1. generative AI Energy consumption Planetary scale computing More than 1% of global greenhouse gas emissions 1 to 1.3% of the world’s electricity Big Data platforms are increasingly becoming partners in of oil and gas resources Global warming Large amounts of radioactive waste Fossil fuels (+80%) Mostly powered by coal, natural gas or oil Calculation power Green colonialism: Land use conflicts with communities 8,000 datacenters worldwide predicted to grow a continuing 10% per year (to maintain cheap data training guarantee) Power consumption A single data center can consume the equivalent electricity of 50,000 homes Renewable energy Cooling systems (CRACs/CRAHs) Greater than 40 percent of data centers electricity usage Cobalt is used in rechargeable lithium-ion batteries. There are reports of the use of child labour and human rights abuses in the Congolese cobalt mines Water Water towers Mercury, copper, lead and arsenic leach into soil and waterways, cau- sing a build-up of harmful chemicals in the ecosystem and its food chains 4-9 litres per kWh of server power Google says that in 2022, its data centres around the world will have used 5.2 billion gallons of water One year’s worth of global e-waste: 7,3 kg per capita 82,6% landfilled or informally recycled Nvidia: Proprietary CUDA system for AI Intel AMD Supply chain of raw materials More than 50,000 tonnes of e-waste, which takes millennia to decompose, is sent to Africa every year. It ends up in informal dumps like those in Agbogbloshie, Ghana, where hazardous materials are burned, exposing people to toxic fumes and radioactive elements Droughts and local conflicts In the year 2023, residents of Montevideo, suffering from a severe water shortage, protest against plans to build a Google data centre Local water table affections Heat Water cooling As the extra power consumption generates more heat, AI data centers also need alternative cooling methods, such as liquid cooling systems ↑ local electrical prices (Taiwan Semiconductor Manufacturing Company) Google TPU AWS Trainium Waste Cables, batteries, uninterruptible power supplies (UPS), air conditioners (CRACs and CRAHs), power distribution units (PDUs), and transformers are also periodically decommissioned and disposed of, when warranties expire Servers Plate Heat exchanger Training data Strategic importance: Increasing fears of a Chinese invasion of Taiwan (US-China technology battle) Azure Cloud AWS Big Data platforms Graphics processing units (GPU) Microsoft Google Deep sea mining: Impacts: destruction of ecosistems, compaction of the sea floor, and creation of sediment plumes that disrupt aquatic life Mining industries Metals and minerals extraction exported to the Global South Salt flat Central depression Produces over 60% of the world’s chips and over 90% of the world’s most advanced chips 20 24 Digital Colonialism Data centers Nuclear fission Oklo (Sam Altman) TSMC Indigenous and local communities: 300,000 million tonnes of carbon are stored on land managed by indigenous communities worldwide (due to lower rates of deforestation). These communities are often excluded from land-use decisions. In Kenya, thousands of people were evicted from their ancestral lands as part of a programme to reduce deforestation Carbon offsetting economy Companies offset their greenhouse gas emissions by paying other parties to prevent emissions or remove carbon dioxide from the air. Offsetting is done through questionable forestry and wind projects that are more symbolic than real Greenwashing strategies 3 Lithium Atacama desert Andes Mountains MMG Cusco 29 Manufacturers use copper in the most powerful computer chips because of its superior electrical conductivity. The 'mining corridor' of southern Peru (exploited by China's MMG Ltd, Switzerland's Glencore and Canada's Hubbay) is one of the mainstays of the economy, but it is also one of the main sources of conflict due to unequal distribution of income from mining and health problems due to water pollution Copper Hudbay Puno Glencore Noise pollution Data center’s hum causes acute and longitudinal physiological effects 63.546 Cu Apurímac Amazon Pump station 6.9410 Li The Salar de Atacama, a salt flat nearly four times the size of New York City, is one of the world’s largest lithium mines, a key component in battery production. Rising demand is increasingly affecting local communities, threatening their access to water and impoverishing biodiversity Arequipa Copper 'mining corridor' (southern Peru) 79 196.97 Au Brazilian law does not officially allow mining on indigenous land. But illegal gold mining in the Brazilian Amazon has skyrocketed since 2019. Researchers have documented tens of thousands of small-scale miners and more than 320 illegal mines, with the actual number likely to be much higher. Small-scale gold mining has led to widespread deforestation, mercury pollution and sedimentation of vital rivers Gold AI data centers need to be built with additional infrastructure that can deliver much more power Silicon Valley venture capital Research and map design — @tallerestampa AI start-ups I enA Op Wave of tech speculation ind epM De Current data centers 5 to 10 kilowatts per rack AI requirements 60 or more kilowatts per rack Software as a service (SaaS) and platform as a service (PaaS) Application programming interfaces (APIs) Race to deploy new AI Diesel generators Air pollution and climate impacts 50 118.71 Sn Weld China, Indonesia, Burma, Peru Tin 3 6.9410 Li Battery Lithium Triangle (Bolivia, Chile, Argentina), Australia, China Lithium 73 opic thr An 74 180.95 Ta Circuits Tantalum Human labour 79 Working markets: Kenya (36% below the poverty international line), India (68%), Philippines (18%), Venezuela, Colombia (39%), Uganda (28%) Copper China 19 26.982 50 49 118.71 Filtering • content moderation • banned prompts • “toxicity” detection Deep neural networks AI ecologies Institutes for future Criticism through arts Alignment Critical expertise Since the year 2007, over seven billion smartphones have since been manufactured 64% of e-waste sent to recycling centres in Europe is then shipped to Africa Public Regulations 60 162.5 63 162.5 64 60 162.5 EU AI Act Transparency requirements in datasets US companies voluntary agreement Massive and private latent spaces 144.24 59 140.91 Pr Neodymium Praseodymium Academic research Statistical text Data Sciences Statistical images Chat GPT Generative AI engines Bard Personalized entertainment Synthesia Adobe Firefly Dall·e Information systems News Recommendation Content categorisation Algorithmic governance Midjourney Musicians Stable Diffusion blogs images comments videos posts illustrations reviews paintings papers brands shares styles Product designer Voice artist Film editors Screenwriter Technological determinism Services dependence In August 2023: 34 million images/day UX designer Texture artist scores Storyboard artist Writer Web developers engineering equipment technical expertise 8 12k 2k Posts Followers Following Image recognition (labeling, segmentation, pose, depth) Web2 Weekly Monthly 67 52 467 46 7 Feb Apr Jun -174 54 +16 53 +72 -247 51 -369 190 -67 550 +114 555 +209 416 -47 +376 Profiling and content filters Social media Descriptors to image and texts < Aesthetic Turn Rise of the attention economy and its technologies Motion graphics Set designer Illustrators Web designer Data extractivism Photographers Stylist Rotoscopers AI Large scale models that run in private infrastructures 2021 OpenAI introduces the neural network called CLIP, that classifies images using textual descriptions 2021 Latent Diffusion Process of diffusion within a latent space 2019 GPT-2 (Generative pre-trained transformer) 2016 Generating images (32 x 32 píxels) from captions Elman Mansimov et al. Global computation Actor Render wrangler Their lifespans average less than two years, a consequence of designed obsolescence Gaming Copywriter Freelancers jobs Digital gadgets Internet service providers 83 Graphic designer Continued hype voices labor Digital internet content Image, text & sound Social media Silicon Valley imaginaries photography likes labels Fashion Architect music 08:22 Compositor Mystification TSMC Taiwan Semiconductor Manufacturing Company controls a large share of the global market for advanced chips and has a strong position in the manufacture of leading-edge semiconductors sounds books Github Copilot Bing Claude Disinformation and propaganda Image banks Aiva Terbium Assembly Devices are assembled in factories like Foxconn located in the Global South, where labor can be procured cheaply and legal protections for workers are scant Project Gutenberg Replica 158.93 Tb 420 submarine cables distributed following the routes of telegraphs and colonial navigations Techniques for automatically downloading and processing web content. Common Crawl and OpenWebText2: scientific research institutions for data mining in non-commercial contexts Wikipedia Statistical sound 65 Advanced chips Web scraping Extraction and privatisation “The right to read is the right to mine”: appropriation of the discourse “Information Wants to Be Free” mid- 90’s version of Internet Freedom • Extreme racial and gender stereotypes, ideologies, viewpoints • Narrative archetypes • Mainstream culture • Styles as statistical patternsthat can be replicated automatically The manufacture of server racks involves copper, silicon and lithium. Cabling, busbars, switchgear and so on all use large amounts of copper Terbium 157.25 Infringement of intellectual property laws Authoritarian populisms Political polarisation 158.93 Tb Gadolinium Nd Dysprosium China 65 Gd Europium Dy 140.91 Praseodymium 151.96 Eu Dy 66 Microphone 59 Pr Neodymium Dysprosium 66 Screen 144.24 Nd Dy ASML Dutch manufacturer of equipment for the production of integrated circuits. It is the world’s largest supplier of photolithography systems for the semiconductor industry, with more than 5/6 of the world market Encoding worldviews Synthetic contents Regimes of truth: public discourses and affectations Outsourced workers Unwaged AI trainers Control of communication through images > Image to descriptors Terms and conditions Research → AI and human extintion 66 Vibration Dysprosium Largest possible dataset Refugee camps (Kenya and Lebanon’s Shatila) AI Photography Search engines AI assistants Word processors Customer service China, Indonesia, Burma, Peru Gallium Lack of transparency AI ethics AI Panic 69.723 Ga Lobbies AI and human extintion Demands for self-regulation Big Tech innovation 31 Indium The information in datasets is always embedded in the contexts that create and use them. They remain tied to the communities, individuals, organisms and environments in which they were created Deepen AI, trains Syrian refugees to annotate data for the likes of Google and Amazon. Sama trains refugees in Uganda, Kenya, and India to complete short data tasks (RLHF) 114.82 In Rare-earth elements Datasets Policy agenda Canada, Russia, Belarus Potassium Sn LCD 39.098 K Al Aluminium CommonPool 12.8 billion image-text pairs Discourses on AI Congo, Zambia, Chile, Peru, Mexico, China, Brasil Silver Silicon 13 Screen LAION-5B 5.8 billion image-text pairs “Visionary” leader covered by the media 107.87 Ag 28.086 Tin Fine-tuning • data scoring • annotating data • labeling 47 Si Processor Companies: Sama, Appen, Clickworker, Scale AI AI Model 63.546 Cu Gold 14 Rwanda, Brazil, Congo, China Tungsten 29 196.97 Au Conductors 183.84 W User AI Open source or publicly available paper-based Local computation 2003 First feed-forward neural network language model Yoshua Bengio et al. Caltech 101 dataset 10,000 images Fei-Fei Li et al. 2009 ImageNet dataset 14 million images and more than 20,000 categories Fei-Fei Li et al. 2012 Alexnet. First fast GPU-implementation of a CNN to win an image recognition contest Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton 2014 Generative Adversarial Network (GAN) Ian Goodfellow et al. 2015 Diffusion model Used for image generation by starting with an image composed of random noise for the network to iteratively denoise