course, in the current majority culture, participants may find it difficult to feel what they genuinely want, or to express their needs: for this reason, wellness practitioners should explore communication approaches that aim at empowering participants. The general idea is that participants should not be asked to ”let go” and passively receive what is offered, but that they should be actively choosing what will happen.
• Wellness should be horizontal rather than vertical. It should not be organized as an all-knowing teacher having hierarchical power over students. It should be more horizontal, with teachers and participants working towards a common goal. It should also aim at empowering students, making them autonomous in the practice, instead of making them dependent on the teacher's knowledge.
• Wellness should be financially available to those who need it most, e.g., with fees that depend on revenue. Wellness professionals should also think about how they structure their activity, e.g., investigating cooperative models.
• Wellness practitioners should be careful about which ideology they promote. For instance, they should eschew the law of attraction which claims that wanting something is enough to get it (hence making individuals responsible for their misery by not “wanting” the right things); they should avoid gender essentialism; etc.
• Wellness practitioners should be aware of the risk of cultural appropriation. This is the situation where Western practitioners teach non-Western practices (e.g., yoga) and often misunderstand them, or bowdlerize them to be compatible with a Western capitalistic worldview.

The book thus identifies deeper positive functions of wellness that are needed in today's society and in activist movements:

• Wellness approaches can reconnect us to traditional roots that we have lost in the name of modernity and rationality. Indeed, the use of non-Western esotericism in wellness can be understood in a colonial way. Specifically, we have assimilated non-Western cultures, we have essentialized their exoticism, and we give them the role of helping us escape from our purely rationalist worldview, while still considering them as inferior. In the face of this, wellness practices can help us acknowledge the fact that the path to our own well-being is sometimes beyond the reach of rational thinking.
• Wellness approaches can reconnect us to our body, and remind us that we are embodied beings. They should make us rediscover pleasure, especially physical pleasure, e.g., via sex positivity. They should encourage us take our time and do less, e.g., via silent walks, meditation, etc. They should free us from the tyranny of having to constantly improve, and leave us space to just live. This goes against a productivist view which makes us forget the needs of our body and tame it to follow the rules of the system. They can also help us change our mindset by making our bodies stronger, training them to take more space rather than being subdued (especially for women and other minorities), and teaching them to defend us (e.g., with self-defense, in particular for women).
• Wellness approaches can make us aware of the true causes of our suffering, in particular systemic causes. They can also help us support one another with our vulnerabilities. They can also help us notice and accept how we are currently influenced by the system that we wish to change. In so doing, they can lead people towards collective political action.
• Last, wellness approaches can reconnect us to the importance of spirituality. The book argues for a personal form of spirituality, in particular one intended to connect us to nature. It perceives spirituality as a way to give importance to what matters to us, and thus to support political action.

This post is about the following hypothetical scenario: Alice is suspected to be a spy from a foreign country, and so to be fluent in the language spoken in that country, called language L. Alice claims to be an innocent citizen with no knowledge whatsoever of language L. Of course, this is just what a spy would say!

You are in charge of interrogating Alice, and to figure out whether she is fluent in L or not. How do you do it?

The easy case is if you are lucky and Alice inadvertently outs herself as an L speaker. For instance, she may use an L word accidentally¹ (e.g., cursing, or counting out loud², etc.³). Your task is also very easy if Alice has a thick L accent in the language that you share with her. But what is Alice is perfectly prepared, perfectly fluent in your language, and you cannot count on her to make a spontaneous mistake? Of course we want to avoid mistreating Alice (she might be innocent), and we can assume that she will reasonably cooperate with the interrogator (like an innocently accused citizen who wants to prove their innocence).

Here are some ideas of solutions to this problem: thanks to louis, Ted, olasd, and Tito for contributing some. There are also some solutions here which are already mentioned in a TVtropes entry with a wider scope⁴; but maybe there are other possibilities.

Brain imaging

With access to enough technology, there is a solution which is probably foolproof and requires no cooperation from Alice. Just put her in a neuroimaging machine, and have her listen to spoken recordings in language L. By comparing with a language that she knows, and with a language that she doesn't, it should be possible to detect whether her brain is making sense of what she hears or not. (I'm no expert in neurosciences though, so I cannot promise this would work.)

In what follows, I consider this to be cheating: I assume we don't have out-of-band access to Alice's brain, and that we must test her via normal sensory interfaces.

The Stroop test

The Stroop test is the following task: you are presented with a sequence of color names which are themselves written in a different color (e.g., it could contain the word "blue" but written in red), and you must read the sequence of the colors in which the words are written (not read the words themselves). This turns out to be difficult, with people often messing up and reading the words instead of naming the colors. But of course the Stroop effect only works if the person knows the language in which the words are written.

So if Alice's performance on this task is different between color words in language L and nonsense words, then you can find out that she is familiar with language L. The Stroop test may in fact have been used for this purpose historically, as discussed in this skeptics.SE question.

Alice may be able to evade detection by deliberately slowing down on nonsense words to try to match her performance on L words -- and in particular to make sure that she never ever translates an L word, which would be a dead giveaway. However, I would assume that maintaining exactly the same performance (and same error profile) between genuine mistakes (on nonsense words) and contrived mistakes (on words of L) would be very challenging for Alice. Especially if you collect precise timings on her performance that she herself does not have access to.

Timing attacks

A generalization of this idea is to measure Alice's performance on other tasks involving both L words and nonsense words, and seeing if her performance is different across both classes. Like, test Alice's ability to memorize short phrases, some of which are made of nonsense words and some of which are well-formed sentences in L. Of course if Alice is fluent in L the second task would be far easier for her.

I guess one can come up with similar tasks, e.g., if L uses an ideogram writing system, I would expect that it would be easier, say, to find differences between two figures, or find occurrences of a figure in another figure, or other such tasks, whenever the figures used in the task are genuine L ideograms as opposed to similar-looking but nonsense ideograms.

This is a bit like a timing attack in computer security. It also resembles a bit the implicit association test, though I find it difficult to adapt this specific test to a task for Alice that would also make sense if Alice has no knowledge of L whatsoever.

Following verbal instructions

Another idea is to have Alice play a game where she must follow instructions as quickly as possible. The catch is that some of the instructions are nonsense instructions, and some are instructions in language L. If Alice spontaneously reacts to one of the language L instructions, then she is unmasked. Playing this correctly if you understand language L is a bit like a Simons Says game, whereas of course it is not especially difficult if you do not know language L. Of course, Alice may again be able to avoid detection by deliberately slowing down.

A similar idea is mentioned in the TVtropes entry mentioned above, about a possibly apocryphal practice by the British of shouting "Achtung!" to identify German spies. Of course variations of this idea can work if you are not overtly interrogating Alice but watching her without her knowledge, which I would also consider as cheating.

A related idea: inserting words from L in conversation, using them as loanwords, and seeing in conversation whether Alice understands one of them (i.e., forgot to pretend she didn't). A related strategy (asking a question in French in the middle of an interview in Spanish) was used by Ladislas de Hoyos to identify Klaus Barbie while he posed as the non-French-speaking Klaus Altmann.

Watermarked language

Another idea hidden in plain sight: how about giving some classes to Alice where you teach her language L, and see how she performs? If she doesn't react like a real beginner would, in particular if she uses just one word that you hadn't already taught her, then she is unmasked.

There is a meaner variation on this idea, but which requires much more preparation. You could design and teach Alice a constructed language L', which is very similar to L except that it is "watermarked" in many small ways that are difficult to remember⁵, e.g., the orthography of some words is subtly different, some words have been exchanged, etc. If Alice is a spy, you would expect her to mess up at least some of the time with errors influenced by L. By contrast, if Alice is innocent, the errors she makes would not be correlated to L.

This is not a very practical solution, and also I don't know whether it would work in practice. Still, if it does, I find it interesting that knowing something (L) may be a handicap in properly learning something else (L').

An explicit solution

To finish with an entirely different idea, another way is watch for a physiological response: read some erotic literature in language L to Alice and see if she becomes excited. Ironically, this very low-tech solution comes from a science fiction short story: I'm in Marsport Without Hilda, by Isaac Asimov.

Other involuntary reactions to language could also work, e.g., laughter (e.g., with jokes), disgust (e.g., with gruesome descriptions), anger (e.g., with insults), etc.

More and more conferences in theoretical computer science (TCS) are moving to double-blind peer review. This is the implementation of peer review where the reviewers who evaluate submitted articles do not know the identity of paper authors, at least initially. The two major database theory conferences have adopted double-blind peer review (at least in an "experimental" way): PODS since 2023 and ICDT since 2024. Among general theoretical computer science conferences, we have ESA since 2019, STACS since 2021, ICALP since 2022, FOCS since 2022, SODA since 2022, STOC since 2024, and ITCS since 2024. An up-to-date list is maintained on double-blind.org. However, double-blind reviewing is not yet used in all conferences in all subareas of TCS. Further, it is also not commonly used in journals. In fact, I do not know of TCS journals that use double-blind peer review, except TODS which has used it since 2006 with a detailed rationale. See also the discussion of this issue at SoCG, which uses double-blind reviewing since 2023.

I think that this move to double-blind reviewing is a welcome change, and thought I'd try to summarize some of my thoughts about it.

First: should conferences and journals adopt double-blind reviewing? The implicit premise behind double-blind reviewing is that reviewers should evaluate articles based on their intrinsic merits, so that the identity and affiliations of the authors are not relevant for evaluation. When discussing double-blind reviewing, I think it is important to check first if all parties agree about this point. Indeed, this view is not universal: some researchers insist that it would be normal for conferences to evaluate submissions by newcomers with higher standards (see, for instance, this answer on academia.SE); or to the contrary conferences could be more welcoming towards outsiders¹. However, if you believe, like TODS, that "every submission should be judged on its own merits", then information about the authors and their affiliations is indeed irrelevant to reviewers (assuming they have no conflicts of interest — see below). The question becomes: does hiding this information make a difference, and is the difference sufficient to justify the change.

Second: how much of a difference does double-blinding make? This is not, in fact, straightforward to evaluate. There has been decades of debate on this topic, and policies have been adopted by scholarly associations in many fields, given that double-blind peer review has been around since the 1950s (according to Wikipedia). Further, many scientific studies have attempted to quantify its effectiveness. The TODS rationale contains some pointers to this literature, and after 2006 one can find many other articles arguing for or against double-blind peer review (see for instance this one and the many articles that it cites or that cite it). Of course, the effectiveness of double-blind reviewing may depend on the community, or on how specifically it is implemented. In computer science, one influential data point was the 2017 WSDM experiment, in which submissions were scored by double-blind reviewers and by single-blind reviewers (i.e., who knew the identity and affiliation of authors). The experiment found that "single-blind reviewing confers a significant advantage to papers with famous authors and authors from high-prestige institutions". Here I must confess that I have not myself read all these works: my point is just that the issue is not simple, and that you cannot dismiss double-blind reviewing simply because you are unfamiliar with it or because you are personally convinced that it does not work.

Third: how should double-blind reviewing be implemented? Here, there is an idea I'd really like to get across: making reviewing "double-blind" does not necessarily mean that it should be impossible for reviewers to deanonymize authors. Indeed, essentially all² conferences in the list above are using one specific implementation, called lightweight double-blind reviewing. This means that, while papers should not contain identifying details, and while reviewers are advised not to try to identify authors, it is OK if reviewers happen to find out about who the authors are. In particular, authors are encouraged to post their work, e.g., on preprint servers, even if this means a reviewer may find it (deliberately, or accidentally, e.g., by stumbling upon the work before reviewing). Lightweight double-blind reviewing still offers two important benefits:

• Reviewers are not immediately biased by reading the author names and affiliations on the first page of the paper. In my personal experience as a reviewer, seeing familiar names or affiliations on a paper will immediately affect my impression of the paper and my expectation about the outcome ("probably accept unless the contents are a bad surprise" vs "probably reject unless the contents are a good surprise"). I would like not to be influenced by this, but I doubt I can avoid it, so I prefer not to see the information³.
• Authors who worry about discrimination and do not trust the reviewers can choose to take steps to be completely anonymous, e.g., they can choose not to post preprints of their work.

By contrast, I do not like double-blind policies that try to guarantee complete anonymity of submissions at all costs, e.g., by prohibiting or discouraging the posting of papers as a preprints, informal talks, etc. I believe this is harmful, because it means that papers are only available when they are finally published⁴ — by contrast, preprints are immediately accessible to everyone. Unfortunately, there are conferences (especially practical conferences) that follow this route, e.g., SIGMOD 2024⁵. Disallowing preprints can have a chilling effect on an entire field: as papers often end up rejected from a conference and resubmitted at another, authors may eschew the posting of preprints because of the risk that some ulterior, unspecified conference may disqualify their work on these grounds.

Fourth: What are the real problems with double-blind reviewing? Many of the criticism I have heard does not make sense to me:

• "Anonymizing papers is tedious work." This I really don't understand: removing author names is trivial, writing self-citations in the third person (e.g., "The previous work of Myself et al." vs "Our previous work") feels weird but takes a few seconds... Altogether the impact seems to be minimal⁶.
• "It it complicated to host/submit supplementary data, e.g., source code or experiments, in an anonymous fashion." But there are tools like Anonymous GitHub, and guides on what you can do.
• "Double-blind reviewing is not perfect, and reviewers can guess who the authors are." Well, first, guessing is not the same as being sure; second, it's still useful if you can remove bias in some cases. An improvement can be valuable even if it is not a perfect solution.
• "It's unnecessary, everyone is honest and exempt of bias." Even assuming that reviewers try to be fair, this misses the point that bias is often unconscious.

To me, the real (minor) shortcomings of double-blind reviewing are:

• Some journals (and conferences?) are "epijournals", where papers are first submitted to a preprint server, and then reviewed by the journal (and endorsed if they pass peer review). I think this is a great practice, that neatly separates the hosting of papers (which is done for free by preprint servers) from the evaluation process. Unfortunately the interaction with double-blind peer review is not perfect: you cannot send the preprint to reviewers, because of course it includes author information. The fix is obvious, just a bit inelegant: simply ask authors for an extra blinded version of the paper when they submit it for evaluation by the journal.
• The management of conflicts of interest (COIs) is more complicated. Many conferences and journals have policies to ensure that papers are not reviewed by colleagues, collaborators, supervisors, or personal friends of the authors — or, of course, the authors themselves! When reviewers know who the authors are, they can easily detect COIs and recuse themselves from reviewing the paper. With double-blind reviewing, this is more complicated. Typical solutions include asking authors upon submission to disclose with which members of the program committee (PC) they are in COI, and/or asking PC members to disclose with which authors they are in COI (during the bidding phase). But these are not easy to adapt to journals, where papers are typically sent to editors not affiliated to the journal. Or, for conferences, it does not address subreviewing, where PC members delegate a paper to an outside expert: this expert could end up being in COI with the authors, or be one of the authors, which is especially embarrassing. This is typically handled either by unblinding papers from PC members if they decide to subreview a paper, or by making subreview invitations pass through an unblinded party (e.g., the PC, or a specific organizer not involved in reviewing) who can check for the absence of COI⁷.

I hope this post can help clarify some points about double-blind reviewing, and maybe encourage more conferences and journals to consider it!

This post is a reprint of the "Viewpoints" column by Theoretical Computer Scientists for Future (TCS4F), published in the bulletin of the EATCS. You can read it in PDF format in number 139 of the Bulletin.

The issue of climate change has been on our collective mind for decades. Each passing year improves our scientific understanding of the problem, and narrows down our uncertainty about the need to drastically reduce worldwide greenhouse gas emissions. As the window of opportunity is closing, and concrete action is slow to materialize, more and more groups from seemingly unrelated areas find themselves advocating for change.

TCS4F is one such initiative: it is lead by computer scientists, and aims at making research in theoretical computer science environmentally sustainable. It started in 2020 with a manifesto that can be signed by researchers, conferences, and research groups. The pledge taken by signers is to follow a sustainable emissions trajectory: reduce emissions by at least 50% before 2030 relative to pre-2020 levels. The TCS4F manifesto was signed by 199 individual researchers (and counting!), 3 research groups, the 2022 edition of the ICALP conference, and the 3 conferences CSL, STACS, and Highlights of Logic, Games, and Automata.

The contribution of theoretical computer science research to the climate crisis is two-fold. On the one hand, we may be able to improve the situation through our research. For instance, we can improve the efficiency of algorithms and hope to reduce the footprint of the ICT sector — though our efforts may well have the opposite effect because of the Jevons paradox! On the other hand, we should also think about the present impact of our research activities on the environment, and try to adapt our practices to be more sustainable.

It may be unclear at first how theoretical research harms the environment — is it about the consumption of draft paper? Whiteboard markers? In fact, while our activities can emit greenhouse gases in many ways, the main factor in our climate impact is probably long-haul plane trips. Indeed, our research field is structured around international conferences. Their stated aim is to give the community a place to meet, discuss, and exchange new ideas. Prestigious conferences are also the most important means of recognition in our community: they are a must-have on one's CV when applying for research positions. For PhDs and researchers on short-term positions, publishing there is not a choice but has become a vital professional necessity. And, until recently, publishing at international conferences naturally meant that you had to fly across the world and be there.

It is in this context that we launched TCS4F in early 2020. This coincidentally followed Vardi's “Publish and Perish” CACM column, which advocated for optional attendance to conferences in the name of environmental sustainability. As we all know, shortly afterwards, the COVID-19 epidemic moved all conferences online almost overnight. This forced experiment gave us a taste of what could be the closest online replacement for traditional conferences — if organized on short notice and by necessity rather than choice. The situation left us yearning for the golden days of in-person conferences and lively bar discussions in exotic locations, and the question of flight-induced climate change was not very pressing while we were stuck at home during lockdowns.

Once the COVID situation improved, many conferences adopted some kind of hybrid format, pragmatically acknowledging the fact that travel was not possible for everyone. These experiments revealed that it is comparatively easy to accommodate remote speakers, and to stream talks to a remote audience, which some conferences already had experience with. However, integrating the in-person and remote worlds proved challenging, especially for coffee breaks and social events. Based on this, some conferences are now back to firm requirements for in-person attendance, and are making explicit what used to be an implicit rule: “all talks are in-person” at ICALP'23, online talks will be for “travel restrictions or other exceptional situations” at ICDT'24... The intent may be to encourage participants to travel so everyone can enjoy a better conference... or to ensure that universities will continue to reimburse trips. Also, a fully in-person conference is of course simpler to organize, and closer to what we are used to.

These rules arguably reveal an inconvenient truth: many conferences are now attracting participants whose main goal is to have their paper published (at a prestigious venue, and on a predictable timeline), and not necessarily to attend the event. Of course, the general will to travel and meet is still very much alive — as can be seen at events without formal proceedings, such as the Highlights workshop series. But coupling formal publications with an in-person gathering no longer makes sense for everyone.

We argue at TCS4F that decoupling the two is necessary, because plane travel is unsustainable at the scale at which we practice it. Flying across the world to a conference can amount to several tons of CO2-equivalent emissions, exceeding sustainable targets for individual yearly footprints in 2030 (source), and there are no plausible technological pathways for low-carbon intercontinental travel by then. Thus, our position at TCS4F is that, if everyone is to do their part to mitigate climate change, we must fly less — and attend less international conferences in person.

However, I believe that mandatory travel is also a question of diversity and inclusion. In-person conferences are an exclusive club for frequent travelers, and exclude people with insufficient funding to travel, people from countries who cannot easily obtain visas, people with disabilities, and people with caretaking obligations (which disproportionately affect women). For instance, the relative proportion of women participants at the 2020 International Conference on Learning Representations (online) was 20%, versus 15% at ICLR'2019 (in-person) — a 33% increase (source). Our focus on in-person conferences thus overlooks a silent majority of people for which online attendance is the only feasible way to participate. Further, if prestigious conferences are in-person only, then recognition in our community is reserved to the privileged few who can meet this obligation.

Of course, my point is not that in-person conferences should be eliminated altogether. As we all know from the COVID era, online events are not perfect, and in-person socializing has no known replacement. Traveling to conferences is still important, and can be done responsibly — going there by train if possible, picking geographically closer locations, or simply going there less often. Online and hybrid events can also play a role, as do other forms of online research: online videos (for instance on ScienceCast), online seminars (for instance on researchseminars.org), the Theoretical Computer Science Stack Exchange, etc. These new formats are especially promising when they do not try to mimic what already exists, but instead leverage features specific to the Internet: asynchronicity, low friction, low cost, machine interpretability, long-term archival... Overall, it is very challenging to balance the scientific value of international in-person meetings with their environmental impact. But every member of our community should have a say in this choice, and it should be guided by careful deliberation — not simply by reverting to the default 20th-century-style conference culture.

It is not yet clear how the conference landscape will evolve after COVID: which conferences will settle on a new format in the long run, and which ones will revert to the pre-COVID rule of mandatory participation barring extenuating circumstances. We have tried to survey this at TCS4F. For conferences with optional in-person attendance, it is not clear how much organizers will encourage or discourage participants to travel, and how researchers will respond. These questions should probably be debated in our community, so the system can achieve the best compromise between scientific value, inclusivity, and environmental sustainability. But, specifically for prestigious conferences with formal proceedings, our short-term hope is that future call for papers will allow publication without in-person attendance.

We are interested at TCS4F to hear about the views of the community on this important issue. Should conference publication be conditioned to onsite participation? How should our conference culture change to be sustainable and inclusive? You can reach us at: contact<REMOVETHIS>@tcs4f.org.

Suggestions for further reading:

• Laurent Feuilloley. About mandatory attendance
• Moshe Vardi. The paradox of choice in computing-research conferences, Communications of the ACM, 2021
• TCS4F blog
• Flying less in academia: A resource guide
• ALLEA. Towards climate sustainability of the academic system in Europe and beyond

This post features minor corrections relative to the edition published in the bulletin of the EATCS. This post was written with help from the TCS4F team: Thomas Colcombet, Thomas Schwentick, and Tijn de Vos. Thanks to Louis Jachiet for proofreading and suggestions.

Summary: it is possible to play Tetris and guarantee that you will score at least one line, no matter which pieces are given to you, i.e., even assuming they are chosen adversarially.

There are many things that can be found online about the game Tetris: high scores (see for instance this great documentary), implementation details (did you know there was a specification for Tetris games?), intellectual property surprises (this post is not related to or endorsed by The Tetris Company)... But for many years I have not been able to find on the Internet the answer to this question: can you be sure to clear a line at Tetris?

Of course, in practice, most Tetris players seem to be able to clear lines at least some of the time^{[citation needed]}. But maybe they are just being lucky! Maybe an evil computer or extreme bad luck could prevent you from ever clearing a line, no matter how you played. For instance, consider the game Hatetris, which is programmed to give you unpleasant pieces. Clearing a line in Hatetris is much more challenging, though good players can do it. Could a worse version of Hatetris, with a more clever opponent, give you pieces that will always make you lose without scoring a single line?

Mathematically, we can formulate this a sequential zero-sum game with perfect information. We have a standard Tetris board of 10 columns by 20 rows. The computer and human play alternatively: the computer gives a piece to the human (one of the seven tetrominoes), and the human responds by positioning it somewhere according to the usual Tetris rules. If the screen is filled up then the human has lost, and if the human completes a line then the human has won. Mathematically, there are then only two possibilities:

• either the computer has a strategy to give pieces (depending on how the human has played so far) that guarantees that the human will lose without completing a single line, no matter what they do;
• or the human has a strategy to place pieces that guarantee that they will score a line before losing, no matter which pieces they receive from the computer.

My question was to figure out which one of the two is true. The same question was asked by qntm in 2011, but leaving the question open for the standard Tetris board size. After some coding and much computation, it turns out that the human wins: it is possible to guarantee a non-zero score when playing Tetris. If you want to see an example strategy, you can test it with the Tetris board below. You play the computer, i.e., you select which piece to give, and the chosen piece will be placed according to a winning strategy. Try preventing your opponent from scoring a line! (the point of this post is that this is impossible)

An alternative challenge: try preventing the computer from scoring a line in the 5 bottom rows! This is possible for precisely 759 piece sequences out of the 427513 winning sequences. Can you find them?

In the rest of the post, I explain some details about the exact problem statement, and give more details about the strategy and how to check it. Then, I explain how I computed the strategy, possible improvements to the computation, and then present related work and remaining open problems.

Technical details

Here are some detail about the problem and strategy:

• I restrict the human player to only rotate and then drop pieces vertically, i.e., playing a piece means choosing a rotation and a column where to drop it. (You cannot slide a block under another block, no t-spins, etc.) Of course, if the human can win under this restriction, they can win even when we allow more moves.
• I do not require the computer to show the next piece, i.e., commit to what the next piece will be before the human has dropped their current piece. Of course, if the human can win in such conditions, they can also win with a next piece display.
• Under such circumstances, the human can play so as to guarantee that they can complete one of the lowest 6 rows of the board. The exploration also shows that the computer can prevent the human from scoring one of the bottom 5 lines, subject to the above restrictions, and assuming that the human does not play "above" the fifth line, i.e., dropped pieces cannot touch a full column which has height at least 5. (See below to know more about this limitation.¹)
• In the worst case, the human needs to position 13 pieces to complete its first line among the bottom 6 rows. This is optimal under the above restrictions: when only considering completion of the first 6 rows, dropping pieces without sliding them, and prohibiting dropped pieces from touching a full column. I think it is very likely that the human can win faster if we allow them to slide pieces; and possible that the human can win faster by going above the 6th row.
• Where several moves achieve the same distance to a win, they are chosen arbitrarily, following the order in which the possible choices of where to put a piece are considered (first by rotation, then by column). This is why the strategy above is not symmetric, e.g., between l-shaped and j-shaped tetrominoes.

Checking the strategy

I provide the strategy as a file describing how the human can play and guarantee a win. Each line of this file corresponds to a possible "game state", numbered from $0$ to 5249 (the file has 5250 lines²). Each line consists of 21 integers, numbered from $0$ to $20$ as $3p+q$ with $0\le p<7$ and $0\le q<3$. The triple $3p,3p+1,3p+2$ explains what to do when receiving piece $p$: the number $3p$ indicates the rotation $0\le r<4$ to use, the number $3p+1$ indicates the column $-2\le c<10$ where to drop the piece, and the number $3p+2$ indicates the game state from which we should follow the rest of the winning strategy (the special value "-1" indicates that the human has won). The pieces, rotations, and column offsets are according to the following description of the pieces:

== piece 0 rotation 0 ==
#...
#...
#...
#...
== piece 0 rotation 1 ==
....
....
....
####
== piece 0 rotation 2 ==
...#
...#
...#
...#
== piece 0 rotation 3 ==
####
....
....
....
== piece 1 rotation 0 ==
.#..
##..
#...
....
== piece 1 rotation 1 ==
....
....
##..
.##.
== piece 1 rotation 2 ==
....
...#
..##
..#.
== piece 1 rotation 3 ==
.##.
..##
....
....
== piece 2 rotation 0 ==
.##.
##..
....
....
== piece 2 rotation 1 ==
....
#...
##..
.#..
== piece 2 rotation 2 ==
....
....
..##
.##.
== piece 2 rotation 3 ==
..#.
..##
...#
....
== piece 3 rotation 0 ==
###.
.#..
....
....
== piece 3 rotation 1 ==
....
#...
##..
#...
== piece 3 rotation 2 ==
....
....
..#.
.###
== piece 3 rotation 3 ==
...#
..##
...#
....
== piece 4 rotation 0 ==
###.
#...
....
....
== piece 4 rotation 1 ==
....
#...
#...
##..
== piece 4 rotation 2 ==
....
....
...#
.###
== piece 4 rotation 3 ==
..##
...#
...#
....
== piece 5 rotation 0 ==
###.
..#.
....
....
== piece 5 rotation 1 ==
....
##..
#...
#...
== piece 5 rotation 2 ==
....
....
.#..
.###
== piece 5 rotation 3 ==
...#
...#
..##
....
== piece 6 rotation 0 ==
##..
##..
....
....
== piece 6 rotation 1 ==
....
....
##..
##..
== piece 6 rotation 2 ==
....
....
..##
..##
== piece 6 rotation 3 ==
..##
..##
....
....

The strategy thus describes a DAG (with labeled edges). In particular, we can get to the same game state by different paths, and indeed the same game state can correspond to different states of the board but from which we can use the same winning strategy.

The strategy in this file is the one used in the Javascript game above. I have written a program (with helper file) that checks the strategy by systematically going over all possible moves by the computer, playing according to the strategy, and checking that the human indeed wins when the strategy says they do. Hence, the mathematical "proof" that the human has a winning strategy would consist of this file and of the verification program.

Finding the strategy

I could stop here and say that the strategy and the verification program are an answer to the question I had posed :) but let me say a few words about how I found it. This is performed using the minimax algorithm, i.e., systematically exploring the game tree. Formally, we define inductively who wins on a given board state, among the human (H) and computer (C):

• Base 1: H wins on board states where one of the bottom 6 rows is filled
• Base 2: C wins on board states where we cannot place a piece without touching a full column, i.e., one where there are already blocks above the 6th row.
• Induction 1: In a given board state $b$ where H is given some piece $p$, if H can drop it and get to a board state which is winning for H, then $(b,p)$ is winning for H; if all moves by H get to a board state which is losing for H, then $(b,p)$ is losing for H.
• Induction 2: In a given board state $b$, if C can give H a piece $p$ such that $(b,p)$ is losing for H, then $b$ is losing for H; if any pieces given by C are such that $(b,p)$ is winning for H then $b$ is winning for H.

This is just a simple recursive exploration. The naive way to implement this would be to compute who wins on every board state, remembering in the board state if each cell of the bottom 6 rows is filled or not: this would be $2^{6\times 10}$ possible board states, where 10 is the number of columns. This amounts to 1153 peta states, which is too much.

Since we are only allowing pieces to be dropped (without sliding them), we can do better: in each column, we only need to remember the height of the highest completed block³. The status of the cells below the topmost filled cell of a column have no influence on which moves are possible. The only extra thing to remember is the set of rows having a "hole", i.e., an empty cell above which there is a filled cell. As we cannot slide pieces, these holes can never be filled, so we must remember that the rows that have a hole cannot be completed. We say that these lines are "sacrificed".

For an example, consider the state of the board after dropping a vertical I-tetromino in the second column, pictured below. We represent this as the sequence of heights written at the bottom.

Imagine now that we drop a vertically oriented z-block, getting to the configuration below:

Columns 2 and 3 are now full, i.e., the blocks stack up to the 6th row. Hence, we store their height as 6, and forget what happens above the 6th row (i.e., we forget the block in column 3 with a black cross) -- because of this, we will no longer allow blocks to touch one of these full columns. Further, the new block has now created holes -- the gray cells with red crosses. The sequence of heights does not account for these holes; to remember that they are there, we store that the bottom 5 rows have been "sacrificed", materialized by the red crosses to the left of the board. This is sufficient, because our way to drop pieces will never allow H to slide a piece and fill these holes. Now, the 6th row is the only row where H can ever hope to complete a line. (In fact, the configuration is now clearly losing for H: if C no longer gives H any I-tetromino, H can never fill the cell at column 1 and row 6, so H can never complete the only non-sacrificed line.)

To summarize, the board state consists of the height of each column (between 0 and 6 inclusive) and the set of sacrificed lines, for a total of $7^{10}\times 2^6$ board states. This is much better: there are now 18 billion states. There are other small savings, e.g., board states where all lines are sacrificed are immediately losing. In practice, a complete exploration considers around 2 billion states, i.e., only around 11% of the possible board states are reached.

For efficient implementation, a board state is represented as a 64-bit integer consisting of a 16-bit mask describing the sacrificed lines (only the 6 lowest bits are useful), and 10 4-bit integers describing the heights (only the 3 lowest bits are useful).

As the same board state can be reached by many different paths in the game tree, we use memoization: when we are done exploring the tree from some board state, we remember whether this state is losing or winning, and if it is winning we remember in how many moves. This is done with a hash table (a C++ unordered set). This means that the program requires a large quantity of RAM to run (around 20-30 GB): for this, I thank the INFRES department at Télécom Paris for giving us access to suitable computing resources.

This explains how the winning strategy is found. It is fast (one hour) to check that a winning strategy exists by terminating the search for a board state and piece as soon as a winning move is found. It is longer (18 hours) to find a strategy that wins as fast as possible, because this requires us to explore the full tree. To speed things up, we use a form of alpha-beta pruning: when trying to put a piece, when we have found a winning option, we consider other options but only exploring them up to a depth that would give a strictly shorter strategy than the currently known best option. This lowers the running time to 10 hours and lowers the number of explored states from 2 billion states to around 750 million states, of which 9 million are winning. The resulting program is here. (Sorry, it is not very clean...)

Once the program has produced a strategy, we need to "compress" it. To do so, we first "trim" it by removing unreachable board states: we go from 9 million states to just 21 thousand states: this is done here. Then we minimize it by merging states from which the strategy is the same: this is done here. As the DAG is acyclic, we can perform this in linear time simply by processing the DAG bottom-up and hashing configurations. We get to the 5250 states of the strategy file.

Other possible optimizations

The following optimizations would have been possible, but I did not implement them:

• There would be an easy saving of a factor 2 by breaking the left-right symmetry, which would also probably make the winning strategy more concise (but it is less convenient to reconstruct it).
• When checking the cache of possible board states, we could check "shifted" configurations where we remove any one of the lowest $d$ lines, provided each column has a height of at least $d$. We could also check configurations with a subset, or superset, of sacrificed rows: if a configuration with identical column heights and with a subset of sacrificed rows is known to be losing then the current configuration also is, and if a configuration with identical column heights and a superset of sacrificed rows is known to be winning then the current configuration also is.
• Rather than exploring all possibilities to the maximal depth before giving up, a better solution is to use iterative deepening. I did this in an earlier version of the code that followed a different approach. Likewise, ordering the possible moves with a piece using some heuristic (e.g., putting the piece as low as possible, or sacrificing as little new lines as possible) would possibly make the search faster (thanks to alpha-beta pruning).

Related work

The most related work is the 2011 study by qntm, which set out to solve the same question. His work did not conclude that the human had a winning strategy for the standard board size of 10 columns, though the cases with less columns were solved. The analysis showed that a height of 6 was not sufficient to have a winning strategy. For qntm's analysis, playing up to height 6 means not allowing a piece to protrude above line 6, whereas for me it means not allowing a piece to land on a column that goes to height 6 and not considering filled lines strictly above line 6. This would imply that qntm should find a solution for height 9 or less. A difference is also that qntm considers more moves than simply rotating and then dropping pieces as I do; this may be part of the reason why I was able to conclude the analysis and qntm was not.

For the related question of whether the human can win at Tetris in the usual sense, i.e., play indefinitely, it is known that this is not the case. The computer can force the player to lose, even with a strategy which is oblivious to how the human plays and alternates S-shaped and Z-shaped pieces: see Burgiel's 1997 paper "How to Lose at Tetris" or this page. This does not contradict the result presented here, which says we can clear a line (but eventually lose). Optimizing the number of lines cleared (in the worse case or in expectation) vs guaranteeing that you make one line are different goals, that may be at odds with each other.

On the other hand, with stronger assumptions on how the computer can propose pieces, it is possible to play forever: see this page.

There are other results of this kind in Brzustowski's 1992 master thesis Can you win at Tetris?.

There is a 2002 study by Demaine et al. of the computational complexity of Tetris play, which does not seem related to the results here. For more related work on Tetris, there is a great literature review in this FUN'2022 paper by Dallant and Iacono.

There are many Tetris implementations that try to give the worst possible pieces to the player: at least Hatetris, bastet, and LTris in "expert mode".

A relevant question seems to be the study of "adversarial Tetris", introduced at a reinforcement learning competition in 2009 (see here), and followup, e.g., Beatris. However, I wasn't able to find works in this area which considered the question of whether the human could guarantee that they score at least one line assuming perfect play.

Open problems

I do not know what is the minimal number of rows and/or maximal necessary height to score a line if the human can slide or rotate pieces as they fall; the latter would depend on the exact rules implemented for rotation which is somewhat unpleasant (similar subtleties are considered in Demaine et al.'s paper).

I do not know what happens for a different number of columns. Odd number of columns clearly allow the computer to win by giving only square tetrominoes, and two columns clearly allow the human to win (every piece can score a line by itself in every configuration except I-tetrominoes where this can be achieved in two moves). Of course one can generalize the problem to different polyomino sets, etc.

I do not know if the human can guarantee scoring multiple lines at once: if the computer only gives O-shaped blocks then the human cannot hope to score three or four lines at once⁴; I don't know if guaranteeing that you can score two lines at once is possible.

I do not know if the result implies that the human can score any arbitrary number of lines provided that the board is sufficiently high. The strategy I presented does not ensure this, because it only works from the empty board; it could be the case (although unlikely) that the board states reached after scoring one line in this strategy no longer themselves have a winning strategy (no matter the height). More generally, one can ask what is the behavior of the function that maps the number of rows of the board to the number of lines that can be guaranteed by the human.