The Factory of Subalterns

Humans, Machines and Subordinations

Marcello Vitali-Rosati

Macerata - April 9th 2024

The point is not to identify the difference between human beings and machines

… but to understand why we are so eager to identify it in the first place

What is so scary about LLM?

Ci sono milioni di algoritmi: riconoscimento di pattern, di caratteri, sistemi esperti, algoritmi di raccomandazione… chi più ne ha più ne metta!

Perché i LLM ci fanno così tanta paura? Perché sono stati tanto mediatizzati?

Since the launch of ChatGPT on November 30, 2022, the public arena has been filled with talk of AI, the human/machine relationship and a series of anxieties of various kinds: one of the issues that seems to be back in the news is the fear of “existential catastrophe” (Nick Bostrom), a scenario in which artificial superintelligences could take control of the world. Educational institutions fear the disappearance of learning capacities, authors and creators see their activities threatened and are trying to defend human originality and creativity, many professions feel threatened and the idea of a massive loss of jobs seems a near possibility. These reactions and media interest in what is, after all, a fairly innocuous object - a conversational agent… - undoubtedly reveal, first and foremost, the great success of openAI’s publicity stunt, but it’s also a sign of something deeper : an unbridled fear of seeing “man’s” place in the world threatened.

Hypothesis

The topos of the difference between humans and machines is always a strategy to produce subalterns

Larry Tesler

Tesler’s Theorem (ca. 1970). My formulation of what others have since called the “AI Effect”. As commonly quoted:

“Artificial Intelligence is whatever hasn’t been done yet”.

What I actually said was:

“Intelligence is whatever machines haven’t done yet”.

Larry Tesler Website

Many people define humanity partly by our allegedly unique intelligence. Whatever a machine—or an animal—can do must (those people say) be something other than intelligence.

Tesler was a very active computer scientist at companies such as Xerox Parc (1970s), then Apple for most of his career (1980-2000) and finally Amazon and Yahoo. Hofstadter, in his book, quotes a phrase that Tesler is said to have used, and says that for this reason he will call this adagio the “Tesler theorem”. The sentence is as follows:

Artificial intelligence is everything that hasn't been done yet.

This idea has been often quoted and reused, and is the basis of the so-called AI effect. It’s a kind of complex of the field, which always pushes you to think that the intelligence you want to achieve is further away than where you’ve managed to get to. You develop a system that does super-complex things, but you tell yourself that AI has yet to be conquered. You create an algorithm that knows how to play chess, but you tell yourself that artificial intelligence has not been achieved, because the system has - for example - no intuition. We create an algorithm that can write sonnets, but we say to ourselves that artificial intelligence would imply that it also has feelings.

Tesler’s theorem is thus an ever-renewed desire to go further, a kind of Faustian Streben Tat in the field of computer science (where the Ewig-Weibliche becomes the Ewig-Menchliche), a complex that ensures that the machine is always chasing the human without ever being able to reach him.

But Tesler never actually said this. He says so himself, on his personal website:

“Intelligence is whatever machines haven’t done yet”.

Something similar has been sayd by John McCarthy: when it works, nobody call it AI anymore.

What is a human being?

A stable essence of human being and the machine tries to achieve it?
A redefined essence every time the machine achieves a result?

Machines and subalterns

The story of computers

Standing: Ester Gerston
Crouching: Gloria Gordon Bolotsky
Jennifer Light, When Computers Were Women
Kathy Kleiman, Proving Ground: The Untold Story of the Six Women Who Programmed the World’s First Modern Computer

To show that this race to see who’s smarter is really about domination, here’s a story to help us.

The story of the ENIAC, a machine designed to automate ballistic calculations during the Second World War. Jennifer S. Light [@light_when_1999] tells this story in the article “When a documentary [@kleiman_proving_2022; @kleiman_computers_2016]. Light and Kleiman demonstrate that, in this project too, the only names put forward are those of the two male engineers who designed the machine. The very large number of female computer scientists who worked on ENIAC are almost systematically ignored. In official photos, we often see men posing in front of the machine

When women are featured, they are not identified in the caption, considered subordinates or mere extras. Yet six women in particular - Kathleen Antonelli McNulty, Jean Jennings Bartik, Betty Snyder Holberton, Marlyn Wescoff Meltzer, Frances V. Bilas Spence and Ruth Lichterman Teitelbaum, chosen for their mathematical skills, had a decisive role in the ENIAC project, as in the history of computing: that of producing all the calculations. Considered as mere operators, to be replaced by men as soon as the war was over and the symbolic value associated with their tasks increased, these women saw their names and contributions long ignored. What’s more, their material activity could be likened to that of the machines themselves: called computers, they carried out trivial activities that could then be entrusted to a machine. The computers were first women before they became computers - the idea referred to in the title of Katherine Hayles’ book [@hayles_my_2005], My mother was a computer.

Wendy Hui Kyong Chun [@chun_software_2005] also points out that ENIAC programming was done by plugging and unplugging cables: it was a manual, material activity. Its materiality - as in the case of perforating the ribbons of the Index Thomisticus - probably induced the low importance attributed to these tasks, and the fact that they were associated with women.

Women and machines are thus assimilated as two entities whose characteristic is to perform material and trivial tasks in the service of men and their higher concerns. This is the same idea we evoked at the start of the chapter when we spoke of Rembrandt’s painting The Philosopher in Meditation, where the female figure is engaged in material tasks at the service of the “great man”. Servants and secretaries become machines: women who calculate become computers, just as, a few years before, typewriters - secretaries who typed - had themselves become “writing machines” [@kittler_gramophone_1999].

Today, this sexism and assimilation of women to machines can be found in the design of intelligent virtual assistants, whose voices are almost systematically feminine. In particular, Lai-Tze Fan [@fan_reverse_2023] analyzes the case of Alexa, the virtual assistant produced by Amazon, clearly identified as a woman – from her first name. Alexa is designed to fit the stereotype of a person at a man’s service, a sort of Rembrandt-like secretary or servant – sometimes even a mother – always ready to do whatever her superior – or son – asks of her. Fan demonstrates that Alexa’s design “reflects traditionally feminized work” and embodies a deeply sexist worldview to the point of implementing highly problematic responses to commands from the user – obviously imagined as male – that include verbal abuse.

Quando la macchina toglie il lavoro ai poveri e ai subalterni siamo contenti!

What do LLMs do?

A probabilistic model
Reinforcement learning and alignment
the distributional hypothesis

the meaning of a word depends on its relationships with the other words

ChatGPT does indeed seem to have abilities that we normally identify as being characteristic of “men”, or more precisely “of man in his highest characteristics”, i.e. of an elite: to manipulate natural language correctly, without making mistakes, to produce “original” speeches, to demonstrate “creativity”. In the case of French, in particular, the very form of the language, with its complex orthography that cannot be deduced from spoken language, is designed to produce elites - who will be able to master this complex object - and subalterns - who will be unable to do so. But what if a simple algorithm, a “machine”, can achieve this feat better than a manager? Or even better than a literary scholar? A skill that used to be used to produce a hierarchy can no longer be used to produce a hierarchy. Hence our enormous anguish. What is threatened by LLM is not the specificity of humans in the face of the coldness of machines, but the tranquillity of elites who see the gap between them and the mass of subalterns narrowing.

Firth and the DI.

Some features

Originality
Creativity
Lack of notion of truth
Excellent manipulation of natural language
Consensual speech
Verisimilar is preferable to true
Speech that represents a majority (in terms of power)

Help! I am…

a politician
a (great) author
an academic
…

Alan Turing

Arguments from Various Disabilities. These arguments take the form, “I grant you that you can make machines do all the things you have mentioned but you will never be able to make one to do X”. Numerous features X are suggested in this connexion. I offer a selection: Be kind, resourceful, beautiful, friendly, have initiative, have a sense of humour, tell right from wrong, make mistakes , fall in love, enjoy strawberries and cream, make some one fall in love with it, learn from experience, use words properly, be the subject of its own thought, have as much diversity of behaviour as a man, do something really new. […] No support is usually offered for these statements. I believe they are mostly founded on the principle of scientific induction. A man has seen thousands of machines in his lifetime. From what he sees of them he draws a number of general conclusions. Turing, Computer Machinery and Intelligence, 1950

Knowing math?

Bene speremus, Hominum enim vestigia video

Edmund Halley, Apollonii Pergaei conicorum (1710).

In some seventeenth-century books on mathematics and geometry (notably a commentary on Euclid’s Elements), we find illustrations of castaways arriving on a beach and, on seeing geometric figures traced on the ground, exclaim: “Bene speremus, hominum enim vestigia video” (Let’s be confident, I can see human traces).

In this case, the typically and essentially human characteristic was the ability to do geometry and mathematics. This image is particularly telling: what the castaways mean is not so much that there are animals belonging to the human species on the land where they have found themselves, but, more precisely, that there are “civilized” human beings, i.e. white, Westerners, and not “savages”. In the good tradition of the imaginary of the great discoveries, this is precisely the question. Geometry is an essential characteristic of “man”, as it enables us to differentiate between superior and inferior beings, the dominant and the dominated. On the one hand, Western colonists with their civilization, on the other, everything else: inanimate things, animals and human beings who don’t share the same culture.

Saying that the ability to do geometry is unique to humans is actually a strategy to justify colonization and to create subalterns: we have the right to dominate you, because you are less human than we are, as demonstrated by the fact that you have no geometric or mathematical skills.

The funny thing - if we can still smile in the face of such immense violence - is that if, in a contemporary novel, we were shipwrecked on an island where we found traces of geometric figures, we could say to ourselves that this might be an island inhabited by machines. Machines” are obviously much better at mathematics than human beings. The ability to count and calculate can certainly no longer be used as a characteristic of an elite - indeed, this is why the task of counting was left to women in the early days of computing, and why “the first computers were women”, as Jennifer Light puts it… (When Computers Were Women).

Making the bed

I’m intrigued that AI has by now succeeded in doing essentially everything that requires “thinking” but has failed to do most of what people and animals do “without thinking” – that, somehow, is much harder! (Nils J. Nilsson, The Quest for Artificial Intelligence, 2009)

Making the bed is what most deeply characterizes the human

What bothers us about LLMs is that they seem to have abilities that have long been used to characterize elites. As long as we automate tasks reserved for subalterns, there’s no problem, because automation helps and makes efficient the mechanism for manufacturing subalterns. The automation of manual skills - such as weaving and harvesting - with the first industrial revolution helped to reinforce the separation between the elites - who own the machines and the capital - and the subalterns, who own only their own labor power, now in competition with the labor power of automatons. But when an algorithm seems capable of doing what is normally an activity of the elite, then… we’re afraid.

The various reports released in recent years on the types of work threatened by automation reveal, once again, this strategy of manufacturing subalterns. In their 2013 report (The future of employment: how susceptible are jobs to computerisation?), Frey and Osborne reassured the elites by asserting:

The expansion in high-skill employment can be explained by the falling price of carrying out routine tasks by means of computers, which complements more abstract and creative services.

The tasks of subordinates will be increasingly devalued (and devaluable), because they can be automated. But rest assured, executives, managers, intellectuals: your jobs are not at risk because you are more human than your subordinates. They can be replaced, not you!

A few years down the line, executives are less tranquil: the World Economic Forum’s 2023 Future of Jobs Report tells us:

But while expectations of the displacement of physical and manual work by machines has decreased, reasoning, communicating and coordinating – all traits with a comparative advantage for humans – are expected to be more automatable in the future.

Aïe aïe aïe! We thought we could continue to disqualify tasks traditionally assigned to subalterns, but LLMs seem to be turning the tables: we can automate the writing of a very well-written, diplomatic, politically correct e-mail, we can - dear colleagues - automate the writing of a funding application, and even the writing of a pretty poem, the reading and summarizing of a complex text, the writing of a rendering for a university exam… but we can’t automate the task of remaking a bed.

We are obliged to note that the production strategies of subalterns as they have stabilized over the last few centuries - particularly since the first industrial revolution - are beginning to be in crisis. Social, cultural and technical changes are restructuring the symbolic values that hold a society together. The cause is probably not just a handful of algorithms. It goes much deeper. If we think of the texts on the end of the modern concept of the author by a Foucault or a Barthes in the 1970s, we can easily understand that the symbolic status associated with the ability to scribble sentences and the production of a figure such as the Author, based on the notion of originality, have been in crisis for at least half a century. ChatGPT doesn’t provoke this crisis, it confirms it.

And now… ?

We could find something which machines cannot do

And now… ?

We could find something which machines cannot do

Or…

And now… ?

We could find something which machines cannot do

Or…

We could question how we make subalterns

Faced with this crisis - which is by no means a revolution, but rather a long, slow, gradual change in the values on which our Western societies are founded - we have two options. The first, the simplest and the one that will undoubtedly come to pass, is to modify the strategies for manufacturing subalterns, identifying new characteristics to define elites and new tasks to assign to subalterns. Let’s follow the reports on the future of work and ensure that the privileged can remain so, by choosing to focus on particular abilities that are not automatable - for the moment - or that we decide we don’t want to automate.

The other possibility would be to realize the huge lie that covers these strategies. We should stop believing that there is something like a human “essence”, and ask ourselves what the real political reasons are for declaring again and again: “and that, a machine will never be able to do”. With the likes of Karen Barad, Katherine Hayles and Rose Braidotti, we could ask ourselves why “the boundary between human and machine stabilizes and destabilizes” (Barad, Meeting the universe halfway), and how and why this boundary defines itself in a certain way. In this way, we might realize that there is no hierarchy in the emergence of meaning: that the “little hands” - as Margot Mellet calls them - are not little at all!