long read | 2025 | en français

how AI is made¶

AI is a tool bag that serves a project; a political project.

Let me explain the motors but first the motivating forces:

intellectual history¶

Cybernetics is: “the study of systems of any nature which are capable of receiving, storing, and processing information so as to use it for control” (Andrey Kolmogorov, a Soviet mathematician). Cybernetics was founded in 1943.

A recurring theme in cybernetics is feedback. Feedback is a process where the observed outcomes are captured and input for further processing. An example is chatBots.

In 1955, John McCarthy, then a young Assistant Professor of Mathematics at Dartmouth College, New Hampshire, decided to organize a group to clarify and develop ideas about “thinking machines”. He picked the name “Artificial Intelligence” for the new field. He chose the name partly to:

• avoid “cybernetics” which already had an assertive influencer in the person of Norbert Wiener;
• obtain funds from the Rockefeller Foundation for a summer workshop at Dartmouth at about ten participants.

The only AI that is currently obtaining success:

• has no learning rules,
• instead “learns” through experiment from observations.

changing relationship to the world¶

‘AI is an ideological project to shift authority and autonomy away from individuals, towards centralized structures of power. Projects that claim to “democratize” AI routinely conflate “democratization” with “commodification”.’

Ali Alkhatib (2024)

“Maybe you could actually unilaterally change the world without having to constantly convince people and beg people and plead with people who are never gonna agree with you through technological means. This is why I think that technology is this incredible alternative to politics.”

Peter Thiel (2010)

logic or generalization¶

Here are two learning paths. We can assess a proposal with logic or generalization; deduction or induction:

• Deduction is inference from premises known or assumed to be true. The laws of valid inference are studied in logic.
• Statistical “inference” is induction: we go from particular evidence to a universal plausibility.

AI does not reason. It applies probabilities. It’s like eating an apple and assuming that everything will be as tasty:

“Sometimes it’s best to explain a concept through a concrete example:

Imagine you grab an apple, take a bite from it and it tastes sweet. Will you conclude based on that bite that the entire apple is sweet? If yes, you will have inferred that the entire apple is sweet based on a single bite from it.

[Statistical] inference is the process of using the part to learn about the whole.

How the part is selected is important in this process: the part needs to be representative of the whole. In other words, the part should be like a mini-me version of the whole. If it is not, our learning will be flawed and possibly incorrect.

Why do we need inference? Because we need to make conclusions and then decisions involving the whole based on partial information about it supplied by the part.”

Isabella Ghement (2020)

coefficients and input¶

As an option, here’s an algebraic intermediary. You may skip it.

Let’s define a coefficient b to translate the result so that the output of the model lands where appropriate:

algebra: y = ax + b is deterministic if input x is fixed.

Now, suppose that we do not not know the value of b: it still is a coefficient but its value is to be determined. The program will assign a working value to b depending on calculations:

linear regression: y = Xß + b is probabilistic where X is a matrix of explanatory variables (to be instanciated by samples).

In classification and generative AI, we repeatedly assess the probability of truth via ̂a, the activation, the result of the sigmoid function sigmoid(wᵀX + b). Our output is in the range of 0 to 1; with sigmoid showing a spillway curve; w being a column of “weights” (unknown coefficients).

data applications¶

What are generally referred to as ‘data’ is values obtained and mined from the environment. The engineer takes them in bulk to manipulate them. That is how we are talking here about ‘data’ in the singular as the pile of values obtained.

Data science extracts large quantities of data and extrapolates information from them:

• data analysis, content-based
• regression, image compression, “you might like”, click on ad
• interpretable AI, expectations, “also liked”
• “Supervised Learning”: image search, speech recognition, translation, driverless car, recommendation concierge

generative AI¶

The method of generative AI is to confabulate; to make up stories. Not hallucinations or lies.

Applications of inventive “artificial intelligence” include: feature design (no consistency, more bugs), materials discovery, AI war (improvisational military robots, military intelligence and planning), self-organizing control systems in robots and space vehicles, entertainment.

LLM¶

Chatbots are based on LLMs (for large language models: representations of written sentences) and are convenient: ChatGPT, DeepSeek.

‘I believe code is a uniquely dangerous deployment site for “AI”, because in programming, the most dangerous thing is code which is almost correct. The human brain has a significant tendency toward seeing what it expects to see and a person can sometimes look directly at something that’s mostly correct but wrong without noticing it. However, creating this almost correct code is exactly what “AI” is best at. The purpose of the content generators people now call “AI” is to create something which can plausibly fool a human into believing they are looking at a real picture, real paragraph of text, or real computer program code. So Copilot might as well be designed to generate bugs a human cannot catch.’

mcc (2025)

More errors will appear in shipped software, leading to more user complaining and changing provider. Corrections will be more difficult; so quality will degrade further, leading to more escape and unstoppable cashflow loss.

business or society 📦¶

Generative AI and LLMs are efficient to industrialize intellectual production. (While degrading quality.)

a magnificent opportunity of cost reduction¶

Organizations choose yield and prefer to reduce cost. There is a billion-dollar case for imagined productivity gains. “The AI jobs crisis is here, now”.

“It was at that very same conference that I learned critical thinking takes up 20 percent, sometimes 30 percent, of company time. It’s clear to me that some of you are not focused on the profit potential of outsourcing all of our thinking to a machine capable of remixing thoughts that have come before.”

“The way I see it, we’re family. It really does disappoint me that so many brilliant colleagues—whose genuine breakthroughs I’ve profited from for years—would be so quick to condemn this newer, stupider way that I and others like me can make money off your life’s work”.

Amanda Bachman (2025)

profit unfolds on AI adoption¶

[x] AI promises are sparkling [x] indicator shows 25 % more [x] directive to make with 20 % fewer [ ] downsizing [ ] pay reduction

(For simplicity, let’s consider that selling prices or not-for-profit output are roughly stable.) We have a sequence:

1. Employers dream that AI would enable an increase of output.

2. So middle managers show delivery indicators at 25 % more.

   The labour cost is the same but the dream is of an output at 125 % times what it was. If material costs are neglected, then the workforce value could be discounted at 100 %/125 % = 80%.

3. So the direction board decides to target the same services with pay at 80 % of today.
   A possibility is downsizing: tending to get rid of 20 % of staff.
   Big firms are already laying out plans to remove median wages.

4. We can pretend that you are paid 25 % too much! Market evaluation of work worth is immediately reassessed at 80 % of what it was.

5. Individual competition and union support.

Original idea

an advance of control¶

‘A technology prevails mainly because of the superiority of the players promoting it,’ writes sociologist Juan Sebastián Carbonell. "AI" gives supervisors more control, ‘that is, the power to decide how we work and with what tools.’

"While AI, like bureaucracy, is presented as a generalised and goal-oriented form of rational process, it actually produces thoughtlessness; the inability to critique instructions, the lack of reflection on consequences, and a commitment to the belief that the correct ordering is being carried out."

Dan MacQuillan (2025)

conflicts of interest¶

Treating persons as resources has long been feasible. (For example building pyramids.)

"What we’re seeing is a continuation of past behavior. Tech companies have been doing layoffs for years now, the trend started before the AI boom. Same with media companies, they’ve been losing revenue for decades, layoffs and Private Equity acquisitions have been constant. The new thing is that they can now spin the layoffs as positive, because they’re AI innovators."

Alan Vezina (2025)

Companies and institutions are whipping the promises of AI up into inescapable prospects. This tactic makes it possible to weaken resistance, to ignore discussions on the challenges of AI and the transformations it brings about in society, and to redirect funding towards the big firms to the detriment of workers and beneficiaries.

AI is being harnessed to deskill, de-spirit, and demean the work and craft that many people have spent years developing (2025)

🧱 On the contrary, resistance to the shrinking of public services opens up a front of counter offensive. The indignation and subsequent mobilisation in the face of budget cuts and the grabbing of personal data show that the aim is not to improve the efficiency of services but to reduce them and to centralise power and money.

We could make tools subject to market authorisation, like medicines. According to the AI Now Institute, AI should be banned for emotional recognition, social rating, price or salary setting, compensation claims, assistance, replacing teachers. A challenge is to increase the range of prohibitions.

bias, teaching 🌱¶

AI does not know anything. It generates ‘artefacts’ called ‘responses’ on the basis of ‘models’ chosen by engineers according to their context. This is mass production, not statistics.

Bias is everywhere:

1. The software does not tell a ‘truth’ but produces an impressive result: the wow effect!
2. Before that, people designed the software with an engineer culture in mind: “We make things, we don’t do ethics”.
3. Even before that, people had chosen the data set: double bias: engineers pick what they want and it only is a record of a past.

If a professor wanted to teach the three moments:

• The audience should understand that the result is a fabrication: the machine is inventing a tale. A demonstration could consist in taking an official dataset (of population or activities) and generating a shocking result. First disasters
• The technicians should be trained to incorporate explainability.
• The technicians should be trained to improve the data set.
• A representative of vulnerable groups can be brought on board at any stage.

a bayesian method¶

“Data Science” is science in that a person comes up with an hypothesis and tests it. Its structure is made of intuition and credibility:

By design, the engineer starts from a “prior” belief, prepares data, chooses a probability function, triggers machine so-called “learning”. The output is not a probability but a likelihood, an “updated prior” belief, a function that seems “plausible” in the context.

The Bayes’ Theorem can then be explicitated in English:
a “Posterior” Belief that a thing Θ will happen | after considering event É, can be estimated by: (Likelihood of event É appearing | given Θ happened) times (“Prior” Belief that Θ will happen, before considering event É) divided by (Probability of event É in any circumstances)

How to use a Bayesian Belief Network:
The software performs inductive inference: from an event that sets variables in known states, it subsequently computes beliefs of interest, conditioned on this event.

Practical examples of the use of Bayesian belief networks include medicine (symptoms and diseases), bioinformatics (traits and genes) and speech recognition (utterances and time).

If wrong prior belief, then wrong outputs. It is a feature of bayesian statistics, not a bug.

“Machine Learning”¶

Prediction is the inductive calculation of a likely outcome depending on new data (observations or prompt). Data models consider outcomes as mathematical functions of “explanatory variables”.

An example of a sequence of machine learning program is:

1. Quantify nature.
2. Represent similarities: constrain “distance” between predicted and observed.
3. Iterate? (neural network)

Python/Numpy now has built-in ease of use and parallelism: vectorization.

efficiency or consideration¶

Black box or interpretable? 🧐

• Precision is offered to the general public: only 20 % of the time is devoted to statistics. ‘Five PhDs is not compatible with our business’ (said my teacher to the class).
• Accuracy can be sought for craftsmanship: 80 % of the time is spent on statistics. (Ethics.)

So we have an ethical tension between delivering quickly and cooperating:

Shall we transform coordinates? A linear space is a set with a structure that enables linear combinations. The elements, often called vectors, can be added together and multiplied (“scaled”) by numbers called scalars. Changing linear space is efficient: it allows the program to take the place of the researcher to manage filling, quality, learning, balance and noise.

On the contrary, we may want to enable our clients to understand how the model works. This is known as interpretability: allow practitioners some space to look at what the programme does. Interpretability allows maintaining a relationship with the occupation represented by the data. To achieve this, we need to:

• preserve explanatory variables ;
• artisanally manipulate at each stage.

Reference for everyone: The explanations can be difficult to convey to end users and line-of-business teams

Reference for engineers: “Interpretable and explainable machine learning”

My mathematical justification: Regression Redress

how a technophile would use AI 🧩¶

“Hallucination” and “agents” are anthropomorphic terms. Putting traits on AI serves to make people fear that AI would replace humans.

Generative AI does not hallucinate: instead it always continues production. For testing/quality purposes, we may get the AI robot to detail how it arrives at the response it generates – including specifying sources and the credibility of those sources. They may not exist; this happens because the program invents references as needed. Indeed conversing with the machine is interesting. No reliability can come out of it because it builds on probabilities.

“LLMs are designed to mimic the way people use language, first through pre-training on next-word prediction and then through additional rounds of redistribution of probability mass, called RLHF and the like.

“It only becomes apparently mysterious when we do a lot of (reflexive, sure) interpretive work on the output and tell ourselves stories that involve the machines doing anything other than repeatedly calculating a likely next word.”

Dr. Emily M. Bender (2025)

AI robots output suboptimal solutions and cannot give context to make better decisions. I would suggest that we keep AI usage to simple tasks of disposable output.

"AI optimism requires you to see yourself and your loved ones as safe from AI; as the passengers in the self-driving car, and not as the pedestrians it might run over."

Josh Collinsworth (2025)

how a praxister would use AI 🌳¶

“i think the best way to use LLMs is this: don't type into the AI chatbot's input field. instead, write a description of the problem in a text file. work through the specifics of what you want to accomplish, and how you'd go about doing it. once you've got a few hundred words of ideas and planning, you can go back to the empty, unused AI tab and close it. then begin doing the work yourself. consider sending your notes to friends or experts. when you're done, remember to thank them for their help!”

Josef (2025)

Praxis:

1. practical application of any branch of learning
2. deliberate action of a rational being

Hence praxister:

1. A person who chooses her practice on purpose
2. A practitioner who intentionally avoids AI

(First occurrence)

For McQuillan, flipping to “decomputation” is a way of preferring the value of situated knowledge and context over transition to scale.

AI damages intelligence¶

We use AI to think less. Therefore our brain untrains as we defer to an AI tool:

Cogitation is costly; so:

• the more we use AI,
• the more we offload processing,
• the less we ponder,
• the shallower our thinking becomes.

On the other hand, education level does not contribute much to critical thinking. More precisely, the statistical model calculates an importance on predicted scores of seven times more from AI usage compared to the level of education attained by participants, ranging from high school to doctoral levels.

“Feature importance analysis underscores AI tool usage as a major negative predictor of critical thinking.”

in “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers”, by Hao-Ping (Hank) Lee et al. (2025).

“Qualitatively, GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship.”

From a study: “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking”, by Michael Gerlich (2025).

“However, that process has already been ongoing for a while even before AI. A precursor of all this was the inability to do mental math. That must have disappeared already before my time because I have run into this problem literally every single time I have had to train students. At the most elite institutions no less.”

via Yves Smith

perspective¶

The AI project is a means of power. It is leveraged by capital. the military, Silicon Valley, Washington, Tel Aviv Jaffa, Beijing, Hangzhou…

On the contrary, we have sketched out ways of improving society.

other references¶

Catherine D’Ignazio and Lauren Klein, Data Feminism
Iris van Rooij on science (2024)
Khrys (in French, 2025). "« IA », Philosophie du Libre, Féminisme"

attribution¶

The present document is authored Eric Maugendre and is available to re-use under the condition of CC BY-SA licence.