ESCAPING THE AI FARM

Experiments with AI September 2024. Phil Hall, using Bing.

---

We Need Modular, Autonomous AI

‘They say, “Sing while you slave” and I just get bored
I ain’t gonna work on Maggie’s farm no more’

Bob Dylan

by Phil Hall

Scanning the horizon carefully reveals a divergence in the application of artificial intelligence that sets a nominally socialist productive model against a parasitical capitalist one.

It has becoming increasingly clear that the most effective applications of AI come from China. There, the focus is on tangible, limited applications integrated directly into manufacturing. The Chinese are using AI to operate robot arms in factories, to optimise logistics, and streamline industrial processes. They are building expert systems trained on specific, high-quality training data to perform defined tasks with increasing efficiency.

This is in contrast to the US capitalism’s dominant model. Thinkers like Radhika Desai have pointed out that huge amounts of money are funnelled into building group AIs for sinister surveillance organisations like Palantir. Into great, all-encompassing AI systems that attempt to be jacks-of-all-trades. The goal with these systems is not utility, but skimming money off people, speculating on financial markets, and intruding into all transactions to take their pound of flesh. In China, AI systems are focused, practical and efficient. Though, of course, they are also used for surveillance.

But what is the philosophical point of creating an all-seeing inanimate computer communication system that has no consciousness? Such a system, consuming vast energy resources, is pointless if its main beneficiary is not the public. Who do these AI systems like Palantir really benefit?

Huge LLM models, trained on the entire unfiltered mess of the internet produce limited benefits for ordinary people and can cause addiction and help cause people to detach from reality and to hallucinate false, often engineered realities. A large part of AI use in the West is as a toy. Even now its use in recycle human experience to create ‘art’ is generating a deep sense of existential nausea.

More sinister is that one of its main functions is as a personal policeman and a shaper of thought, in part to help ensure the continuance of the current ruling elites in power. To generate support for wars against Iran, Venezuela and Russia, to attempt to generate support for genocidal Israel. US corporate AI is used as a tool for social engineering, intelligence control and advertising, constantly nudging our perception of the world.

You don’t need an AI that understands the nuances of kitten videos, the depths of human depression, or the mysteries of quantum mechanics to control a robotic arm on an assembly line. You need a precise, reliable, and specialised tool. This is the engineering mindset: build the right tool for the job.

The motivations behind these two paths are also different.

The Western model, particularly as championed by what can be termed the financial criminal class, is not about creating value, but about extracting it. The AI they deploy is a sophisticated form of scamming and skimming—a digital tax erected on every economic highway. By interjecting itself as a seemingly necessary and convenient intermediary in platforms like Amazon and Uber, it presents itself as the only possible model, rejecting commons alternatives. Remember when AOL tried to control the Internet and make it seem inevitable that we would all have to pay to access content? This LLM technology in combination with satellites acts as a mechanism for wealth parasites like Bezos and the rest of those sharp-mouthed fat tapeworms.

US-based AI siphons value away from the people who actually create it—the workers, the drivers, the small business owners—and redirects it upward to shareholders and financial engineers. It is a tool for defrauding ordinary and poor people of a fair share of their own income.

We need modular, autonomous and trainable intelligences

A general AI that can ‘draw’, philosophise, and write pastiche poetry and novels is useless to most of us humans. We need a practical, serviceable and trustworthy AI, not huge, unreliable, opinionated, dumb-gods. We should cultivate an ecosystem of modular, autonomous and trainable intelligences that develop and embed the specific expertise we require, protected from the dangers of outside intervention. We need useful AI tools for everyone from electricians to teachers, to vets and lawyers that don’t come at the price of giving away a piece of your soul to some Mephistophelean company.

Rulers fear an educated public because education enables people to see through the bosh to the base motives. The educated sections of the citizenry will ultimately be inspired to take decisive direct action and lead the people in a revolution against a government that harms the interests of the majority of its people. Ask Charles Stuart, Louis Bourbon, Nikolai Alexandrovich Romanov, Henry Puyi Huantong and Fulgencio Batista.

Traditionally the ruling class has used systems of ideas and knowledge to maintain control: religion, nationalism and racism, the histories of kings and queens, the subject focus of university departments funded by Microsoft, economic theories that rest on Malthusianism and social Darwinist market philosophies, theory heavy books vaccinating against revolution, the castrated rebelliousness of ‘radical’ experimenters.

The idea of winning control of storylines of inducement and fear and possibility comes through Hollywood, through art and mass media. The purpose of the money is to buy as much complicity as possible in as many people as possible in order to bring as many people as possible over to the side of the exploiter and oppressor; the rich and powerful pay intellectual flunkies, Salieris, to justify their system so full of inequality in order to maintain their position.

The neo-imperialist, capitalist ruling class are the real true Marxists of conviction. They wage relentless class warfare and intervene all around the world without scruples while feigning liberal values, and AI is now one of their favoured tools.

---

Sovereign (we control it) autonomous modular AI is the tool the ruling class fears; to download a private, autonomous copy of Deepseek sealed off from the Internet on a laptop using Linux is to take back the information and knowledge commons. Meanwhile, general AI becomes an all-powerful seeing eye capable of dominating discourse, targeting individuals and spreading disinformation. Do you really need to use a ‘free’ AI owned by Palantir, or by an Israeli company? Should you really give it all your personal information so that it schedules your day?

There are profound political and epistemological implications of current AI. The architecture of information systems is never neutral. It is deeply political and relies on definitions of orthodoxy. Gemini will spin you a line on Gaza, Ukraine, Yemen, Iran, Cuba, Venezuela, Nigeria, Taiwan and any other country the US corporations target.

The future of practical, robust and trustworthy AI in the physical world is not about building a single, all-seeing general intelligences that talk shit, paint new Rembrandts and write pastiche poems.

By controlling the World Wide Web like a spider Microsoft Copilot, Google Gemini, Facebook Meta, and Grok, can win an information war and their owners dream of a future where, in controlling AI, they will also finally win the global knowledge war; against socialism and shared wealth and well being, and capture the wriggling minds of everyone in their sticky chatbot threads, guaranteeing social stasis and inequality, shoring up systems of control and ownership, continuing to extract wealth, and resisting the natural demands for security, food and health of ordinary people. The capitalist class oppose all real substantive democratic change.

AI and its associated algorithms provide dopamine-stuffed content for the naturally gluttonous human imagination: push-button pictures of food, of sex, triggers for love and status seeking, identity-affirming words; all promoted to maximise clicking, watching, scrolling and listening.

The shit you are fed from the World Wide Web is copious and almost inescapable. The slush pile is infinite. The AI can generate more unique, persuasive fakes, articles, social posts and inflammatory snippets in a second than a human could see through in a lifetime. Universal corporate-controlled AI in the hands of US corporations is an existential threat to autonomous human thought.

---

Defeat for ‘we, the people’ and the planet in this latest iteration of the class war does not come when AI loses its training wheels and goes out of control and turns the whole world into paperclips. Defeat for humans, and all life on the planet comes when AI helps accelerate the transformation the entire living world into a machine to generate money, status and power for a microscopically small number of dysfunctional billionaires and their hangers on; when AI is allowed by corporate-captured governments to optimise every system to produce shareholder profits, and makes the world into one big mushroom farm, keeping everyone in the dark and feeding them shit.

By encouraging people to rely on AI chatbots to give them their bearings, corporations want to create lobotomised dependencies. We are all cottoning on to this, now. This is why, those people who are more wide awake to AIs dangers now are increasingly turning away from its injudicious use.

A population that can understand its own painful reality and verify its own facts, and understand its own infrastructure and systems of control is a population that will be in a better position to exert democratic, social and even revolutionary pressure for change: to organise together to fight for social justice and the redistribution of wealth and power. Rather than using corporate AI chatbots, nowadays we should make an effort to use closed off specialised, modular autonomous open source AI tools that do not violate privacy, monitor our every move and try to manipulate our political outlook.

---

A single, autonomous AI on your laptop does not need to know everything, or to be a Jack-of-all-Trades. An AI just needs to be an expert tool at its specific set of jobs, on a specific machine, (or a closed-off network of machines) in one specific context. Modular autonomous AI can exist disconnected from the corporate cloud on a laptop and that is what makes this kind of AI both more useful and safer.

Citizens running specialist autonomous AI systems locally on their own machines are citizens thinking for themselves, solving their own problems, and building their own models of the world; running specialist autonomous AI systems is anathema to surveillance capitalism and behavioural prediction.

Modular autonomous intelligence represents local, people-controlled intelligence. Those who need this kind of AI are not supplanted by it, but will use it as a tool and deploy it as they will. Processing happens only your device or a set of devices. You can choose whether you connect or not to the wider Internet. Your queries, your data and your conclusions are not syphoned into a cloud to be mined and used to model you. Rather, your local AI is a tool for you, not a sensor placed on you.

---

You do not need a planet’s worth of data. You need high-quality, context-rich data from the specific domain. A plumber using an AI model trained exclusively on 500,000 images of pipes and leaks will be more accurate at detecting them than a general AI model that is also trained on 50 million images of cats, cars, and street signs.

Instead of a single, overweening (preening) intelligence managed from Silicon Valley, we need local, modular AIs that are independent with optional connectivity and specific to different tasks. AI machines can be highly efficient AI specialists, not generalists with ridiculous reactionary opinions picked up from a mountain of shitty training data and twisted by algorithms inserted into them by intelligence agencies.

Modular autonomous AI systems result in a massive increase in reliability because of the homogeneity and reliability and relevance of the training data. The result is a revolutionary transition to predictive maintenance where expert AI systems reliably anticipate failures. Perhaps they automatically order parts or take actions, and schedules precise repairs or suggest actions before a breakdown occurs in any process, minimising downtime. Maintenance is next to Godliness—self-healing, autonomous systems will provide cascading societal benefits and mean unparalleled convenience, security, reliability and sustainability and give machines independence in space, under sea, and in harsh environments like the desert and the Antarctic.

In contrast, the current AI model is a Palantir-style universal, all-access, AI system that overtly embraces evil (see the quote below) while attempting to flatter its users and manipulate them.

Free corporate AI is a panopticon and social surveillance operation that is currently embedding itself everywhere in Europe with the consent of the transatlantic traitors, the bought-off corporate stooges who pass for members of government in Europe.

Starmer and his cabinet, with the lowest popularity ratings of all time, give the impression of being far more concerned about securing their futures on the boards of these military industrial tech companies after office than about the economic and social well being of Great Britain. The doors revolve between the corporations and government in the EU and especially in Britain. Tony Blair, a case in point.

---

To quote the speed-talking, wispy-haired Alex Karp of Palantir:

“You can have privacy and secrecy, or you can have security and transparency. But you can’t have all four… In a world where you are being targeted by adversaries using the most advanced tools, the idea that you can have perfect privacy and perfect security is a fantasy.” (2020 letter to shareholders)

And

“The ‘Don’t be evil’ framework is a luxury good… It is a framework for a time when there was no existential conflict. We are now in a time of existential conflict, and you have to be willing to do things that are controversial.” (Interview with CNBC’s Andrew Ross Sorkin, November 2023.)

And

“The idea that in a modern society you can be anonymous is a fairy tale… Our enemies do not operate under these constraints. We need to have tools that allow our institutions to function effectively and protect our citizens.” (Bloomberg Panel at the World Economic Forum in Davos, January 2023)

To summarise: One AI to rule them all, one AI to find them, One AI to bring them all and in the darkness bind them.

Modular, autonomous AI, in contrast, is quite bad at being Sauron; much better at doing what it is designed to do. Cobbler, cobbler, mend my shoe! Modular, autonomous AI only knows what it knows within specific limited domains. This type of sealed-off AI is a threat to inefficient machines and systems, but it is not an instrument of social control.

---

The future of practical, robust, and trustworthy AI in the physical world is not about building a single, all-seeing general intelligences that talk shit, paint new Rembrandts and write pastiche poems. Corporate chatbot AI has far too many political opinions. We are forced to access them whenever we use a search engine. Modular, autonomous AI, in contrast, is about creating ecosystems of trainable intelligences with deep, reliable independent expertise. They are intranets.

Clearly, we need to shift away from monolithic, general-purpose AI to modular, specialised, autonomous, contextual intelligence, but, this practical shift will mainly involve prying the claws of the corporate vultures like Palantir embedded in our necks after the establishment of a society based on humane socialist principles.

At the moment this action feels like a very difficult thing to do, but in the United Kingdom with the oncoming rebellion against digital ID and the economic crisis caused by the UK and European governments prosecution of their speculative proxy war in Ukraine, the situation could start to change very quickly. We have to be clear! We can embrace the usefulness of a more sovereign, specialist AI without necessarily becoming modern Luddites.

Modular sovereign AI platforms will provide a less sinister and more resilient technological foundation for a future humane socialist society. For thoroughly good advice on how to put this easily into practice in the interim, it is strongly recommended that you consult Rob Braxman on YouTube, here.

Begin by downloading and installing the Linux operating system and pay for a local version of DeepSeek. Install it on an air gapped laptop disconnected from the Internet and use Retrieval-Augmented Generation to train your local AI with your own private documents and books and pictures and videos, so that your AI speaks only from your trusted private source.

---

Phil Hall was born in South Africa into an ANC family with British, French, Austrian, and German roots. After his parents were exiled, they lived in East Africa and India before returning overland to the UK. In the UK he studied Russian and Spanish literature, politics, and economics. After graduating he specialised in descriptive and applied linguistics. Phil has lived and worked in Spain, the USSR, Mexico and the Gulf. Returning to London during the pandemic, he co-founded the Humane Socialist magazine, Ars Notoria (the Art of the Noteworthy) and the micropublisher, AN Editions.