Lydia H. Liu, "Toward AI Socialism": A Summary

Originally published in Chinese in Wenhua Zongheng (Beijing Cultural Review), October 2026. 

A personal starting point. Lydia H. Liu, a professor at Columbia University, opens with something she saw firsthand. In 2004, while she was teaching at the University of Michigan, Google signed a deal to digitize the university library's seven million volumes under the slogan "democratizing knowledge." The terms were kept secret from faculty. Years later, a similar contract with the British Library became public. It showed that libraries were barred from using their own digital copies commercially, while Google kept wide commercial freedom. Two decades on, Liu argues, those scanned books have become training data for Silicon Valley's commercial AI models.

A digital enclosure movement. Liu compares this to the English enclosures, which fenced off common land for private profit. Today the "commons" is humanity's accumulated knowledge: libraries, the internet, open-source code, academic research. The "fences" are intellectual-property and trade-secret law. She traces how the courts cleared the way. In Authors Guild v. Google, mass scanning was ruled "fair use," a decision that became final in 2016. In Bartz v. Anthropic, training on books was also held to be fair use, but downloading pirated copies was not; the case ended in a $1.5 billion settlement in 2025. By treating "acquiring" data and "using" it as separate legal questions, Liu argues, the courts let AI companies turn public knowledge into private model weights, which they then sell back to the public through paid APIs.

Hidden labor. Behind the "intelligence" of AI, Liu describes a global chain of invisible work:

• crowdworkers on Amazon's Mechanical Turk, who labeled the ImageNet dataset;

• low-paid data workers in Kenya and elsewhere in the Global South, who clean data, label images and moderate traumatic content;

• the human ratings that train models through reinforcement learning from human feedback;

• the unpaid labor of ordinary users, whose posts, clicks and corrections become free training material.

Intellectual monopoly capitalism. Liu traces today's global IP regime to a 1986 coalition of thirteen multinational corporations, including IBM and Pfizer. That coalition drafted the blueprint that became the WTO's TRIPS agreement. Drawing on Marx's analysis of rent and Lenin's analysis of finance capital, she calls the AI giants "rentiers" who collect rent on access to knowledge. She rejects Yanis Varoufakis's thesis of "technofeudalism": what we face, she argues, is not feudalism but capitalism's latest monopoly form, fused with finance capital. Echoing the Communist Manifesto, she writes that "a specter, the specter of AI socialism," is haunting transnational capitalism. She notes how doomsday warnings about AI have tended to lift the share prices of AI companies, and proposes a new novel: Frankenstein on Wall Street.

The alternative: AI socialism. By AI socialism Liu does not mean AI-run central planning. She means opening AI's means of production (data, models, algorithms, computing power) to the many instead of financial oligarchs. Her program includes:

• public AI infrastructure, treated like highways or power grids;

• data trusts and cooperatives that let data producers share in the gains;

• a global charter protecting the knowledge commons;

• open-source models, shared computing power and jointly built training data.

She points out that the US government has funded AI research from the start, through DARPA and other agencies, and asks why its fruits should belong to private firms. She also cites Wikipedia, OpenStreetMap, GenBank and open-access publishing as working commons, and Chinese open-source models such as DeepSeek as evidence that open development can compete.

The obstacles. Liu names three. Computing power is concentrated in a few hands and tied up with chip export controls. Data is unequal: AI is dominated by "global English," while thousands of languages are data-poor. And geopolitical rivalry turns open-versus-closed models into a contest between nations. In the end, she concludes, the choice depends not on technology but on political will: "The digital knowledge commons belongs to all humanity; AI socialism is no utopian dream."