A letter in the following week’s FT comments on this article:

Time to dust off a Dutch com­mun­ist blue­print

Mar­tin Sandbu is nearly right when he asks “Could AI revive the social­ist dream?” (Life & Arts, August 29). AI could be empower­ing, but the real cen­tury-old debate isn’t about free mar­kets versus state con­trol, which turned out to be two sides of one bent coin. The choice is between a more-or-less state-reg­u­lated cap­it­al­ism, and repla­cing com­mod­ity pro­duc­tion entirely with a sys­tem built on labour time account­ing.

Defeated by counter-revolu­tion in Ger­many, but also strongly opposed to Rus­sian-style state social­ism, the GIK (Groep van Inter­na­tionale Com­mun­isten) in the Neth­er­lands in the 1930s pub­lished what it called “Fun­da­mental Prin­ciples of Com­mun­ist Pro­duc­tion and Dis­tri­bu­tion”, a hard-nosed mani­festo for an eco­nomy that would not only reduce neces­sary labour time while meet­ing peoples’ real and grow­ing needs, but also pre­vent the ree­m­er­gence of com­mod­ity value and a para­sitic social class to feed off it, whether plu­to­cratic or bur­eau­cratic.

GIK’s blue­print needs updat­ing, but AI is made for the large-scale inter­pret­a­tion of real-time data their pro­pos­als would involve. And for a future where pro­duc­tion and dis­tri­bu­tion are both glob­ally diver­si­fied and hyper­local, an updated polit­ical coun­ter­part to their eco­nomic ideas — which in the 20th cen­tury they called the work­ers’, sol­diers’ and house­wives’ coun­cils — could be the fix for our scler­otic and lethal sys­tem of nation states.

Siôn Whel­lens
Lon­don E2, UK

I asked Claude to review the GIK’s 410-page blueprint and to “Act as a political economist. Review the attached document and suggest 5 ways in which the AI capabilities available today could enable more efficient book-keeping required in order to implement the System of General Social Accounting the 1930 paper recommends.”

The response directly addresses, in very specific ways, Sandbu’s insight that AI can, indeed, revive the socialist dream–with one remaining conundrum.

Here’s the picture the document paints, and where AI genuinely changes the calculus.

What the GIK system requires

Chapters 9–10 lay out a decentralized labor-hour accounting scheme with no central planning authority: each establishment tracks its own inputs — fixed means of production (P), circulating means/materials (C), and living labor (L) — converts them into labor-hours, and “hands on” that value to the next establishment in the production chain via a registration of the stream of products, not unlike bank clearing-house settlement. A parallel ledger issues labor certificates to workers and computes the Factor of Individual Consumption. Crucially, the authors insist control must be objective — arising from public registration of the production data itself — rather than subjective, i.e., not dependent on inspectors’ discretionary judgment (they’re explicit that Varga’s Hungarian inspectorate model, which fused technical rationalization with account-book control under one authority, is “fundamentally wrong” for a communist economy). And they stress there can be “no economic secrets” — periodic public reports from the book-keeping office are what make control a non-issue.

The bottleneck the authors themselves acknowledge: even once technical staff have fixed production times, “office workers” are still needed to manually record the debit/credit movements between thousands of establishments, continuously, across an entire economy. That’s the actual bookkeeping labor problem 1930 couldn’t solve. Here’s where today’s AI capabilities could plausibly close that gap:

1. Automated real-time conversion of physical inputs into labor-hour units. The formula (p + c) + L = product requires translating heterogeneous inputs (machine-hours, materials, energy, subsidiary labor) into a single accounting unit at every establishment, continuously. Sensor/IoT data feeding into ML-based activity-costing models could do this conversion automatically at the point of production, rather than through periodic manual cost-accounting exercises — essentially industrializing what interwar “scientific cost-accountancy” did by hand.

2. A distributed ledger for inter-establishment settlement. The clearing-house mechanism the authors describe (settlement between establishments without labor-certificate “spot payment”) maps closely onto a shared, tamper-evident ledger architecture: each establishment’s outgoing product automatically debits/credits the receiving establishment’s p/c/L accounts. This removes manual reconciliation while satisfying their “no economic secrets” requirement, since the ledger is inspectable by any workers’ council, not just a central office.

3. Continuous computation of socially necessary labor time. The text’s “horizontal coordination of similar productive establishments” implies constantly comparing production times across many factories to determine the average — a statistical task the authors could only imagine handled periodically by a “social statistician.” AI systems can run this comparison continuously across the entire economy, updating the social average as techniques change, rather than through episodic manual survey.

4. Algorithmic, non-discretionary deviation detection. The book insists control must come from registering the production stream itself, not from inspectors’ subjective judgment (their critique of Varga’s Hungarian model turns exactly on this point). Anomaly-detection models applied to the registered data stream could flag departures from established productivity norms automatically and publish them — a standing, algorithmic “Inspectors’ Gazette” that never depends on a human inspector’s discretion, preserving the objectivity the authors are at pains to establish.

5. Legible, queryable personal accounts for workers. The Factor of Individual Consumption and labor-certificate balance are, in the text, opaque bureaucratic calculations. Natural-language interfaces over the ledger would let any worker query their own labor-hour contribution, consumption entitlement, and the establishment’s books directly — reinforcing the workers’-council autonomy the authors are protecting (recall: the book-keeping establishment “wields no power” over the economy, only its own accounts) rather than recreating a technocratic layer that mediates between workers and their own data.

Worth flagging the tension underneath all this: the authors’ whole point is that objective, non-discretionary accounting is what makes centralized political power over production unnecessary. AI-driven bookkeeping could realize their technical vision far more completely than 1930s manual ledgers ever could — but who controls the models, sensors, and infrastructure doing this conversion becomes its own site of the very power question the GIK were trying to design out of the system. A system this legible is also a system whose legibility can be captured.”

By Theboss on 09.04.26 5:56 pm

My friend Richard S. recommended Aaron Bastani’s 2020 book ‘Fully Automated Luxury Communism‘ as a relevant to this discussion. I have a copy on order.

By Theboss on 09.04.26 7:12 pm

Further to this, Strategic ghostwriter Neil Hrab commented on LinkedIn:

Not such a far-fetched scenario.

Indeed, I am sure some of the novels published by those giants of Soviet-era science fiction, Arkady and Boris Strugatsky, projected such a development far into the future.

Maybe Martin E. Sandbu can mine the relevant Strugatsky works and return to this topic to ponder some of their novel-based imaginative takes on AI.

This prompted me to ask Perplexity “Which science fiction novels anticipate the current ability of expert A.I. systems to predict our desires and material needs? List books that focus on broader societal prediction (governance, welfare, health).” In addition to classics like 1984, Brave New World, and Vonnegut’s Player Piano, suggestions included:

1. Consider Phlebas (1987) and the rest of the Culture series – Iain M. Banks

The Culture is a post‑scarcity, spacefaring society effectively run by hyperintelligent AIs called Minds. These Minds administer habitats, starships, and social systems, continuously modeling citizens’ preferences and needs so that material welfare, security, and even leisure are arranged before anyone has to ask. The series is arguably the most sustained fictional exploration of benevolent, predictive AI as the backbone of governance and social welfare.

2. Angels and Other Cows: A Celestial Adventure into AI Worlds, the Social Good, and Unknown Connections (2024) – Petra Ahrweiler

A recent open‑access novel that uses fantasy and romance to explore how AI technologies might be deployed for social justice and welfare across diverse political contexts. It directly engages with the idea of AI systems that anticipate needs in health, social services, and community support, rather than just optimizing commerce.

3. AI 2041: Ten Visions for Our Future (2021) – Kai‑Fu Lee & Qiufan Chen

This hybrid work pairs nonfiction analysis with speculative short stories that imagine AI embedded in governance, healthcare, and social services across different countries. Several vignettes explicitly show predictive AI allocating medical resources, monitoring public health, and shaping welfare policy—anticipating needs and risks before they become visible to humans.

By Ian Griffin on 09.06.26 9:39 am

Further to #2 (above)

A 300+-page open access book by German sociologist Petra Ahrweiler reports on the international project “Artificial Intelligence for Assessment” (AI FORA):

Participatory Artificial Intelligence in Public Social Services: From Bias to Fairness in Assessing Beneficiaries (2025)

The report is derived from an empirical study that collected data on social assessment practices and respective AI use in the welfare systems of Spain, Estonia, Germany, Iran, India, Nigeria, Ukraine, China, and the USA. They directly address issues raised by Sandbu’s vision of ‘socialist dream’. Specifically:

AI-based social assessment systems, because they are based on machine learning from historical data, are accused of perpetuating old and introducing new bias and discrimination, often to the detriment of the most vulnerable groups in society. Although types and degrees of AI implementation vary between countries, delegating decisions about the distribution of scarce resources to machines leads everywhere to important questions about ethics, justice, quality, responsibility, accountability, and transparency of welfare decisions triggering reflection about the values behind them. The public discourse on AI-based social assessment for public service provisions often is highly controversial and emotional: Societal core values are affected and at stake.

Cultural context is key:

Not only do societies differ in the values and norms they facilitate, foster, and propagate, but also political, economic, and societal dependencies and pressures vary between societies. Perceptions, attitudes, discussions, and acceptance of AI use for public policy vary between these countries, as do the types and degrees of AI implementation, with reference to norms and values in use, but also related to technology status, economic models, civil society sentiments, and legislative, executive, and judicial characteristics.

Ahrweiler’s novel ‘Angels and Other Cows’, listed in the comment above presents these same issues in a work of fiction:

For me, the best learning experience came from cultural differences. Think about what it’d be like in Germany if kindergarten placements were allocated according to membership in an Indian caste system or according to political compliance like in the Chinese social credit system. OK, the German system has its own biases and discriminations, but at least we are culturally used to them,” Tilda said. Gabriel laughed. “Yes, Mrs Toelz, social service provision is always about social justice, which is a value. It is about what people think needful and deserving people should get from the state. But who is considered as needful and deserving completely depends on cultural context. If our project title were ‘Social Service Provision Using Artificial Intelligence,’ then any good AI should be responsive to values in societies and cultural context, right? (p.250)

By Theboss on 09.06.26 10:24 am