<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Jo Lindsay Walton on Medium]]></title>
        <description><![CDATA[Stories by Jo Lindsay Walton on Medium]]></description>
        <link>https://medium.com/@jolindsaywalton?source=rss-25ffba2f748e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Axvu-3lTT18a36FV5fXJaA.jpeg</url>
            <title>Stories by Jo Lindsay Walton on Medium</title>
            <link>https://medium.com/@jolindsaywalton?source=rss-25ffba2f748e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 14:44:07 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jolindsaywalton/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How AI is improving the productivity of knowledge workers]]></title>
            <link>https://medium.com/@jolindsaywalton/how-ai-is-improving-the-productivity-of-knowledge-workers-ddfd8f56f8c7?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ddfd8f56f8c7</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[knowledge-workers]]></category>
            <category><![CDATA[consultancy]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Fri, 06 Nov 2026 13:49:07 GMT</pubDate>
            <atom:updated>2026-11-09T21:04:58.337Z</atom:updated>
            <content:encoded><![CDATA[<p>Just another example of what you get when you scratch beneath the surface of thought leadership. In June 2023, McKinsey published <a href="https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier"><em>The economic potential of generative AI</em></a>, a report estimating that generative AI could add trillions of dollars annually to the global economy.</p><p>A key example concerned information retrieval. Knowledge workers, McKinsey suggested, spend “about a fifth of their time” searching for and gathering information. Generative AI could take over much of this work, searching “vast corporate libraries” and allowing employees to concentrate on more productive tasks.</p><p>This sounded intriguing. Where did that one-fifth figure come from?</p><p>McKinsey cited its own 2012 report, <a href="https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy"><em>The social economy: Unlocking value and productivity through social technologies</em></a>. Already, there was a subtle change in emphasis. The 2012 estimate concerned searching for and gathering <em>all kinds</em> of information, not simply finding documents in corporate knowledge systems.</p><p>There was also the age of the evidence. The 2012 report relied on an IDC publication from 2009, drawing on an even older tradition of workplace surveys. Related IDC research had been conducted as far back as 2004. The year TheFacebook launched. Not exactly fresh data to evidence the generative AI revolution.</p><p>But it gets even worse.</p><p>The original IDC estimates broke down the working week into activities including reading emails, searching for information, analysing information, creating content, managing projects and communicating with colleagues.</p><p>The average number of hours reported for each activity added up to <strong>65.8 hours a week,</strong> considerably more than the reported average working week of 46.5 hours. Obviously, at least some of the categories overlapped. So it’s not an unusual result from the survey, you just have to deal with it appropriately. (Someone analysing information might simultaneously be searching for it, for example).</p><p>This was too many hours, thought the 2012 authors, since the average working week was only 46.5 hours. They then did something remarkable.</p><p>They took the three categories they were interested in (emails, information searching and internal communication), added them together, and subtracted the result from 46.5 hours.</p><ul><li>Reading and answering emails: <strong>13 hours</strong></li><li>Searching and gathering information: <strong>8.8 hours</strong></li><li>Internal communication and collaboration: <strong>6.4 hours</strong></li></ul><p>The remainder, 18.3 hours, became a catch-all category called “role-specific tasks”.</p><p>The seven categories now lumped together as “role-specific tasks” had originally accounted for 37.6 hours of reported activities. So McKinsey had simply discarded 19.3 hours, reducing these activities by more than half. Meanwhile, the three activities that the report was arguing took up too much time were left entirely untouched.</p><p>There was no evidence that the overlap occurred exclusively in the other seven categories. McKinsey simply assumed this, apparently because those weren’t the activities it was interested in.</p><p>The resulting figures later fed directly into McKinsey’s estimates of how much time its proposed technologies could save. The three protected categories were also those assigned the largest potential efficiency improvements.</p><p>In 2012, the technology in question was <strong>enterprise social networking</strong>, which McKinsey suggested could improve productivity by 20–25%.</p><p>Eleven years later, the same figure for time spent searching and gathering information reappeared in the case for <strong>generative AI</strong>. The technological revolution had changed, but the stupid statistic was apparently still good enough.</p><p>These stats then feed estimates, given in trillions, of how much GenAI is going to contribute to the economy.</p><p>All this is completely unremarkable — the thought leadership of companies like McKinsey are absolutely littered with obvious, objective errors, a weaponised incompetence that allows them to make whatever arguments are convenient to them. These days, a lot of it’s about AI.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ddfd8f56f8c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Game Theoretic Approach to Collective Fractionalization]]></title>
            <link>https://medium.com/@jolindsaywalton/a-game-theoretic-approach-to-collective-fractionalization-e75856f9f4aa?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/e75856f9f4aa</guid>
            <category><![CDATA[mechanism-design]]></category>
            <category><![CDATA[game-theory]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Sun, 11 Oct 2026 12:05:15 GMT</pubDate>
            <atom:updated>2026-10-11T12:05:15.379Z</atom:updated>
            <content:encoded><![CDATA[<p>A group of workers (say, a department), threatened with a Compulsory Redundancy (CR) process, may instead offer to management to undertake voluntary fractionalization to meet the required savings.</p><p>The department is therefore faced with a coordination problem. Can they assemble a package of FTE reductions that are acceptable to everyone in the department, and meet management’s target? Individuals will differ in their financial circumstances, working arrangements, and willingness or ability to absorb a reduction in hours and pay.</p><p>There is also a procedural problem. While the workers are in an unchosen situation (the need for fractionalization is imposed from outside), they want to organise their response without reproducing that coercion among themselves, or placing undue pressure on individual colleagues to make sacrifices. They want to be caring, sensitive, and compassionate, and to minimise any sense of judgment or shame. This requires active attention, since colleagues are likely to feel guilty about what they can contribute even when nobody is judging them.</p><p>Let’s assume that the savings target is fixed. There is a <em>lot </em>of messy, real-world complexity that will be left out in this post — you see “game theoretic” in the title, right. For example, whether partially meeting management’s savings target, alongside the threat of industrial action, might persuade management to revisit its assumptions and find savings elsewhere (e.g. fractionalizing or taking pay cuts themselves).</p><p>Game theory and the literature on institution design and mechanism design offer a range of procedures for allocating costs and coordinating voluntary contributions. These might involve successive rounds of simultaneous confidential bids, with a rule for reallocating any excess pledges (e.g. scaling down all pledges proportionately, or reducing the largest pledges first). In a university, I would imagine that an economics or mathematics department, or perhaps a development studies department, would be instinctively familiar with such processes and might be able to mobilise collectively more quickly and effectively than other departments.</p><p><a href="https://www.aeaweb.org/articles?id=10.1257/aer.20160425">Li’s work on obvious strategyproofness </a>(OSP) is interesting in this respect. What is obvious to one community is not necessarily obvious to another. More subtly, different kinds of mechanisms may attract, select for, and cultivate participants with different intuitions about what is obvious. People accustomed to particular forms of collective decision-making may find some strategic choices immediately transparent and others surprisingly difficult to grasp. There may therefore be a reciprocal relationship between the mechanisms a community tends to use and the forms of strategic reasoning its members find obvious. It feels like there may be a research question here? — but I am focusing on something else.</p><p>Other departments might, of course, bring different strengths in collective action. To stereotype a little more, economics departments might have less experience of grassroots organising, workplace solidarity and collective mobilisation than English departments.</p><p>A while ago I wrote my devastating critique of the toolkit-industrial complex <a href="https://docs.google.com/document/d/1Xe0Le1tShDeEDnO8fEapXiUkMPAULUHjn3uxDyfvdXQ/edit?usp=sharing">in the form of science fiction</a> (published in <em>Phase Change: Imagining Energy Futures</em>); obviously I am also part of the problem, because what I am playing with here is the idea of yet another toolkit.</p><p>Assume three different Collective Action Processes (CAPs) for eliciting pledges in an attempt to solve the coordination problem. I’m not sure whether we need to get into their details; perhaps not. But let’s assume three possible CAPs …</p><ul><li><strong>CAP #1 — O — One-shot.</strong> Everyone submits one confidential pledge simultaneously. The collective allocation is calculated after this single round.</li><li><strong>CAP #2 — T — Two-shot.</strong> Everyone submits a confidential pledge, receives anonymised information about the aggregate pledges, and then has one opportunity to revise their pledge before the allocation is calculated.</li><li><strong>CAP #3 — R — Repeated.</strong> Everyone submits confidential pledges in successive rounds, receiving anonymised aggregate feedback between rounds. The process ends when every individual&#39;s pledge has remained unchanged for three consecutive rounds, subject to a maximum number of rounds and a fallback rule if agreement is not reached.</li></ul><p>Assume that players can have different intuitions about the legitimacy of each CAP from the standpoint of sensitivity, care, respect etc.— which, in the barbarous terminology of this discipline, we will formalize as <em>cost </em>incurred. (Toolkit note: Where this is going eventually, but slowly, is the possibility of <em>different workers inputting according to their preferred CAP</em>. But set that aside for now).</p><p>A preliminary model might include, say, twenty players and seven ideal templates (similar to types in game theory). Each template is specified by four variables: 1) private cost of each 0.1 FTE reduction (keep it linear to keep it simple) 2) order of most preferred to least preferred CAP, <strong>3) the spread between the most preferred and the least preferred (do you care a lot or a little about how the decision is reached) </strong>and 4) private cost of failing to meet the threshold and triggering the CR fallback outcome (which deliberately conflates factors like “How likely am I personally to be made redundant?” and “How much do I care about colleagues who I believe are likely to be made redundant?”).</p><p>I’ve bolded 3) because in my actual experimental run, that’s the thing I altered. (So OK if you’re a stickler, there are fourteen ideal templates here, not seven).</p><p>For instance: Financially Stretched Carer (high private cost associated with FTE loss), Time-Seeking Colleague (low private cost associated with FTE loss), Anxious Comrade (high cost associated with triggering CR, strong preference for CAP #1), Broken Yet Confident Desperado (low cost associated with triggering CR, strong preference for CAP #1), Conditional Cooperator (strong preference for CAP #2), Deliberation Enthusiast (moderate costs associated with FTE loss and triggering CR, strong preference for CAP #3), Exhausted Solidarist (high private cost associated with FTE loss, high cost associated with triggering CR). These aren’t spread representatively over the possibility space or anything; they are just speculation and storytelling.</p><p>As a sample run, assume a department of 4 Anxious Comrades, 4 Financially Stretched Carers, 4 Exhausted Solidarists, 3 Broken Yet Confident Desperados, 3 Deliberative Colleagues, 1 Conditional Cooperator, 1 Time-Seeking Colleague. Rather than doing the private costs (or even utility functions) properly, let’s just assign each type a maximum FTE they can give up, and assume everybody goes for 75% of that on their first pledge, and then increases by 50% of the remaining headroom in each successive pledge (i.e. once for CAP #2 or ongoing for CAP #3) … with some kind of minimum increment to stop CAP #3 from going on forever.</p><p>The first interesting thing I found in this model is that the granularity makes a <em>huge </em>difference. I’m using a fairly unrealistic 0.01 FTE, but pushing that up to even 0.25 FTE makes a difference between solving the coordination and not solving it under two CAPs. This is pretty intuitive, to be honest: a department that has flexibility to accept more carefully, finely sculpted offers, with everybody giving exactly what they can afford, is more likely to be able to meet its target.</p><p>I ran two versions of the model, one with weak procedural preferences and one with strong procedural preferences, keeping the same department of 20 workers and a target reduction of 2.0 FTE. Both runs used a minimum increment of 0.01 FTE (which as I understand it would not fly for most HR departments — already way too small).</p><p>I specified procedural preferences numerically for each of the seven worker types. These consist of an ordinal ranking of the three procedures (O, one-shot; T, two-shot; R, repeated) and a participation-cost spread between the most and least preferred procedures. In the <strong>strong-preference scenario</strong>, these spreads ranged from 0.01 to 0.04 FTE-equivalent units, with some types indifferent between procedures. In the <strong>weak-preference scenario</strong>, the spreads were reduced to 25% of these values. I’m flagging this as one of the <em>most made-up bits </em>in the model (there should be more of a practice of this in modelling, I think).</p><p>I effectively assumed that participation costs reduce each worker’s maximum willingness to pledge, since the cost of participation was imposed on the pledge budget (see below). Initial pledges were set at 75% of this maximum, with subsequent pledges increasing by 50% of the remaining headroom, rounded down to increments of 0.01 FTE.</p><ul><li>With weak preferences, O raised 1.84 FTE, T raised 2.12 FTE, and R raised 2.32 FTE.</li><li>With strong preferences, O raised 1.78 FTE, T raised 2.05 FTE, and R raised 2.19 FTE.</li></ul><p>In both scenarios, O failed to meet the target, while T and R succeeded. Any excess pledges could be scaled back to produce a collective reduction of exactly 2.0 FTE.</p><p>In this model stronger procedural preferences reduced aggregate pledges under all three procedures, particularly R, but did not change which procedures succeeded. This is kind of trivial, but perhaps reflects a real world phenomenon, where there is disillusionment and disengagement from a process that some find alienating. The additional opportunities to revise pledges under T and R compensated for the greater participation costs.</p><p>After this first pair of runs, I thought about a department with a lot of Anxious Comrades: people willing to sacrifice quite a lot of FTE to avoid compulsory redundancy, but strongly preferring a one-shot survey to do so. In such a department, you might expect the intensity of procedural preferences to make a bigger difference.</p><p>So I constructed a second department of 20 workers, this time including seven Anxious Comrades, four Financially Stretched Carers, four Exhausted Solidarists, and five others. I ran the same three CAPs again, once with weak procedural preferences and once with strong preferences, keeping the target at 2.0 FTE and the minimum increment at 0.01 FTE.</p><ul><li>With weak preferences, the one-shot procedure raised 1.83 FTE, the two-shot procedure raised 2.05 FTE, and the repeated procedure raised 2.21 FTE. So the first failed, but the other two succeeded.</li><li>With strong preferences, the one-shot procedure raised 1.75 FTE, the two-shot procedure raised 1.97 FTE, and the repeated procedure raised 1.98 FTE. This time, all three failed.</li></ul><p>Obviously, the importance of procedural preferences depends partly on the composition of the department (types), and particularly on whether people willing to contribute substantial FTE reductions also dislike the procedure being used.</p><p>I didn’t actually realise at first that I was assuming that disliking a procedure reduces willingness to pledge!</p><p>That’s because you incur a cost from participating in a disliked procedure, and because each worker has a private maximum “budget” they can devote. It is a sort-of-plausible assumption, although I’m not sure I like it. How else might this be modelled?</p><p>Either way: a new direction might be to derive pledging behaviour from explicit utility functions incorporating private FTE-loss costs, the perceived departmental cost of compulsory redundancy, and participation costs. Then you could keep investigating whether stronger procedural preferences genuinely affect coordination and collective welfare, rather than building their effects into the bidding rules.</p><p>Where I might take this next is towards a hybrid mechanism in which participants have some choice over how they participate, rather than everyone being required to follow the same procedure. I’m interested in whether perceptions of fairness, the transparency of strategic incentives, and differences in what participants find obviously strategyproof affect whether the department can solve its coordination problem. In particular, might allowing people to engage in different ways improve the prospects of collective agreement, even when their underlying willingness to sacrifice FTE remains unchanged?</p><p>Key finding: I hate modelling.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e75856f9f4aa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CO2 AI]]></title>
            <link>https://medium.com/@jolindsaywalton/co2-ai-409f58baa80b?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/409f58baa80b</guid>
            <category><![CDATA[unfair]]></category>
            <category><![CDATA[commercial-practices]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[misleading-advertising]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Tue, 11 Aug 2026 13:42:26 GMT</pubDate>
            <atom:updated>2026-08-11T13:42:26.889Z</atom:updated>
            <content:encoded><![CDATA[<p>CO2 AI’s website states:</p><blockquote>Smart building systems powered by AI agents are already delivering impressive results. These systems continuously monitor occupancy, weather conditions, energy prices, and equipment performance to optimize heating, cooling, and lighting in real time. The result is typically 15–30% reduction in energy consumption without any impact on occupant comfort. Companies deploying IoT solutions, now adopted by 47% of organizations , create the sensor networks that enable these intelligent systems.</blockquote><p>Huge if true! Sounds like agentic AI is very advanced in building energy management. Better not fall behind!</p><p>I guess someone should tell the authors of <a href="https://arxiv.org/abs/2512.25055">this late 2025 preprint paper</a>, that they’re way too late, and it’s cased closed.</p><p>There are no sources cited for the 15–30% figure and the 47% figure. But they do <em>really </em>sound odd. 47% of organizations are deploying Internet of Things solutions? Really? 47% of <em>all </em>organizations?</p><p>A likely source of the 47% figure is <a href="https://newsroom.eclipse.org/news/announcements/eclipse-foundation-releases-2021-iot-and-edge-commercial-adoption-survey-results">a 2021 survey by the Eclipse Foundation</a>. The summary states:</p><blockquote>IoT technologies are being adopted at an accelerated rate. 47% of respondents currently deploy IoT solutions and an additional 39% plan to deploy within the next 12 to 24 months.</blockquote><p>Of course, one question is, “Who was being surveyed?” The answer is, comically, “more than 300 IoT and edge professionals.” People who work in IoT are probably more likely than most to be using IoT in their businesses.</p><p>What about the savings? A “typical” result being “15–30%” tells us that CO2 AI, or some source that they have relied on, have taken a large enough representative sample of organizations deploying agentic AI for smart building systems, trimmed off the outliers, and found that results generally fall between 15% and 30% of savings.</p><p>There could of course be some interesting conversations about the baseline here — should we not be comparing to non-agentic smart buildings systems?</p><p>But we can’t even get to <em>that </em>point.</p><p>One potential source for the 15–30% figure is <a href="https://eprints.whiterose.ac.uk/id/eprint/94475/1/RockettHathway-BR_I-final.pdf">this 2017 paper</a>. 2017: in other words, well before agentic AI was a thing, at least in the sense intended by CO2 AI.</p><p>The 2017 paper’s claim is not about agentic AI, of course, but about Model Predictive Control (MPC).</p><blockquote>Many of the reports, both simulation and for real buildings, estimate energy savings in the range 15–30% while maintaining or even improving internal comfort, which, at face value, is an impressive saving</blockquote><p>The authors moreover immediately cast doubt on the 15–30% paper.</p><blockquote>We conjecture that the reported energy savings with temperature-controlled MPC may be due to the fact the existing BMSs in the test buildings were poorly calibrated, and so almost any ‘care and attention’ lavished on such a building would have produced some energy savings. As further (circumstantial) justification for this suspicion, we are aware of a number of building-energy consultancies in the UK currently advertising that they can achieve 15–30% energy savings for their clients by careful scrutiny of building operation and ‘fine-tuning’ of the existing BMS.</blockquote><p>Beautifully put.</p><p>BMS is Building Management System — in short, the older approach that MPC was touted as improving upon, back in 2017.</p><p>It seems like the same statistic is still being used, in relation to completely different tech.</p><p>Why has this kind of thing become acceptable?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=409f58baa80b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tank URL: Amsterdam Digital Methods Summer School]]></title>
            <link>https://medium.com/@jolindsaywalton/tank-url-amsterdam-digital-methods-summer-school-0cf496046b72?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/0cf496046b72</guid>
            <category><![CDATA[communism]]></category>
            <category><![CDATA[tankie]]></category>
            <category><![CDATA[digital-methods]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[imperialism]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Fri, 17 Jul 2026 17:07:26 GMT</pubDate>
            <atom:updated>2026-07-17T17:56:35.338Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Amsterdam Digital Methods Summer School</strong></p><p>In early July I dropped in on the <a href="https://wiki.digitalmethods.net/Dmi/DmiSummerSchool">Amsterdam Digital Methods Summer School</a>, and contributed to Gavin Mueller’s data sprint on tankies — hopefully not derailing too much with the Sus AI Future project’s obsessions about AI and climate!</p><p>What’s a tankie? The question kept coming up over the week. Finding more precise definitions, or better yet a taxonomy (tankie-onomy) might be worth doing. For now, suffice to say that <strong>the term tankie is a pejorative for a certain kind of authoritarian leftist.</strong> It emerged on the British left after the Soviet Union sent tanks to suppress the Hungarian uprising of 1956, and was later applied to those who defended the crushing of the Prague Spring in 1968. “Tankie” therefore overlaps a lot with “Stalinist” — some might call them the same thing? I’d say the terms are not <em>quite </em>synonymous though, as tankie is a looser and broader jibe for leftists who excuse repression whenever it is carried out by an ostensibly socialist or anti-imperialist state …</p><p>… it’s also a jibe that can have a bit of paranoia about it: a way of saying to someone that they may not be excusing repression, but they’re acting like somebody who would, given half a chance. It’s about having not killed the cop in your head. In this sense, your threshold for who sounds like a tankie may vary. I’m probably The Tankie One of some of my Signal groups. I tend to think that climate change creates a genuine need for large-scale, coordinated mobilisation: rapid renewable-energy build-out, major changes in agriculture, and other forms of planned transition. So to more anarchisty friends, I might sound like a green authoritarian, somebody who <em>would </em>justify repression given half the chance. Send in the solar-powered tanks.</p><p><strong>Tankies might also be understood as a contemporary online subculture,</strong> consisting of statist and authoritarian communists who staunchly defend either the historical record of the USSR and figures such as Stalin and Mao, or present-day nominally communist states including China, Cuba, Vietnam and North Korea (AES: “Actually Existing Socialism”). There is some indication — one of the things Gavin’s research is exploring — that parts of this subculture arrived at being Very Online Leftists by way of being Very Online, at least as much as by being Very Leftist.</p><p>These communities are usually associated with an anti-imperialism organised primarily around opposition to the hegemony of the United States and its allies. Anti-imperialist messaging from within the imperial core, especially amid the global resurgence of fascism, faces an uphill struggle and deserves considerable sympathy. Political messaging is rarely conducted in conditions of perfect information or immaculate rhetoric. Exaggeration may have strategic uses — or epistemological ones; you don’t aim directly at a target if you know the wind is blowing strongly in a particular direction. Affective intensities can sustain solidarity. Movements confronting vastly more powerful institutions and discourses may knowingly fight dirty sometimes. Authoritarianism is bad, but your mileage on statism may vary.</p><p>At its worst, however, tankie-ism can harden into <em>geopolitical campism</em> — the idea that states opposed to Western power should be defended almost regardless of their conduct — while undermining its own credibility through dogmatism, selective scepticism, and analyses that shade into conspiracy theory.</p><p>Exploring <strong>tankie views on AI and climate</strong> therefore offers a revealing window onto broader political struggles over technology, ecological crisis, empire and competing visions of modernity.</p><p><em>First phase</em></p><p>Gavin and Ema had prepared two text corpora, scraped and transcribed from YouTube videos — one from independent YouTube tankie influencers (‘YouTubers’) and the other from traditional Marxist-Leninist organizations with a YouTube presence (‘organizations’). The website ProleWiki was treated in this study as a hub for this kind of tankie-ism.</p><p>There were a huge number of experiments that I won’t go into here (I’ll focus on the bits I was involved in).</p><p>At first we used a randomly downsized data sample, too small to draw firm conclusions, but interesting. It did look plausible that ‘tankie’ YouTubers think about AI and climate a little differently from the wider online left.</p><p>This corpus didn’t come out as either consistently pro- or anti-AI. YouTubers celebrate AI’s potential to reduce toil and raise productivity, while criticising environmental costs, job displacement, and dehumanising effects. More distinctively, they sometimes read AI as symptomatic of the contradiction between highly developed forces of production, and increasingly obsolete capitalist relations of production. Some use AI tools artistically — for example, to depict Marx as a bodybuilder.</p><p>Most distinctively of all, AI is also frequently linked to geopolitics — US, China, Russia, Ukraine, Israel, Palestine. This includes the IDF’s use of military AI in Palestine, and AI’s role in dis/misinformation and genocide denial. Tankies may contrast US and Chinese approaches, presenting socialist AI as more economically competitive and socially beneficial, sometimes linking this to China’s open-source model.</p><p>A crude estimate of how much AI appeared in a ‘geopolitics’ frame (how often do any of a set of AI-related words co-occur with any of a set of geopolitics-related words?) gave a fairly striking result: organisations 19% of the time, YouTubers 35% of the time.</p><p>Nathalia Henao beautifully visualised this, along with other findings, for a poster presentation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/1*iISMK3JwfOGcLs8zPPfx6g.png" /></figure><p><em>Second phase</em></p><p>Next I went back to the big dataset, and downsampled in a more targeted way to create four smaller datasets:</p><ul><li>individual YouTubers talking about climate,</li><li>individual YouTubers talking about AI,</li><li>organisations talking about climate,</li><li>organisations talking about AI.</li></ul><p>I tried some topic modelling. So according to outcomes, what do tankies talk about, when they talk about AI?</p><p>It sure looks like <strong>there’s a distinct Israel / Palestine topic (YTAI-T7) in the YouTuber dataset, that doesn’t appear so clearly in the organisation dataset</strong> (ORGAI-T0 is the closest, containing “Israel,” “imperialism,” “imperialist”). It is plausible that this might include mentions of systems like Lavender and Gospel / Habsora targeting tools, Where’s Daddy?, Unit 8200, Project Nimbus, Pegasus, as well as the use of genAI in Hasbara and more informal grassroots propaganda and advocacy — further probes, or some close-reading, could confirm.</p><p>There is also a notable absence across both corpora —<strong> neither AI corpus connects AI and climate consistently enough for any climate-related keywords whatsoever to turn up in the topics.</strong> This seems to go both ways — the climate corpora do not conspicuously feature AI or tech.</p><p>Other than that, everything is a bit more uncertain — interesting things to explore further, rather than solid results.</p><p>One obvious angle on AI for any leftist is automation, technological unemployment, and the impact on labour organising. Of the eight topics of the YouTuber AI corpus, two of them (YTAI-T4 and YTAI-T6) seem to include some of this focus. My impression is that the organisation AI corpus has one such topic (ORGAI-T2), but that labour issues are also more present in a more diffuse way across its other topics.</p><p><strong>YouTuber corpus: climate</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/949/1*N83Auu_hM7XehVY99_isLA.png" /></figure><p><strong>YouTuber corpus: AI</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*at3v6MqBJnNJEIqKrD-X2w.png" /></figure><p><strong>Organizations corpus: Climate</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1007/1*R2SULWobnIVOdq1yVCbLng.png" /></figure><p><strong>Organizations corpus: AI</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/994/1*oc-dVa1oEJyKwLtBw9o6GQ.png" /></figure><p>The data sprint was exhilarating, informative, and highly entertaining. It was also a useful reminder of how much uncertainty, interpretation, assumption, and ambiguity lies behind apparently polished computational results. Once data has been processed and presented through technical systems, those caveats and contingencies can fade from view beside the brightly gleaming output. This is a recurrent concern across digital humanities, critical AI studies, science and technology studies, and our own BRAID project Sustainable AI Futures: how proximity to computation can lend findings an impression of precision and security that exceeds their evidential basis.</p><p>In some cases, this effect is stronger for audiences than for analysts themselves. Modellers, for example, often feel in their bones the limitations of their art, distributed across the various assumptions, simplifications, uncertainties, validations-yet-to-be-run and so forth built into their models, but may underestimate how readily those qualifications disappear once the results circulate beyond the modelling context.</p><p>Perhaps we need more ‘soft’ humanistic tools for governing these encounters between technology and meaning? Here’s a tool we could use, that emphasises the situated, relational qualities of knowledge.</p><p><em>1. Before conducting the analysis, describe some of the differences you would expect to find. Give a sense of how confident you are in these expectations. Describe your positionality / situatedness a little bit too, in the sense of where these expectations come from: Where have you picked up the knowledge that lets you make these predictions? Think about what you might hope or want to find and why you hope or want to find it.</em></p><p><em>2. After the analysis, describe how your understanding has shifted. Have some things been confirmed, strongly or weakly? What surprised you? Do you have new expectations or questions?</em></p><p>I thought of this too late to do it properly! But here’s a version reconstructed after the fact.</p><ol><li>Most of my understanding of AI and climate has been formed through a mixture of academic commentary, predominantly from Europe and North America (sometimes seeking to elevate Indigenous and decolonial approaches; environmental-justice movements in the Global South); UK civil-society organisations; a burgeoning grassroots anti-AI movement primarily in the UK and US; and practitioners working in green software and digital sustainability. Some of the latter seemed to have begun a few years ago with relatively technical questions — energy-efficient code, carbon measurement, cloud architecture, hardware utilisation — but have become increasingly interested in political economy as they encounter friction and limitations in their technical intervention. This movement beyond narrowly technical approaches is also visible in sustainable-software research, which increasingly treats environmental outcomes as products of interacting social, economic, and technical systems.</li><li>Among the sources I’m in regular contact with, green and environmentalist traditions predominate, with a political spectrum that spans the liberal and centre-left to more anarchist and anti-authoritarian communist (workerist and autonomist type traditions) positions. Many are receptive to post-growth and degrowth arguments, and interested in decentralisation, democratic control of infrastructure and forms of participatory decision-making. There is some continuity here with the climate movement, e.g. Extinction Rebellion and its emphasis on protest, people’s assemblies, distributed organisation, and deliberative participation. These traditions also connect with early grassroots resistance to contemporary AI: sometimes involving some of the same people, similar tactics and strategy, local opposition to data centres, campaigns around water and electricity use, resistance to surveillance and automation, and demands that affected communities should have meaningful power over digital infrastructure. Insofar as I’m regularly in touch with more statist presuppositions on this topic, those tend to be liberal and social democratic, rather than socialist or communist. I also try to read climate science straight from the horse’s mouth (IPCC) which complicates all this a little. I also have some familiarity with more right-leaning accounts, via my own sceptical and critical readings, and quite a bit of familiarity with corporate sustainability literature, again primarily through a critical lens — somewhere between constructively critical and outright antagonistic.</li><li>This leads to a discursive situation where any mention of AI and/or climate can quite quickly lead to a critique of US big tech, either from some kind of centre-left position — these firms need to be better regulated, or broken up — or from a more radical left position — the problem is capitalism. The equivocal phrase ‘system change,’ used a lot within the climate movement, encompasses both possibilities. Are you changing the institutions? The laws? The relations of production?</li></ol><p>Against this background, my expectations were:</p><ol><li>that the tankies might be more inclined to talk about AI in relation to <strong>geopolitics</strong>, especially US-China rivalry, than to connect AI with extractivism and climate change.</li><li>I also expected <strong>greater optimism about AI’s productive capacities</strong>, and a stronger tendency to treat automation as potentially emancipatory, with its harms arising primarily from capitalist ownership and control rather than from the technology itself. I expected greater <strong>techno-solutionism</strong>, i.e. greater faith in AI’s potential to contribute to climate mitigation and adaptation.</li><li>I expected less critique of <strong>unemployment</strong>, and perhaps for the organizations to be more interested in unemployment than the YouTubers.</li><li>I was curious as to whether there would be <strong>climate denialism</strong>, or indications of the seriousness of climate change being downplayed.</li><li>I was curious about <strong>antisemitism</strong> and <strong>conspiracy theories.</strong></li></ol><p>How have my intuitions shifted after the investigation? There is nothing that very strongly contradicts them.</p><ol><li>Expectation 1 has actually strengthened quite a bit: there was a lot of geopolitics! And I was surprised by the absence of <em>any </em>connection between AI and climate!</li><li>Nothing to contradict the greater optimism about AI, and maybe some weak confirmation as well. I was also slightly surprised by how readily AI became an occasion to return to Marxian economics, although in retrospect this should perhaps have been entirely predictable: “as Marx himself wrote…” I had also expected a clearer emphasis on public ownership, state planning and large-scale infrastructural coordination, alongside less interest in decentralisation, local consent, participatory governance or limits to technological expansion. That expectation now seems too simple. Praise for Chinese AI, for example, may include praise for open-source models, which complicates any straightforward opposition between centralisation and openness. I didn’t get a clear read on climate techno-solutionism as such.</li><li>No clear signal, really. I didn’t really see much of a difference in the way YouTubers and organizations talked about jobs and the impact of AI on the labour market. Feels like one to keep investigating.</li><li>Didn’t see any evidence of climate denialism. Not sure if it would have shown up.</li><li>Didn’t see any evidence of antisemitism (unless you count the word “Jew” itself), or of conspiracy theories, but again, not sure this would have shown up here.</li><li>The differences between the YouTuber and organisation datasets were less pronounced than I had anticipated. That, in itself, suggests the value of building comparator datasets representing other parts of the left, or perhaps a wider political spectrum.</li></ol><p>The repertoire of contention with which I am most familiar includes: despondency about agency or the sense that things remain at an early stage — “we need to move from awareness to action”; attempts to influence policy through evidence and argument; demands for greater transparency from powerful actors, especially technology companies; the articulation of alternative AI imaginaries, often more decentralised, democratic and environmentally regenerative; demonstrations and direct actions with recognisable climate-movement DNA; and grassroots experiments in democratic accountability, particularly through people’s assemblies.</p><p>The exercise also made me realise that I lack a concrete sense of the tankie repertoire of contention, or what tankie optimism looks like in practice. I understand the broad historical claim that Actually Existing Socialism will, or already does, outperform capitalism. I am less clear about what tankies expect to unfold, or about the political practices that follow from their expectations.</p><p>Special thanks to the Sussex Digital Methods Accelerator and HEIF for supporting participation in the Summer School.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0cf496046b72" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Some resources for AI and the environment]]></title>
            <link>https://medium.com/@jolindsaywalton/some-resources-for-ai-and-the-environment-92454b2cafba?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/92454b2cafba</guid>
            <category><![CDATA[carbon-emissions]]></category>
            <category><![CDATA[sustainability]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[energy]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Tue, 21 Apr 2026 23:26:00 GMT</pubDate>
            <atom:updated>2026-04-21T23:46:18.701Z</atom:updated>
            <content:encoded><![CDATA[<p>If you’ve recently been to a Sustainable AI Futures event, we’d appreciate your feedback: <a href="https://docs.google.com/forms/d/e/1FAIpQLSegxozpnvIIkWyVVLA6zNFHRfdi0QnYCfucdhXPMwbgTvMYcQ/viewform?usp=header">https://docs.google.com/forms/d/e/1FAIpQLSegxozpnvIIkWyVVLA6zNFHRfdi0QnYCfucdhXPMwbgTvMYcQ/viewform?usp=header</a></p><p>Meanwhile, some resources …</p><ul><li><strong>Green Software Foundation: </strong>non-profit creating a trusted ecosystem of resources for building green software — <a href="https://greensoftware.foundation/">https://greensoftware.foundation/</a></li><li><strong>Branch: </strong>online magazine from the Green Web Foundation, dreaming of a sustainable and just internet for all — <a href="https://branch.climateaction.tech/">https://branch.climateaction.tech/</a></li><li><strong>The Green Web Foundation:</strong> <a href="https://www.thegreenwebfoundation.org/">https://www.thegreenwebfoundation.org/</a></li><li><strong>Green Screen Coalition: </strong>coalition at the intersection of climate justice and digital rights, funding projects for equitable sustainable infrastructure — <a href="https://greenscreen.network/en/">https://greenscreen.network/en/</a></li><li><strong>Carbon Market Watch: </strong>watchdog reporting on carbon markets (offsetting with carbon credits, carbon pricing etc.) — <a href="https://carbonmarketwatch.org/">https://carbonmarketwatch.org/</a></li><li><strong>CarbonBrief: </strong>pretty good news site for general carbon and climate issues, a bit easier to read than IPCC reports — <a href="https://www.carbonbrief.org/">https://www.carbonbrief.org/</a></li><li><strong>Distributed AI Research Centre: </strong>one example of a research group in the field of Critical AI Studies — <a href="https://www.dair-institute.org/">https://www.dair-institute.org/</a></li><li>Data Center Watch: <a href="https://datacenterwatch.org/">https://datacenterwatch.org/</a></li><li>Data Center Impact Dashboard: <a href="https://datacenterimpactdashboard.com/">https://datacenterimpactdashboard.com/</a></li><li>No Data Centers on Native Land: <a href="https://www.honorearth.org/stopdatacolonialism">https://www.honorearth.org/stopdatacolonialism</a></li><li>Beyond Fossil Fuels: <a href="https://beyondfossilfuels.org/">https://beyondfossilfuels.org/</a></li></ul><p>More to come …</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=92454b2cafba" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Translating alternative AI imaginaries into reality]]></title>
            <link>https://medium.com/@jolindsaywalton/translating-alternative-ai-imaginaries-into-reality-663f03634f52?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/663f03634f52</guid>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[sustainability]]></category>
            <category><![CDATA[climate]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Fri, 13 Mar 2026 12:24:51 GMT</pubDate>
            <atom:updated>2026-03-13T12:27:03.269Z</atom:updated>
            <content:encoded><![CDATA[<p>The <a href="https://impact.indiaai.gov.in/">AI Impact Summit</a> is in full swing in New Delhi [<em>at least it was when I wrote this</em>], bringing politicians and policymakers, industry leaders, and researchers together at the Bharat Mandapam to debate the future of AI.</p><p>It arrives at a moment when AI is increasingly framed not just as an economic engine, but as an indispensable tool for <a href="https://www.pwc.com/gx/en/issues/value-in-motion/three-tomorrows.html">stopping climate change</a>. Yet at the same time, concern has been growing about AI’s own climate costs, including <a href="https://www.theguardian.com/environment/2026/jan/29/gas-power-ai-climate">growth of gas-fired plants</a> in the US to power data centres, alongside mounting pressure on water systems for cooling.</p><p>Will AI save the planet? The AI Impact Summit comes hard on the heels of a new report, <a href="https://beyondfossilfuels.org/wp-content/uploads/2026/02/AI-for-climate-claims-Report_FEB-2026_FINAL-2-16.pdf">‘The AI Climate Hoax: Behind the Curtains of How Billionaires Greenwash AI,’ </a>commissioned by a coalition including Friends of the Earth, the Green Web Foundation, and the Green Screen Coalition.</p><p>“An environmentally destructive heavy industry has emerged in the blink of an eye,” says the report, lead-authored by investigative journalist Ketan Joshi. It goes on offer evidence of greenwashing by major AI companies.</p><p>Last year, our own report <em>The Cloud and the Climate</em> cast doubt on some evidence that AI will benefit the climate. For example, we found that one peer-reviewed paper, which has since been retracted, <a href="https://theconversation.com/will-ai-save-the-planet-why-the-evidence-is-flawed-240133">was filled with errors</a> probably caused by AI writing tools.</p><p>This new report confirms these suspicions with a more systematic analysis. Examining claims that AI will benefit the climate, it finds that: “only 27% cited published academic papers. 37% did not cite any evidence. Of the rest, 28% cite corporate publications, and 8% cited media, NGOs or unpublished academic papers.”</p><p><strong>What’s wrong with ‘net climate impacts’</strong></p><p>The report also shines light on what it describes as a new greenwashing tactic. The emphasis on a “net climate impact” of AI, in which costs and benefits are lumped together, is a misleading distraction. The problem arises because many environmental costs are already real, whereas the benefits are future projections — and perhaps, empty promises.</p><p>Of course, climate policy cannot ignore potential gains. Climate policy is, by definition, future-facing. The question is not whether to consider possible benefits, but how to treat them prudently and fairly.</p><p>But the ‘net benefit’ argument has another problem. It collapses very different AI systems into a single category. Picking them apart, the report found that “the overwhelming majority of AI claims relate to ‘traditional’ forms of AI rather than generative AI.”</p><p>In other words, many AI systems that could deliver real environmental benefits are <em>not </em>the same large, well-known AI models like ChatGPT, Copilot, or Claude. They can run on far more modest hardware, with dramatically lower energy use and emissions.</p><p><strong>A turning point for AI?</strong></p><p>“Concern is growing” may be putting it mildly: an AI backlash is brewing. A grassroots movement, Pull The Plug, is organising a protest march at the end of this month, calling for citizens’ assemblies to democratise AI. Global Action Plan are holding rallies outside data centres, while the campaigners Foxglove are challenging new-build data centres through the courts. There are also hints that resistance to AI does not map neatly onto the political spectrum: concerns about ‘sovereign AI,’ and the influence of US tech, appear to resonate at least as strongly on the right as on the left.</p><p>This popular response unfolds amidst growing anxieties about a US tech bubble. <a href="https://www.ft.com/content/82100215-18d0-4850-8dcd-d9745ad342d0">The European Central Bank has warned</a> that valuations for companies such as Nvidia, Alphabet, and Microsoft have become “stretched”, and some fund managers <a href="https://www.ft.com/content/48d9c100-0ec6-4edf-9395-eb44879ea5c6">quoted by the <em>Financial Times</em></a> have gone further, describing “insane” valuations and investment in data centres and AI infrastructure racing ahead of near-term returns. In January, even <a href="https://www.ft.com/content/a1f04b0e-73c5-4358-a65e-09e9a6bba857">the chief executive of Google DeepMind</a> conceded that parts of the sector “looked bubble-like”, while emphasising that “if the bubble bursts, we” — meaning Google DeepMind — “will be fine.”</p><p>When financial bubbles burst, the world does not reset. <a href="https://theconversation.com/the-ai-boom-feels-eerily-similar-to-2000s-dotcom-crash-with-some-important-differences-269472">The Dotcom Crash of 2000</a> didn’t put a stop to the internet. But it did reshape the internet we got. In the aftermath of a bubble, a dominant narrative can lose its hold, and space may open for alternative stories. To date, the loudest claims about AI and climate have been closely linked to the companies building the largest models.</p><p>If the tables turn, one question is: will there be credible alternative narratives waiting in the wings? Convivial technology, neo-luddism, solarpunk, Critical AI Studies, minimal computing, decomputing, AI abolition, pluriversal politics, appropriate technology, slow AI, degrowth AI, decolonial AI: there’s a rich mix of thinking out there. But is it in any shape to seize the moment, if the moment comes?</p><p><em>More: </em><a href="https://susaifutures.com/"><em>susAIfutures.com</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=663f03634f52" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why we need “AI” for climate transition]]></title>
            <link>https://medium.com/@jolindsaywalton/why-we-need-ai-for-climate-transition-9c03a311ac1c?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/9c03a311ac1c</guid>
            <category><![CDATA[climate-action]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Mon, 23 Feb 2026 23:07:19 GMT</pubDate>
            <atom:updated>2026-06-26T10:07:17.938Z</atom:updated>
            <content:encoded><![CDATA[<p>Why we need “AI” for climate transition</p><p>And by “AI” I mean Assured Interventions (also known as Actual Impacts).</p><p>AI’s resource use — energy, water, minerals — is attracting plenty of attention and resistance, as temperatures soar this summer in the north.</p><p>The AI apologists will point to AI’s potential to deliver “climate solutions.” They may also remind us that data center emissions, in the grand scheme of things, are not making that huge a contribution to climate change.</p><p>Are they right? Not <em>really</em>.</p><p>On the first point, Ketan Joshi’s recent devastating report <a href="https://foe.org/wp-content/uploads/2026/02/AI-for-climate-claims-Report_FEB-2026_FINAL.pdf"><em>The AI Climate Hoax</em></a><em> </em>shows how the most promising climate use cases have little to do with the overall strategic direction of the AI industry. They are also highly uncertain — AI systems with climate benefits might emerge, they might scale, but they might not. It’s not the first time technological fixes to climate change have been confidently proposed.</p><p>The second point is a bit more nuanced. The energy demand of data centers is significant, growing, and uncertain — because of a stubborn lack of transparency on the part of major data center operators — but sure, compared to things like food production, or building energy use, or road transport, or even cement production, it’s not all that high.</p><p>For that reason, I am a <em>little </em>concerned that grassroots climate activism is placing <em>quite </em>so much emphasis on AI and data centers these days.</p><p>Then again — as many grassroots organisers clearly understand — the direct impacts of AI systems are not the heart of the issue.</p><p>Nor is it even really the use of AI systems to prolong fossil fuel extraction that would otherwise be untenable (as important as that is too).</p><p>At the real heart of the issue is governments zealously clinging to a climate transition that is uncertain, tech-centric, and politically weak because it is inequitable. This approach asks sceptical publics to accept real burdens now — “behaviour change,” pressure on prices for food, energy, transport, housing— in return for later benefits. The costs and benefits are not even joined up in most people’s minds.</p><p>The answer isn’t better communications for the same unjust climate transition. It isn’t leaving the door ajar to climate denying right-wing political parties, or pandering to their core supporters on topics like immigration.</p><p>The answer is interventions that will actually work, as part of broader investment in public goods and services, including public luxuries. The problem isn’t imagining such futures — there are by now kajillions of projects that do so. It’s about putting them into practice.</p><p>So I propose a category of climate policy tools called <strong>Assured Interventions </strong>(“AI”). They are policy packages in areas where there is high certainty that:</p><ul><li>there are large potential CO2e savings,</li><li>these savings could begin in the near term, and</li><li>the only missing pieces are legal and financial.</li></ul><p>These are interventions that <em>don’t </em>rely on technologies that have yet to be proven, or yet to scale.</p><p>We need Assured Interventions for climate transition. What would <em>you </em>count as Assured Interventions?</p><p>Sidebar. Why the ‘not technologies that have yet to scale’ qualification? Isn’t that just the same as ‘legal and financial barriers’ — when you throw political will and the investment at a proven technique, doesn’t it always scale?</p><p>Maybe, but not necessarily — technologies don’t always work logistically or behaviourally at a small scale and a big one.</p><p>Climate science and climate policy is filled with uncertainty. This is something we often don’t talk about, because it gets so easily weaponised by climate change deniers. But it’s pretty obvious — we’re talking about the future here! Of the whole world!</p><p>Read any report from the IPCC (the global scientific body responsible for climate change science), you’ll quickly see it. It’s a patchwork of various degrees of confidence and likelihood.</p><p>Put another way. If we could somehow pull it all off, and meet 1.5 degrees after all — like the sudden last-minute kinky progress of a Curling stone — how would we do it?</p><p>What would you include?</p><p>You know more than me about this.</p><p>But here are my suggestions.</p><p>Rapid switch to <strong>plant-based diets</strong> (sorry) is a strong candidate. It seems like word is out: using the key metric “What source of global warming do my students remember the most” the clear winner is burping cows. Short-lived climate pollutants, e.g. methane, have a strong warming effect in the near term. Let’s start treating lentils and chickpeas as critical planetary infrastructure. Let’s demand Universal Basic Oowie Burger.</p><p>But I’m not primarily talking about behaviour change here. Policy levers include procurement, tax and subsidies, hospitality sector regulation, and public-sector defaults. Schools, hospitals, universities, police, militaries, shape demand at scale and shape norms through routine provision.</p><p>Why isn’t there a proper Sustainable Food Incentive for the hospitality sector yet? I know chefs are scary, but hotels and restaurants are another key lever.</p><p>Cutting methane can slow temperature rise within years. That buys time while we cut carbon more deeply.</p><p>Then there are <strong>fugitive emissions</strong>. According to Global Methane Tracker, the fossil fuel sector is responsible for nearly one-third of anthropogenic methane emission. Emissions have been over 120 million tonnes (Mt) annually. Abandoned wells and mines contributed around 8 Mt to these emissions in 2024. Is this an area where environmentalists haven’t focused sufficient attention, because the temptation is to focus on ending fossil fuel production, rather than tweaking it? Sure, but there are huge savings to be made in the short term.</p><p>What do we want? Tweak the leaks! When do we want it? Before 2030! Leak detection and repair requirements reduce emissions when inspections occur frequently and repairs follow quickly. The Global Methane Tracker reckon 5% of production is already near a zero emissions standard: in that sense there aren’t technological barriers. Policies and practices with a proven record already exist and can scale up — mandating zero-emitting pneumatics and other low-emissions kit; publicly funded identification and remediation programmes for orphan wells; binding restrictions on routine flaring and venting, maybe — while measurement systems and disclosures continue to improve.</p><p>But this doesn’t go far enough. There is a long history of underreported methane emissions. There is a clear mandate for <strong>direct, continuous, independent observation as the default regulatory approach to fugitive emissions</strong>. Independent satellite and aerial measurements have shown emissions far above inventory estimates in major producing regions. That gap between reported and observed emissions signals a structural weakness in a system built around self-reporting.</p><p>So the Assured Intervention here is publicly funding data collection and management infrastructure to ensure emissions are visible to regulators in close to real time. Regulatory independence (including proper resourcing), direct observation, transparent and open data, meaningful penalties, binding repair requirements. When the plume appears, the very next thing to appear should be the environmental agency officers, in a puff of smoke of their own.</p><p>Then what? Repair deadlines can follow directly observed and measured releases, with financial penalties scaled to the warming impact of methane and indexed upward for repeat exceedances, and criminal prosecution right there if you need it. Operating permits can incorporate methane intensity benchmarks verified through independent observation. For what it’s worth, investors also gain more robust information for risk assessment.</p><p>If regulators could hunt down leaks with half the zeal of <a href="https://www.bbc.co.uk/news/articles/cp9142nlpvyo">Manchester City Council hunting down parking violations</a> …</p><p>There is something curious here: this methane has a market value. Some estimates suggest that a lot of this could be cost-neutral or <em>cost negative</em>. But let’s move on.</p><p>Current financial and political incentives make clearing forests profitable. Afforestation and reforestation is great. But bolder policy to prevent deforestation<strong> </strong>could be another strong candidate for Assured Intervention.</p><p>This is somewhat tricky territory, since REDD+ has had such a troubled and complex history. Ethically complex, as much as anything. Climate transition also needs to be just, so we might focus on a subset of AI which I call <strong>Global Equity-Normed Assured Interventions</strong> or “genAI.” We don’t just need AI for climate transition: we need genAI for just transition.</p><p>Deforestation is very developmentwashed: it is often described as the unavoidable cost of development in the Global South. At the level of individual farming and pastoralist communities, yes, clearing land often increases income and food security, especially where land rights are insecure and support is limited. But there are inequalities within countries as well as between countries, and there’s a systemic picture here. It’s clear that most large-scale, permanent deforestation today is not mainly serving the stars of those small-scale stories.</p><p>On balance, I think <strong>bold anti-deforestation measures</strong> are a strong candidate not just for Assured Interventions, but Global Equity Normed Assured Interventions. Adopting food sovereignty principles, securing land rights for local communities, meeting and expanding loss and damages commitments, empowering the Indigenous communities stewarding some of the planet’s deepest carbon sinks, improving productivity and transport infrastructure on land that is already cleared, investing in rural public services and reforming supply chains can raise incomes without pushing agricultural frontiers further into forests.</p><p>Cookstoves are controversial, and the additionality of some cookstove-driven offsetting has been rightly roasted. But there is evidence black carbon and incomplete combustion of traditional biomass used in cooking and heating in developing economies may be a pretty big contributor. We’re not quite in Assured Intervention territory there, I think, but close.</p><p><a href="https://ccaf.io/cbeci/"><strong>Ban cryptocurrency</strong></a><strong>.</strong> Crypto, especially proof-of-work crypto, is a large, discretionary energy use. Phasing it out would free up substantial power — and maybe even infrastructure.</p><p>Joshi’s <a href="https://ketanjoshi.co/2026/02/17/big-tech-greenwashing-report/">recent report on AI greenwashing</a> indicates that “Artificial Intelligence for sustainability” discourse is based far more on optimism than robust evidence. More importantly for our purposes, the report shows how “Artificial Intelligence for sustainability” often conflates distinct forms of Artificial Intelligence, frontier models (for which they’re building all the new hyperscaler data centers) with ‘traditional’ Artificial Intelligence (with far lighter infrastructural requirements).</p><p>My back-of-napkin estimates suggest that by sunsetting the global crypto industry, we <em>comfortably</em> free up enough energy to train and deploy core climate-critical Artificial Intelligence to 2035.</p><p>Stretch goal: could former crypto infrastructure <em>itself </em>be used for climate mitigation and adaptation compute purposes? The reusable components of a crypto mining operation sit around the ASICs (alien palaentologists will dig these gizmos from our ruins and marvel, just marvel) rather than inside them. Anecdotal evidence of miners pivoting toward HPC illustrate this point, since they typically replace mining rigs with new GPUs while retaining site infrastructure.</p><p>Use a baby-cut carrot and stick approach: prohibit high-energy mining, while offering former operators preferential access to the new booming industry in government-subsidised climate data centres. Maybe something like:</p><ul><li><strong>Interconnection priority</strong> for providers / sites that exit proof-of-work and convert to HPC under clean-power conditions. Interconnection queues can be bottlenecks.</li><li><strong>Public compute offtake contracts.</strong> Governments could bundle demand from relevant actors like universities and research institutes, meteorological services, established climate adaptation services, into long-term contracts that make retrofits financeable.</li><li><strong>Concessional retrofit finance</strong> tied to covenants. Incentivise permanent exit from proof-of-work, public interest compute, and meeting renewable energy standards (probably including 24/7 hourly matching. See also e.g. Switzerland’s SDEA eco-label).</li></ul><p>Who knows, a managed transition from crypto data centres to climate data centres could just be the start.</p><p>So there you go, that’s what I’m calling ‘AI’: massive and delicious demand-side mitigation of meat methane, nationalise LDAR for fugitive emissions, stop deforestation with community rights and loss and damages funding, ban cryptocurrency. All within the context of a broader move toward publicly-funded sustainable systems that meet basic needs and deliver basic luxuries too — because without that, people <em>know </em>you’re ripping them off to futher bloat the value of the AI grifter class and their closest lackeys and alcolytes.</p><p>Clearly these need tailoring for different local contexts, and clearly there’s more to climate transition than AI — renewables, storage, building retrofits, EVs and public transport, fossil fuel phase-out, an industrial net zero green jobs and green skills pipeline strategy, etc. etc. — but AI is a promising space to advocate for urgent and bold policy now.</p><p>If we are serious, we should take actions that change the physical system reliably and quickly. What are these priorities? What do you think?</p><p><em>More: </em><a href="https://susaifutures.com/"><em>Sus AI Futures</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9c03a311ac1c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Modelling diverse futures for AI and the Climate]]></title>
            <link>https://medium.com/@jolindsaywalton/modelling-diverse-futures-for-ai-and-the-climate-e28aa3503ac8?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/e28aa3503ac8</guid>
            <category><![CDATA[sustainability]]></category>
            <category><![CDATA[greenwashing]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Tue, 17 Feb 2026 09:34:11 GMT</pubDate>
            <atom:updated>2026-03-13T13:34:59.080Z</atom:updated>
            <content:encoded><![CDATA[<p>We are pleased to share this preprint from the research project Sustainable AI Futures:</p><p><a href="https://drive.google.com/file/d/1hFOGihLFEb7EVx3VlBInFKSfDeOaKOzZ/view?usp=drivesdk">Modelling diverse futures for AI and the Climate</a></p><p>Feedback very welcome.</p><p>What’s it all about, roughly speaking? It is about trying to push forward how we think about AI-related policymaking, civil society advocacy, and grassroots action.</p><p>Big tech, responding to its own failing climate pledges, and concerns over the wasteful and ominous impacts of recent AI systems, has aggressively marketed the purported climate benefits of AI. Some of the claims, as you might expect, are at best exaggerations.</p><p>This has understandably led to a lot of interest in doing the calculations credibly and properly, and discovering the ‘real’ climate impact of AI. However, we argue that this is a dangerous framing, for at least two reasons.</p><p>First, it quickly gets you mired in methodological disputes. For instance, do we count emissions from AI-enabled oil and gas exploration in the ‘costs’? These are disputes that can’t be answered on purely technical grounds. The question of how to keep well below 2 degrees, given current and reasonably projected technologies, cannot wait for several years while researchers contest each other’s scope boundaries, and (if we are lucky) gradually get consensus on how different methodologies might be appropriate for different purposes.</p><p>Second, this framing uses a linguistic and cultural sleight of hand to present a set of very different technologies — which require different resources, do different things, and can be subjected to different laws or acts of refusal and resistance — as though they were more or less “one thing,” AI. This does not help policymakers, who need to intervene in a more granular way. It also undermines public understanding of AI, and grassroots movement building to resist AI harms. It also plays into a longer history of positioning technological “solutions” to climate change in ways that sideline effective, justice-oriented social and economic climate policy.</p><p>There is <em>some</em> feeling within our project team that this paper might not go far enough — that it is still swallowing too much AI hype, and accepting the implausible premise — implausible when you consider AI’s overall political economy, material infrastructure, and impacts on how we relate to eachother as humans — that some lighter-weight AI systems can work for the common good of people and planet.</p><p>What do you think? Is our preprint still stuck in the paradigm of ‘some AI bad, some AI good, we can tell the difference’ that we often see in Responsible AI? Does it actually accept a violent, extractive, colonialist future, while giving Big Tech sustainability leads of the future a few tips on how to better sell them?</p><p>We also look in depth in what is by far the best and most transparent model out there, and find some quite serious problems with it. We hope to be constructive with these criticisms. We are a very small team, and to date it has been frustrating to have spent so much time trying to track down sources that may not even exist. We have recently provided this preprint to Lord Stern and some of his co-authors, and are hopeful for a dialogue and even a collaborative approach.</p><p>Some expert advice from the modelling community suggests that we are being unfair here — that this technology adoption model is entirely standard and fit-for-purpose within the field, and the kind of nitpicking we are doing mostly just demonstrates our unfamiliarity with modelling norms, and the intrinsically difficult and uncertain nature of modeling work. There is also concern for the risk of undermining climate modeling, particularly at a moment of populist backlash against climate science. What do you think?</p><p>I expect this paper to change, maybe quite a lot, before it is published (if it is published). But it feels timely to share it now. Here is the formal abstract:</p><blockquote>How should recent models of interactions between AI and the climate be understood from the perspective of policy design and public accountability? We take as our case study Stern et al. (2025), a study that projects that 2022-2035 global carbon savings enabled by artificial intelligence will exceed AI’s own energy-related emissions. The study methodologically advances earlier estimates, combining expert elicitation with historical technology-adoption data across three domains, energy, food and transport, and concluding that AI is likely to be a net climate benefit, contingent on active and effective public governance. Although we concur with the recommendation of active and effective public governance, we challenge the framing of ‘net climate benefit / cost.’ This framing risks fostering an impression of AI as broadly self-offsetting, in ways that may obscure the conditions under which net climate benefits fail to materialise. Similarly, it risks obscuring crucial distinctions between diverse AI systems with diverse infrastructural requirements. Stern et al.’s accounting boundary focused on electricity use while excluding embodied and enabled emissions is moreover potentially overly narrow. However, in this response, we use this to reinforce our position that AI’s net carbon impact is deeply uncertain, rather than to propose improved boundaries. Against the background of deep uncertainty, modelling has the potential to help identify where, when, under what constraints, and with what uncertainties particular AI applications deliver genuine decarbonisation benefits, especially if deployed as an explicitly uncertain and imaginative practice for generating multiple, radically different future pathways. Contrary to current practice, a robust approach to climate governance requires open data, explicit communication of deep uncertainty, and inclusion of a broad range of stakeholders.</blockquote><p><em>More: </em><a href="https://susaifutures.com/"><em>susAIfutures.com</em></a><em> (BRAID UK)</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e28aa3503ac8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Two AI and the Environment Workshops]]></title>
            <link>https://medium.com/@jolindsaywalton/two-ai-and-the-environment-workshops-0711c0c2437d?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/0711c0c2437d</guid>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[climate-action]]></category>
            <category><![CDATA[policy]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Wed, 04 Feb 2026 17:16:40 GMT</pubDate>
            <atom:updated>2026-03-13T12:21:27.630Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>#1)</strong> On 1 April, we’re having a small, industry-focused event about <strong>AI and sustainability reporting</strong>. We’d be very interested in hearing from you if you are a sustainability professional, if you work with AI and sustainability reporting, or if you are a policymaker or a policy-focused CSO with interests in this area. E-mail: j.c.walton@sussex.ac.uk.</p><p><strong>#2)</strong> Some time in the Autumn, we’re planning another workshop focused on <strong>alternative AI futures</strong>. If you’ve got thoughts about how to make such an event worthwhile, we’d love to hear from you. In the meanwhile, we’d also be very interested in what you think we should be asking for from legislators and policymakers (and to some extent industry and other actors). Please add your ideas here: <a href="https://docs.google.com/document/d/1OSuB_D707aJlT_YwjgmemLOzVf2F7LbjNu8E-eU-txQ/edit?usp=sharing">https://docs.google.com/document/d/1OSuB_D707aJlT_YwjgmemLOzVf2F7LbjNu8E-eU-txQ/edit?usp=sharing</a></p><p><em>More: </em><a href="https://susaifutures.com/"><em>Sus AI Futures</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0711c0c2437d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An easy win for AI and climate policymaking]]></title>
            <link>https://medium.com/@jolindsaywalton/an-easy-win-for-ai-and-climate-policymaking-a6f3643a607a?source=rss-25ffba2f748e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a6f3643a607a</guid>
            <category><![CDATA[sd]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[secr]]></category>
            <category><![CDATA[csrd]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jo Lindsay Walton]]></dc:creator>
            <pubDate>Wed, 04 Feb 2026 13:50:29 GMT</pubDate>
            <atom:updated>2026-02-04T16:45:57.792Z</atom:updated>
            <content:encoded><![CDATA[<p>Let’s try an experiment. <strong>What concrete measures should lawmakers introduce to improve transparency in the tech sector?</strong></p><p>Add your answers here:</p><p><a href="https://docs.google.com/document/d/1OSuB_D707aJlT_YwjgmemLOzVf2F7LbjNu8E-eU-txQ/edit?usp=sharing">https://docs.google.com/document/d/1OSuB_D707aJlT_YwjgmemLOzVf2F7LbjNu8E-eU-txQ/edit?usp=sharing</a></p><p>One piece of low-hanging fruit.</p><p>As data centres expand rapidly and their environmental footprint grows, there is a strong case for greater transparency.</p><p>The power grids serving the data centres of companies such as Apple, Google, and Meta, vary widely in carbon intensity.</p><p>At a minimum, large technology firms should be required to clearly separate total corporate electricity use from the electricity consumed by data centres, and provide these figures at the level of individual facilities.</p><p>Aggregated data-centre metrics can offer a rough starting point (De Vries-Gao 2026), but fall far short of what is needed.</p><p>Next steps:</p><ul><li><strong>Who should be in scope?</strong></li><li><strong>How should this fit with existing disclosure regimes?</strong></li><li><strong>How should this fit with carbon emissions reporting?</strong></li><li><strong>What level of detail is proportionate?</strong></li><li><strong>What cadence is appropriate?</strong></li><li><strong>How should responsibility be allocated across complex supply chains?</strong></li><li><strong>What safeguards might be needed for confidentiality and security?</strong></li><li><strong>What’s the direction of travel? How should compliance be monitored and updated over time?</strong></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a6f3643a607a" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>