Mapping Time with Kate Crawford

概览

Kate Crawford frames artificial intelligence through a 500-year history of technology, power, extraction, and representation. She argues that AI is not an immaterial intelligence but a deeply material industry built from data, minerals, energy, water, labor, and infrastructures of control.

The talk moves from Renaissance perspective to generative AI, suggesting that representational power has shifted from God, to the human observer, and now toward machines. Crawford connects this shift to “AI slop,” synthetic media, model collapse, and the political economy of computational capitalism.

A central conclusion is that AI repeats older imperial patterns: extraction from peripheries, enclosure of commons, concentration of power, and externalization of costs. In the Q&A, she argues for public-interest AI, consent-based data, renewable energy, democratic oversight, and more precise use of AI where it is genuinely effective.

分段落总结

[00:00] Introduction and Long Now framing

[事实] Host Rebecca Lendl introduces Kate Crawford as a leading AI scholar whose work maps the genealogy of technology and power over the last 500 years. [事实] The introduction highlights Crawford’s claim that power has moved from a God-centered cosmology, to a human-centered era, and now toward a machine-centered era. [事实] Lendl previews AI as a material and metabolic technology competing with humans for land, water, and other planetary resources.

[02:31] Looking backward to understand AI’s present

[事实] Crawford says the talk will use a longer historical horizon than public talks usually allow. [事实] She presents the goal as looking backward to understand the present and possible futures of AI. [事实] She names topics including Renaissance painting, steam engines, bird guano, and AI slop.

[03:36] Linear perspective as a technology of truth

[事实] Crawford describes Brunelleschi’s 15th-century optical demonstration in Florence as an early technology of artificial or linear perspective. [事实] She argues that linear perspective became accepted as “the way things look,” despite being a constructed technique. [事实] She says perspective helped make the world measurable, controllable, and commodifiable.

[06:07] From God, to the human eye, to the machine

[事实] Crawford says medieval cosmology placed God at the center, while artificial perspective made the individual human eye the center. [事实] She argues that AI shifts representational power again, toward neural networks trained on billions of scraped images and texts. [推测] The historical comparison suggests that AI may change not only media production but also what societies accept as reality.

[07:01] Shrimp Jesus and the AI slop aesthetic

[事实] Crawford identifies “Shrimp Jesus” as a widely shared AI-generated image and an example of AI slop. [事实] She describes AI slop as hyperreal, uncanny, and shaped by computational capitalism. [事实] She argues that slop is produced by turning human culture into data and metabolizing it through energy-intensive infrastructure.

[08:04] AI as a metabolic technology

[事实] Crawford compares generative AI to a Krebs cycle: models ingest images, videos, and texts, process them through large neural networks, produce synthetic outputs and carbon dioxide, and then ingest those outputs again. [事实] She says AI is competing with humans for basic resources such as land and water. [推测] The metabolic framing positions AI’s environmental costs as central to the technology, not as an external side issue.

[09:18] Mapping empires across space and time

[事实] Crawford says her talk has six chapters covering historical empires, metabolic machines, AI data, planetary resources, waste, model collapse, and possible futures. [事实] She describes earlier work mapping the full life cycle of an Amazon Echo device, including mining, manufacturing, data processing, and e-waste. [事实] She introduces “Calculating Empires” as a 24-meter-long timeline mapping technologies of communication, computation, classification, and control since 1500.

[12:43] Why the map begins in 1500

[事实] Crawford says the map begins around 1500 because capitalism, European colonization, and the printing press began articulating together. [事实] She argues that these systems created structures of power that still shape the world. [事实] She uses Fernand Braudel’s idea of the longue durée to argue for looking beyond events to slower historical forces.

[15:38] Soil depletion, guano, and metabolic rift

[事实] Crawford recounts Justus von Liebig’s discovery that European agriculture was depleting soil nutrients faster than they could be replaced. [事实] She explains that Europe turned to Peruvian guano as fertilizer, leading to geopolitical conflict over bird excrement. [事实] She says Marx saw this as a “metabolic rift,” a disruption of ecological cycles by industrial capitalism.

[18:26] Gutta-percha and the extraction behind communication networks

[事实] Crawford describes how gutta-percha latex from Malaysian trees became essential for insulating undersea telegraph cables. [事实] She says the first transatlantic cable required the equivalent of about 900,000 tree trunks. [事实] She argues that technological booms can drive extraction beyond what ecosystems can sustain.

[20:54] The data appetite of AI

[事实] Crawford says AI companies have scraped enormous amounts of data from the open web and are approaching the limits of what can be collected there. [事实] She traces the statistical turn in AI to IBM’s speech recognition lab in the 1970s, where researchers shifted from rule-based understanding to prediction from large corpora. [事实] She cites Robert Mercer’s phrase: “There’s no data like more data.”

[23:42] Commercially biased datasets and world models

[事实] Crawford says modern AI datasets overrepresent commercial sites such as stock image platforms, Shopify, and Pinterest. [事实] She argues that training on these datasets means seeing through the eyes of online commerce. [事实] She says newer “world model” approaches require surveillance-scale data from cameras, sensors, bodies, gestures, and physical spaces. [推测] Crawford treats this as a deeper enclosure of life itself, extending extraction beyond online expression.

[26:22] Lithium, minerals, and deep-time extraction

[事实] Crawford discusses lithium mining in Chile’s Salar de Atacama and says local communities described falling water tables, collapsing ecosystems, and loss of land. [事实] She says AI infrastructure depends on critical minerals including rare earths, cobalt, copper, and lithium. [事实] She frames this as another metabolic rift: minerals formed over billions of years are used in AI chips that may last only one or two years.

[28:26] Energy, data centers, and environmental racism

[事实] Crawford says AI prompts use far more electricity than typical web searches, and image or video generation uses still more. [事实] She cites estimates that data center energy demand could rise sharply as a share of U.S. electricity use by 2030. [事实] She uses Elon Musk’s xAI data center in South Memphis as an example of local grid strain, methane generators, and pollution affecting predominantly Black and brown communities.

[30:34] Water use and Jevons paradox

[事实] Crawford says AI data centers require large amounts of fresh water to cool chips. [事实] She cites a recent Nature study estimating future data center water demand at 5.2 billion cubic meters. [事实] She explains Jevons paradox: making a resource use more efficient can increase total consumption if it expands usage. [推测] Efficiency improvements alone are unlikely to solve AI’s resource problem if AI is embedded into more platforms, schools, and workplaces.

[32:46] Abstraction, extraction, and distraction

[事实] Crawford builds on Michael Hardt and Antonio Negri’s idea that the information economy abstracts material conditions while extracting resources and data. [事实] She adds a third term: distraction. [事实] She argues that AI models target attention by tracking interests, pleasures, writing styles, communication, and thought.

[33:53] Slop, slopaganda, and the content economy

[事实] Crawford defines slop as low-effort AI output, including mass-produced videos and synthetic content designed to farm engagement. [事实] She identifies satirical slop, commercial slop, and political “slopaganda.” [事实] She argues that AI-generated content is overtaking human work online and cannibalizing the human content economy. [推测] The slop economy may reward virality over quality, truth, or human authorship.

[37:18] Model collapse and AI eating itself

[事实] Crawford says models trained repeatedly on their own outputs can degenerate into noise. [事实] She references “model autophagy disorder,” described as AI self-consuming, and connects it to model collapse. [事实] She says self-training can reduce diversity, flatten outputs, and erase outliers, minority patterns, and edge cases. [推测] Model collapse could affect not only image aesthetics but also high-stakes systems such as medical models.

[39:53] Possible futures for AI

[事实] Crawford says the current approach to AI may not sustain itself indefinitely because of capital investment, energy demand, and infrastructure strain. [事实] She presents three possible futures: cascading collapse, radical regeneration, or imperfect coexistence. [事实] She says the most likely path may be an imperfect coexistence that abandons fantasies of quick technical fixes.

[42:33] Alternative computing does not automatically decentralize power

[事实] Crawford surveys hydraulic computing, water integrators, biochemical computing, DNA computing, plant DNA storage, and quantum computing. [事实] She says quantum computing may someday improve efficiency but currently requires energy-intensive cryogenic and vacuum infrastructure. [事实] She warns that alternative computing paradigms do not inherently challenge Silicon Valley’s concentration of power.

[44:47] Local resistance and democratic agency

[事实] Crawford says federal AI transparency and accountability are not meaningfully happening in 2025. [事实] She identifies local opposition to hyperscale data centers as an emerging source of resistance. [事实] She says 16 hyperscale data center projects in the United States have been delayed or rejected through bipartisan opposition. [推测] Crawford sees infrastructure politics as a practical site for democratic intervention.

[45:51] Carl Sagan’s warning and Crawford’s conclusion

[事实] Crawford plays a Carl Sagan interview warning that societies based on science and technology become dangerous when the public does not understand them. [事实] Sagan emphasizes skepticism, education, and questioning authority. [事实] Crawford concludes that AI must be mapped, understood, and shaped collectively to avoid environmental devastation and concentrated AI empires.

[48:17] Public AI and technological sovereignty

[事实] In Q&A, Kevin Kelly asks whether a publicly accessible, publicly funded, and publicly accountable AI could exist. [事实] Crawford says public AI and tech sovereignty are important ideas but warns that competing with the largest AI companies requires staggering money and resources. [事实] She says building more resource-intensive LLMs could accentuate the problem rather than solve it.

[50:37] Sovereign AI, Europe, and sustainability

[事实] Kelly mentions sovereign AI programs outside the United States. [事实] Crawford discusses the European “Euro stack” as a public-interest response to American AI. [事实] She warns that if every country builds its own full AI stack, the planet could be burned at a superlinear rate. [推测] Crawford favors combining public-interest AI with strict sustainability requirements rather than simple national replication.

[51:50] AI, capitalism, and tech feudalism

[事实] Kelly asks whether much AI criticism is really criticism of capitalism. [事实] Crawford says AI is possible because of a hyper-accelerated, lightly regulated form of capitalism. [事实] She suggests current dynamics may resemble tech feudalism, with companies functioning above nation-states as power states.

[53:20] What a better capitalism would require

[事实] Crawford references local democratic socialist proposals in New York, such as free buses and childcare. [事实] She says Australia had free healthcare and free education when she was a child while still operating within capitalism. [事实] She argues for an economic system centered on human flourishing and ecological support.

[54:43] Conditions for AI to work broadly

[事实] Kelly asks what would need to change in 2025 for AI to benefit most people over the next 25 to 50 years. [事实] Crawford says one must ask “works for whom” and consider minorities harmed by supposedly average benefits. [事实] She says AI would need renewable energy, consent-based data frameworks, and use in domains where it is actually effective, such as translation, astronomy, and climate research.

[57:26] Good uses of AI and data gaps

[事实] Crawford says AI could help analyze gaps in data: who is missing, whose perspectives are lost, and what forms of bias appear. [事实] She says these gaps can reveal deeper histories of uneven resource distribution. [事实] She resists listing many applications because trillion-dollar industries already market AI as useful for everything.

[58:46] Art, AI, and the future of creative labor

[事实] Kelly asks whether art might help show a better path forward. [事实] Crawford says art can reveal different stories and mirror the world back to us. [事实] She warns that AI is already affecting design, illustration, and art jobs, just as platforms like Spotify damaged the music economy. [推测] The future role of art depends partly on whether artists can still be materially supported.

[60:30] Crawford’s own artistic practice

[事实] Crawford says “Calculating Empires” is part of her artistic practice and was released in 2023. [事实] She describes her art as research-based critical cartography. [事实] She says her work sits between art, research, design, and public mapping.

[61:46] Research methods

[事实] Crawford describes herself as interdisciplinary and influenced by social scientific methodology. [事实] She emphasizes interviewing people, visiting sites, and observing how systems work in the world. [事实] She says ethnographic work helps counter screen-based fragmentation and gives a deeper understanding of technological systems.

[63:02] What visualization revealed

[事实] Crawford says people returned to the “Calculating Empires” installation like a library, following different centuries or themes. [事实] She says the installation includes handmade books where visitors can write in their own histories and views. [事实] She says this feedback loop taught her that public engagement can shape future work.

[64:26] Recurring patterns in the longue durée

[事实] Crawford identifies automation, militarization, and enclosure as recurring patterns around major technologies. [事实] She gives long-range military shipping and European colonization as an example of technology expanding imperial reach. [事实] She says empires may look solid but always fall.

[65:55] Crawford’s personal use of AI

[事实] Crawford says she and her 13-year-old son red-team new AI models. [事实] She gives examples of checking whether image models reproduce stereotypes, such as doctors as white men and criminals as dark-skinned people in hoodies. [事实] She says the key question is what kind of world AI allows users to see and how that differs from what they know.

[68:01] Can slop and model collapse self-correct?

[事实] Kelly asks whether model collapse might be self-correcting because companies investing billions would be sensitive to it. [事实] Crawford says slop and slopaganda are not temporary but part of an emerging visual and political language. [事实] She says AI labs are aware of model collapse and are using synthetic data to address peak data, but hallucinations may replicate through training. [事实] She invokes Paul Virilio’s idea that every invention also invents its negative.

[71:04] Doomer scenarios and present harms

[事实] Kelly notes that Crawford did not focus on AI doomer scenarios in which AI takes over and kills humanity. [事实] Crawford says she used to be very skeptical of such hypotheses and still worries about shifting focus away from real and present harms. [事实] She says the collapse of federal attention to AI safety and harms has made people concerned about any AI harm more aligned. [推测] Her position is not that existential risks are impossible, but that present harms and lack of brakes are already urgent.

[72:50] What comes next

[事实] Crawford says she has never been more convinced that this is the moment for democratic conversations about AI. [事实] She says she is working on more research, another book, and continued public discussion. [事实] She wants people’s voices to be heard while short-, medium-, and long-term consequences are considered.

播客点评/总结

[推测] This episode is valuable because it refuses to treat AI as only software, intelligence, or productivity tooling. Its strongest contribution is the historical and material frame: AI appears as part of a long pattern of empire, extraction, enclosure, and representational power.

[推测] The highlight is Crawford’s ability to connect vivid examples, from Brunelleschi’s mirror and Peruvian guano to lithium mining, data centers, AI slop, and model collapse. The talk gives listeners a broad conceptual map rather than a narrow technical update.

[推测] The limitation is that many claims are presented at a sweeping scale, so listeners wanting detailed empirical debate on specific forecasts, energy estimates, or technical mitigation strategies may need additional sources. The transcript itself is a public talk and Q&A, not a technical paper.

[推测] This episode is best suited for listeners interested in AI ethics, infrastructure, environmental politics, media studies, history of technology, and long-term thinking. It is less suited for listeners looking for hands-on AI implementation advice or product-level guidance.