The RCP Interprets AI: The Point, However, Is to Change It
They say what the machine is, but cannot raise a demand. The Vanguard in Retreat, Part Four
They get it half right for the first third: the class divide, the forced adoption, the readers who preferred the machine poem. Then ninety minutes on what the machine is, and not one demand. On an organisation that can read a temperature and cannot write a demand.Recently, the Revolutionary Communist Party streamed ninety minutes on artificial intelligence, on the night of France against Spain, and their broadband gave out twice. They blamed British capitalism, which under the circumstances was the correct analysis. Somebody in the chat wrote in to say their home internet had just gone down as well.
It is a good stream, and I want to be specific about that before I take it apart, because most of what follows is not a complaint about their politics. They are right that this is a class question rather than a moral one, and they have the figure for it: the Americans most optimistic about the technology are the ones earning two hundred thousand dollars a year and up, and seventy per cent of Americans oppose a data centre near them. They are right that adoption is forced rather than chosen, with sixty per cent of bosses now requiring the tools and one in ten saying they have fired or would fire someone who refuses. They are right to refuse the boycott, since you cannot boycott a condition of your employment. And the best five minutes of the broadcast is on the study in which three-quarters of ordinary readers, shown a real poem and a machine imitation, preferred the imitation. The presenter refuses the easy conclusion that people are getting stupider. She says the system has withheld art from them, that everybody has it in them to appreciate the real thing and has never been given the room, and that the finding indicts capitalism rather than the readers. That is a properly Marxist reading and I have not seen a better one this year.
Then, an hour and a quarter in, a viewer asks the only question that matters.
Are there any legal tools to stop this? How do we stop it when most of society is against it, and nobody voted for it, and it is happening anyway?
Watch what happens, because the whole argument is in the next four minutes. First they answer it well, and with evidence. Eighteen proposed data centres in Scotland are on pause, they say, because of protests. In the United States a quarter of proposed projects were cancelled in the first quarter of this year, because of protests. There is the Brick Lane fight in the middle of London. And these, they add, are small-scale protests, so imagine what a larger movement could do.
Then, within a minute, they answer it again, and the second answer cancels the first. It is a question of who owns and controls the resources. The state exists to protect the ruling class. We cannot appeal to the government and protest with appeals, because the government has absolutely no control over this situation, it is completely out of their hands. Every government in the world is looking to AI as the saviour of capitalism.
Both of those things were said by the same people about the same question inside the same minute, and only one of them can be true. If appealing to government is pointless because government has no control, then the eighteen Scottish moratoria they have just finished reporting did not happen. They did happen. They are the results of exactly the method being waved away.
That contradiction is not a slip. It is the shape of the whole thing, and it has a name in our tradition.
Notice what every concrete fight in the stream is doing there. Brick Lane, the Scottish moratoria, the cancelled American projects, the electricity bills up two hundred and fifty per cent in one Virginia county, the student walkout at Leeds Arts University, the crafting circles that turned into anti-data-centre protests. Not one of them is treated as terrain. Every one of them is entered as evidence of mood, proof that the anger is out there and electric and rising fast. The stream can read a temperature with real skill. It cannot write a demand.
And so, when the question of what to do finally arrives, here is the entire answer. How do we stop the rollout in the way it is being done? We have to own it. How do we own it? We have to be in power as a class. How do we do that? Through a revolution. Three rungs, and the ladder reaches from Monday morning to the seizure of state power without touching anything in between.
In ninety minutes on a technology they say is destroying entry-level graduate work, nobody says the word hours. Not the shorter week. Not the four-day week. Not thirty-five, not thirty-two, not thirty. The demand does not appear.
That absence is worth dwelling on, because it is not an eccentric demand nobody could have thought of. It is the demand the Fourth International has arrived at every single time this argument has come round, and it has come round four times.
It came round in 1955, when Harold Robins went through Iron Age and Mill and Factory and the auto workers’ own research and laid out Buick’s automated cylinder-block lines with no production workers on them at all, and a refinery spokesman conceding that eight hundred employees could become twelve. Robins weighed Walter Reuther’s answer, the Guaranteed Annual Wage, and found it a year’s severance for the displaced, which is a cushion and not a solution. What would actually spread the work, he wrote, is a much shorter week. It came round in 1962, when Ed Beecher reported that the US Census Bureau had done with fifty statisticians what a decade earlier had taken four thousand one hundred, and that the office was the cheapest place in the economy to automate. His answer was thirty hours. It came round in January 1984, when the West German section’s paper carried a report from inside the car plants on the second generation of reprogrammable robots, quoting a Volkswagen engineer to the effect that applying every one of these innovations to a single plant would give you a factory empty of people. In May the metalworkers struck for the thirty-five-hour week, stayed out seven weeks, and came back with 38.5 hours from 1985 and a phased path that reached thirty-five in October 1995. And it came round in February of this year, when Daniel Tanuro’s ecosocialist theses on AI reached the workplace and asked for a union veto on introduction, workers’ control over the workload, and against layoffs a reduction of hours without loss of pay.
Four times, on four machines, by people who probably would have had a row if they were all in the same room. It is in the founding programme of the International the RCP claims descent from, where it is called the sliding scale of hours, and it is there for precisely the situation the stream spends an hour describing.
Why can’t they reach it? Because of what they spent the hour on. Is it conscious? Can it make art? Does it atrophy critical thinking? Is the bubble seventeen times the dot-com bubble? One of the two hot takes is that artificial intelligence will never develop human consciousness, and it is argued carefully, with a good passage about a program that had to play thirty million games to reach superhuman play in one narrow domain while a person has to eat and sleep and talk to other people.
I have no quarrel with the argument. I have a quarrel with the hour, because if your question is what this machine is and where it is going, then the only political conclusion available at the end is that the system which owns it has to go. Nothing else follows from a description of a machine. Demands only become available when you ask the other question, the distributional one: who takes the time it saves, who sets the workload, who pays for the electricity. That is the question the archive answers, and it is the question the stream never asks.
The proof is that the machine argument has been had before and settled nothing. It appeared in the Spring 1954 Fourth International, at similar length and with similar confidence, from a young member of the American SWP writing as Lyn Marcus, who closed by observing that an automated machine has less intelligence than what a biologist finds under the microscope in an amoeba. He was right. It saved no jobs in Detroit, and nobody remembers it, partly because its author went on to become Lyndon LaRouche. Tanuro disposes of the same question in two sentences this February and gives the remaining thirty theses to the workplace, the value chain and the hours. That is the correct ratio.
They know how to do it, which is what makes all this worth writing down rather than shrugging at. The best story in the stream is Leeds Arts University, where management pushed generative image tools at students whose whole prospective trade was making images, and between fifty and a hundred walked out. The demand they raised was open the books: who is investing in this, and what did they get. Sharp, local, answerable, and it worked. And in the closing stretch they attack Stand Up to Racism for a mobilising slogan too narrow to cover the second and third generation being attacked in Belfast, which is a serious criticism, correctly made. They can dismantle somebody else’s slogan. They put forward none of their own on AI.
The name for this is propagandism, and the charge is not new, and it is not ours first. We’ll make it again in these pages about this tendency in another context: that the international Ted Grant founded has walked away from concrete conjunctural analysis toward a portable propaganda that is the same in every country because its object is recruitment rather than analysis. The sharpest form of that charge is that the ground was Grant’s own. Arguing against Pierre Frank in 1946, he insisted that generalisations would not suffice and each stage must be examined concretely. A method that answers every conjuncture with the same question, are you a communist, is exactly the generalisation that will not suffice. They tell us, near the end, that one of them had been reading The Class, the Party and the Leadership that morning, and that it gave her a kick up the backside about how much rests on communists in a revolutionary situation. It does. What rests on them is having something to propose when the question comes.
It shows in small ways too. Asked how many people a data centre employs, one presenter says it is a very good question and they will leave it open; later she says she doubts it is many, and adds that she does not know this for a fact and is guessing on instinct. Robins read the trade press. Beecher read congressional testimony. The West German comrades read a Commerzbank study. If your object is analysis, you need the number. If your object is recruitment, you do not, because the number was never going to change what you were going to say.
And there is one more cost, the one our new Red Mole Guide is built to say. The stream’s positive programme is that under socialism, with the workers owning the means of production, we will decide how the technology is distributed. Our library is where that goes to be tested, and it does not come back intact. Leon Sedov looked at the Stakhanov records in 1936, in a state where the expropriation had already happened, and found the productivity real, the mechanism not the machine but a redivision of labour and an intensity nobody could sustain, and the gain taken as revised norms and cut piece rates. In 1981 a Soviet economist writing for Progress Publishers to celebrate the scientific and technical revolution recorded that nearly half of all Soviet workers in industry and building were still working by hand, and explained why: abolishing unemployment had removed the automatic incentive to technical progress and nothing had been found to replace the reserve army of labour. Mandel, on Gorbachev’s computerisation drive in 1985, found the obstacle was not technical either. Making the machines pay required managerial powers that would undo the right to a job, and every previous attempt had run into the class solidarity and instinctive egalitarianism of Soviet workers. To automate, the bureaucracy needed a reserve army it did not have. Tanuro puts it in a line: replacing capitalist property with collective ownership, in and of itself, would not be enough.
So even the good half of the RCP’s answer is only half an answer. Change the owner and you change what the gain is for. You have not yet settled who commands the labour process, and that is where the gain is distributed. Which is why the shorter week is not a stepping stone to be skipped on the way to the real question. It is the form the real question takes on Monday, and it is the thing you can say to somebody who is furious about their electricity bill and their nephew’s job prospects and has no intention whatever of joining a communist party this year.
That is what a united front on this looks like, and it already exists on paper in our own press. Union recognition along the whole value chain, consultation and a union veto on introduction, workers’ control over workload, no layoffs and hours reduction without loss of pay, a moratorium on new data centres pending a real plan for water, minerals and emissions. Every item is something a non-communist can sign. Half of them are what the campaigns in Scotland and Virginia and Brick Lane are already demanding, which is why those campaigns have wins to report and the stream has only a temperature.
They end on the old mole, burrowing under the surface where you cannot see it. Trust me, one of them says, it’s coming. Every generation of Trotskyists has said that too, and what actually arrived was decided by whether anybody had been organised to ask for the hours, to ask for control, to demand the better pay. The mole does the digging. Somebody still has to say where.
Sources. The stream is the RCP’s World on Fire livestream on artificial intelligence, July 2026; figures quoted are as given on air. On working time: Harold Robins, “Automation: Menace or Promise?”, Fourth International, Spring 1955; Ed Beecher, “Automation and the Trade Unions”, International Socialist Review, Summer 1962; Winfried Wolf’s report from the West German car plants, International Viewpoint 45, 30 January 1984; the IG Metall settlement of 28 June 1984 and the phased reduction to 35 hours in October 1995, per Eurofound. On the workers’ states: N. Markin (Leon Sedov), “The Stakhanovist Movement”, The New International, February 1936; S. A. Heinman, Economic Aspects of the Scientific and Technical Revolution (Progress Publishers, Moscow, 1981); Mandel on Gorbachev’s reforms, International Viewpoint 84, 14 October 1985. Daniel Tanuro, “Theses for an ecosocialist critique of artificial intelligence”, International Viewpoint 616, February 2026. The 1954 article is Lyn Marcus, “Automation: The New Industrial Revolution”, Fourth International, Spring 1954. On propagandism and the Grant tradition, our own The Method That Changed Sides is scheduled for August 28. All the archival texts are reprinted, in extract, in Red Mole Guide No. 16.