An article published by The Wall Street Journal offers a deep critical perspective on how modern artificial intelligence systems work, especially the large language models behind smart chat programs like ChatGPT. The article presents a personal experience of a software engineer in the newspaper’s newsroom, revealing how AI “thinks” and what happens behind the shiny interfaces users interact with daily without understanding the internal mechanisms.

Between Admiration and Criticism

The article narrates how AI has become an amazing tool to accelerate work and analyze huge amounts of data, but at the same time raises growing concerns due to the opacity surrounding its operation. The companies producing these technologies do not allow users to access the details of the data that trained their models, nor do they allow understanding of how decisions are made or answers generated.

The author compares the relationship with AI to a washing machine: we can use it without understanding how it works. But the difference here is that what we hand over to AI is not dirty clothes, but our ideas and creative and intellectual output.

Race Towards “Artificial General Intelligence”

The article criticizes what it calls the “race of big companies towards artificial general intelligence,” a model capable of performing most human tasks more efficiently than humans. It points out that the definition of this term by some industry leaders — such as OpenAI CEO Sam Altman — focuses more on economic value than on the essence of “intelligence” itself.

The article argues that such a definition reduces AI to a tool for increasing profits rather than a means to understand the nature of human thinking. Is intelligence merely the ability to achieve financial gains? The article poses this question to bring the discussion back to its deep human meaning.

Inside the “Machine’s Mind”

The article explains in a simplified way how large language models work internally, clarifying that they do not “think” like humans but process language through mathematics and statistics.

Words are converted into numerical codes known as “vectors,” arranged within a multi-dimensional space where the model can measure relationships between them. Through this digital representation, AI learns to predict the next word in a sentence, giving it the ability to produce coherent and natural texts.

To simplify the idea, the article gives an example from English poetry, showing how the model can perceive connections between words like “sweet” and “sour,” and how meaning changes if directions in this digital space shift.

Small Model Experiment

The article reviews a previous experiment by the author in 2017 when he created a simple language model to imitate the style of American writer Herman Melville in his famous story Bartleby the Scrivener.

The model started writing incomprehensible sentences, then gradually evolved until it almost repeated the story’s famous phrase “I would prefer not to,” but instead wrote “I would prefer to yes.”

The article sees in this error a deep human significance; even in its moments of failure, AI reflects a strange imitation of human nature — that fallibility that distinguishes us from machines.

A Call for Transparency and Understanding

The article concludes with a clear call for more transparency and openness in the field of AI. It suggests that companies allocate a small portion of their massive computing investments to explain how their models work, make training data publicly available, and provide interactive environments where users can adjust parameters and see results for themselves.

The article believes that opening the “digital mind” to the public not only builds trust but also gives people a chance to understand this technology that increasingly influences their creativity, decisions, and even consciousness. According to the article, AI is not just a commercial product but a cognitive experience touching the essence of what it means to be human in the technological age.