Hypertrophic AI – can libraries lead the pushback?
Abstract
A keynote given at Focus on Open Science, Budapest, November 6th, 2025 Abstract: "The ongoing hype surrounding the newest generation of AI methods -- notably generative AI -- creates enormous pressure both for individuals and institutions not to get left behind in terms of technology and competence. This makes it easier for commercial tech companies to push AI-based solutions into their services, sometimes even before they have reached a certain maturity and not always in the best interest of their users. Both the training processes and productive operations of many of those solutions are using up a staggering amount of resources and creating or aggravating all kinds of risks for various minorities, the climate, and the information landscape. The full extent of the impact is a moving target and therefore hard to gauge but even the most concrete and imminent implications have barely entered the awareness of the wider public to date. In this talk we will survey a range of aspects how the use of large AI (LAI), especially proprietary LAI, could have and indeed already has a critical influence on society and its environment. We will then ask ourselves if there are alternatives to this massive commercial push of LAI applications that follow the principles of sustainability, transparency, and openness for the benefit of everyone. Libraries as publicly funded, non-profit information infrastructure institutions are uniquely qualified to promote best practices towards a more "civilized" use of AI and also to act as reliable data providers or brokers for research and productive applications. However, in order to fulfil that role, libraries have to manage the balancing act of boldly and unapologetically representing those values on the one hand and navigating the necessities of a world shaped by big profit-oriented players on the other. This is a long-term mission which will require initiative, tenacity, and strategic networking on a global scale."
Full text
The ZBW is a member of the Leibniz Association. hypertrophic AI – can libraries lead the pushback? Dr. Argie Kasprzik ZBW – Leibniz Information Centre for Economics Focus on Open Science, Budapest, November 6th, 2025 hypertrophic (adj.): abnormally enlarged or overgrown
LAI (large-scale AI) comes at a price. Always. We often do not realize that we are paying it. Also, the biggest price is not paid by us but by others. Do we really want that? page 2
costs for humanity: LAI accelerates climate change page 3 water air electronic waste pictures: https://icon-icons.com/de/symbol/Stecker/39479 ; https://ccnull.de/foto/dry-white-long-rice-inwooden-spoon/1038560 ; https://www.pexels.com/de-de/foto/industrie-nebel-smog-pflanze-7001364/ ; https://commons.wkimedia.org/wiki/File:Elektroschrott.jpg ; https://ccnull.de/foto/computerchip/1006620 e lectricity both the training and the running of LAI uses up an amount of resources which often exceeds that of small states, and the trend goes upwards
training GPT-4: 50 GWh – 200 flights NY-SFO – 50x more than GPT-3 training GPT-3: 5,4 million liters of water; LLaMa-3: 22 million liters – as much as needed for farming 2 tons of rice prognosis 2030: LAI will make up ~21% of global electricity use and 4x to 6x the water consumption of Denmark 2030, example USA: 4x more deaths than 2023 due to LAI-related air pollution; 60.000 more cases of asthma (predominantly in poorer communities) 2030: up to 2,5 million tons of electronic waste more ~ one smartphone per human costs for humanity: some examples for resource use page 4 all references see link list at the end screenshot warning from: https://ai-impact-risk.com/ai_energy_water_impact.html
page 5 source: IEA "Energy and AI" report https://www.iea.org/reports/energy-and-ai
costs for humanity: exploitation page 6 screenshot from: https://data-workers.org/ •Large AI continues to depend substantially on humans sorting, cleaning, and annotating data •this kind of work is often oursourced by big tech to less privileged countries (e.g. Kenya, Argentina), under exploitative conditions – no safety at work, hourly wages of less than 2 USD •many people doing that kind of work are suffering from health problems and post-traumatic stress disorder (PTSD)
knowledge is power – asymmetry gets worse •paid-subscription ("premium") plans → access to (allegedly) high-quality information and tools becomes a luxury •if you do not have the financial means you pay with your data / privacy •big tech build monopolies and act as gatekeepers – control who gets access and exploit downstream actors •big tech obtain a potential influence that they should not have in use cases that lie in the public interest (e.g., research, education, infrastructure) costs for humanity: gatekeeping and undue influence page 7 picture: CC-BY vectorportal.com also see von Thun, Max and Hanley, Daniel, Stopping Big Tech from Becoming Big AI: A Roadmap for Using Competition Policy to Keep Artificial Intelligence Open for All. Open Markets Institute 2024. Online at http://dx.doi.org/10.2139/ssrn.4990780
often unsollicited, invasive use of data without respect for personal privacy, copyright, terms of licences, provisions to exclude AI-related traffic •example: experiments by OpenAI / researchers of University of Zürich in online community (Reddit) •example: strain on Open Access infrastructure by crawling / bots costs for humanity: privacy, infrastructure – two examples page 8 all references see link list at the end
costs for humanity: change management page 9 screenshot from: https://mapstodon.space/@abel/114975709443152718 •reports from the community suggest that more and more managers actually enforce the use of AI tools •potentially negative impact of misguided change management on productivity (!), psyche, and motivation of staff ↓ → screenshot from: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
What can I do as an individual? page 16
•less is more – is this particular use necessary? do I even need an AI-based tool for that? •deliberate use of AI-based applications – using commercial tools sends a signal to tech firms that further inflating their services is worth it •educate and raise awareness in others – including your leaders! •look out for your own rights – do I really want to •use built-in AI tools? if not, switch them off / opt out wherever possible! •relinquish my input (written, audio, video) to an AI-based tool e.g. in video calls, for a summary? if not, object / opt out! What can I do as an individual? push back against hypertrophic AI! page 17 picture: https://openclipart.org/detail/327430/stop-sign-hand-print-silhouette
•raise awareness, educate (and act accordingly) •mindful change management, use human expertise strategically •activism! •push towards openness (Open Data, Open Source) •push towards improvement of legal situation context: we live in a world in which •people (have to) generate content in order to make a living •enterprises curate information and develop services in order to make a profit and not for the benefit of all / the public good ; point of view: content should be open and free to use for all – including as training data – but ONLY for non-commercial players! ➔ fair use for purposes of education and research What can we do as institutions? page 18 picture: https://publicdomainvectors.org also see Brewster Kahle (Internet Archive) on Fair Use: https://blog.archive.org/2025/05/14/protect-fair-use-especially-now/
•cooperation! •"evaluate & create data sets" ! •reuse models / services that use as few resources as possible, are run locally if possible, and preserve privacy / data security •for centres of competence: •develop models / services that •use as few resources as possible, •can be run locally, •and in a way that preserves privacy / data security and make that process transparent for reusability •inject human expertise at strategic points •optimize own software / hardware for carbon footprint What can we do as institutions? page 19 picture: https://publicdomainvectors.org
page 21 https://cfg.eu/advanced-ai-possible-futures/
Message page 22 screenshot: https://mastodon.online/@street artutopia/115186042059851789 Thank you! contact: [email protected] what my team does with AI at ZBW: https://www.zbw.eu/en/aboutus/knowledge-organisation/automationof-subject-indexing-using-methodsfrom-artificial-intelligence link list see next four slides ↓
overview / general / energy •https://savethe.ai •https://heated.world/p/ai-is-guzzling-gas •https://www.iea.org/reports/energy-and-ai •https://ai-impact-risk.com/ai_energy_water_impact.html •in "An MIT Exploration of Generative AI": https://mit-genai.pubpub.org/pub/8ulgrckc/release/2 •MIT Technology Review: https://www.technologyreview.com/supertopic/ai-energy-package/ •IFLA AI Entry Point for Libraries (incl. risks): https://repository.ifla.org/items/f197f327-dc49-4743-bb57-0a373505da8b •link list about drawbacks of all kinds: https://fmjansen.com/posts/case-for-ai/ •https://arstechnica.com/information-technology/2025/07/ai-in-wyoming-may-soon-use-more-electricity-than-stateshuman-residents/ •https://www.linkedin.com/posts/bgamazay_openai-has-announced-o3-which-appears-to-activity-7276250095019335680sVbW water additional links •https://grist.org/technology/amazon-data-centers-water-positive-energy/ •https://www.tomshardware.com/tech-industry/artificial-intelligence/using-gpt-4-to-generate-100-words-consumes-up-to-3bottles-of-water-ai-data-centers-also-raise-power-and-water-bills-for-nearby-residents air pollution and electronic waste additional links •https://www.theguardian.com/technology/2024/sep/15/data-center-gas-emissions-tech •https://www.technologyreview.com/2025/05/20/1116272/ai-natural-gas-data-centers-energy-power-plants/ •https://www.sciencealert.com/scientists-predict-ai-to-generate-millions-of-tons-of-e-waste link list (last checked 2025-11-03)
reactions of big tech / greenwashing •https://grist.org/accountability/microsoft-employees-spent-years-fighting-the-tech-giants-oil-ties-now-theyre-speaking-out/ •https://www.theatlantic.com/technology/archive/2024/09/microsoft-ai-oil-contracts/679804/ •https://www.scientificamerican.com/article/what-do-googles-ai-answers-cost-the-environment/ annotation by humans / exploitation •Humans in the loop: Labour in the AI supply chain https://www.youtube.com/watch?v=GEuy4USHMGY •https://data-workers.org/Fasica/ •https://www.theguardian.com/media/2024/dec/18/kenya-facebook-moderators-sue-after-diagnoses-of-severe-ptsd •https://www.ihrb.org/latest/content-moderation-is-a-new-factory-floor-of-exploitation-labour-protections-must-catch-up privacy / other legal and ethical aspects •https://desfontain.es/blog/privacy-in-ai.html •https://static1.squarespace.com/static/5e449c8c3ef68d752f3e70dc/t/6710039559ef840f59365bc8/1729102742546/Stopp ing+Big+Tech+from+Becoming+Big+AI.pdf •https://opentools.ai/news/openai-turns-rchangemyview-into-ais-persuasion-playground •https://retractionwatch.com/2025/04/29/ethics-committee-ai-llm-reddit-changemyview-university-zurich/ •https://arstechnica.com/ai/2025/03/devs-say-ai-crawlers-dominate-traffic-forcing-blocks-on-entire-countries/ •https://coar-repositories.org/news-updates/open-repositories-are-being-profoundly-impacted-by-ai-bots-and-othercrawlers-results-of-a-coar-survey/ link list (last checked 2025-11-03)
degradation of information landscape and deskilling •https://en.wikipedia.org/wiki/AI_slop ; https://en.wikipedia.org/wiki/Enshittification •https://edition.cnn.com/2025/01/13/business/enshittification-internet-meta-nightcap/index.html •"Self-Consuming Generative Models Go MAD": https://arxiv.org/pdf/2307.01850 ; https://news.rice.edu/news/2024/breaking-mad-generative-ai-could-break-internet •https://link.springer.com/article/10.1007/s00146-025-02199-9 •https://www.forbes.com/sites/bernardmarr/2023/05/16/the-danger-of-ai-content-farms/ •https://www.wired.com/story/linkedin-ai-generated-influencers/ •two studies about "cognitive debt": https://arxiv.org/abs/2506.08872 ; https://doi.org/10.1016/S2468-1253(25)00133-5 •https://www.arthurperret.fr/blog/2024-11-14-student-guide-not-writing-with-chatgpt.html •https://open.devinci.fr/ressource/etude-2024-impact-ia-generatives-etudiants/ •https://advait.org/files/lee_2025_ai_critical_thinking_survey.pdf •https://www.theverge.com/openai/686748/chatgpt-linguistic-impact-common-word-usage •https://arstechnica.com/ai/2025/04/researchers-find-ai-is-pretty-bad-at-debugging-but-theyre-working-on-it/ •https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study •"Google's AI is Destroying Search, the Internet, and Your Brain" https://archive.ph/tNSO4 link list (last checked 2025-11-03)
alternative approaches •https://hidde.blog/ethical-ai/ •Suominen, Osma. Building Civilized AI. AI Sauna, 6 May 2024. https://www.youtube.com/watch?v=oT8FP1JH5vE&t=1322s ; https://docs.google.com/presentation/d/e/2PACX-1vRA1o11pODoJ0FmFc8dRj-xNZRUs7lsxzDACkiYt6dBdfql1ujw3gGpSedTQnXDG0MrRg3_WAl1GQS/pub •https://greenscreen.network/en/blog/within-bounds-limiting-ai-environmental-impact/ •https://www.thegreenwebfoundation.org/publications/report-ai-environmental-impact/ •https://www.sciencedirect.com/science/article/pii/S2666498424001340 •https://medium.com/@nagidravid/small-language-models-vs-large-language-models-understanding-the-differences-andimplementations-fc91ff208541 •https://tenbluelinks.org/ openness / fair use •https://opensource.org/ai/open-source-ai-definition •https://www.forkable.io/p/metas-new-llama-4-ai-models-arent#%C2%A7openness-is-binary •https://shujisado.org/2025/01/27/why-is-the-llama-license-not-open-source/ •https://shujisado.org/2025/01/27/significant-risks-in-using-ai-models-governed-by-the-llama-license/ •https://blog.archive.org/2025/05/14/protect-fair-use-especially-now/ •https://osai-index.eu/guides/open-llms-education •https://www.swiss-ai.org/apertus general (digital carbon footprint, independence from corporate providers) •https://www.digitalcleanupday.org/ •https://codecarbon.io/ link list (last checked 2025-11-03)