AI: Hype for Some, Labor Revolution for Others

AI lacks understanding, yet it already replaces jobs

One interlocutor replied in italics, with impeccable bullet points and textbook paragraphs. Several panelists responded as if it were a person. It was ChatGPT. The episode captures the core debate: is artificial intelligence an ongoing technological revolution or a bubble? Some argue it merely fuels the leisure of young people.

Two years after a conversational model went mainstream, the discussion has shifted from whether machines can write well to whether meaningful human work will remain. Arguments exist on both sides.

Why do some call AI just a word-stitcher?

The harshest argument does not dispute the results: it disputes the word intelligence. AI lacks any 'I', summarizes one viewpoint. What is offered today is a word-stitcher boosted by statistics, without deep or superficial understanding, only linguistic co-occurrence over massive corpora.

With this premise, examples multiply. A system fails to grasp the irony of someone arriving late from work «and not temporally, precisely»; nor the joke of whether one stops in space or time first. Labeling an image is not comprehending the story behind it. A panelist recalls that Hubert L. Dreyfus made similar arguments in the 1990s, which he believes remain valid.

Thus, some diagnose: it is not that 'deep' understanding is missing, but that there is none. The model orders probabilities. It is often right, which is not the same as knowing.

Where AI already replaces technology and employment

Against conceptual skepticism, there is a catalog of uses that are no longer demonstrations. Video processing (resizing and frame generation), cybersecurity, and big data analysis. Panelists maintain that mid-sized companies have integrated these tools for productivity, and those who do not master them will have less room.

The most cited example is not literary but scientific: an AI has discovered how proteins fold, a problem stalled for decades. In the military field, some claim Russia already uses drones supported by these systems to observe targets in real time. And in the office trench, a thousand-page English technical manual ceases to be a wall: it is queried in natural language and responds.

Here lies the uncomfortable detail. The tool may be a giant parrot, and yet still be a parrot that saves hours. The full breakdown of that saving, item by item, is what turns the discussion into something with real consequences.

Will 60% of jobs disappear?

Some put a figure on the threat: the imminent end of 60% of workers. The projection is not supported by data, only by analogies.

The classic counterargument is historical. The IT boom was supposed to decimate employment and did not; the author of this text started with an Amstrad 128. The tricky accompanying question is simple: if technology destroys jobs, why is there more unemployment in Mauritania than in Japan.

Other voices downplay the drama: they are tools that increase productivity but require human experts to validate the result. With a nuance that reassures no one: when its use becomes general, the price of that report will tend to zero.

What an AI cannot smell or remember

Here the argument leaves technique and enters biography. Someone aged 55 recounts that a smell, a sound, or an image returns them to their nursery school desk. The smell of DACS crayons, Iter Sopena dictionaries with flags and anatomy sheets, children looking up forbidden words to laugh.

No machine knows what that smells like. It lacks its own memories, shared memories, and emotion applied to both. Perception —another intervention suggests— is not accumulating data, but seeing invisible relationships for the inner eye. That, if it exists, precedes data.

One can debate whether all that is needed to call it intelligence. What is not debated is that some maintain it will die without seeing a system with the curiosity, emotion, and learning capacity of a ten-year-old child.

The bubble, energy, and printed money

The final front is financial. The energy consumption of these systems is high and the bill, uncomfortable. The moment is compared with other recent promises —the metaverse, which lasted less than a trend— and with technologies that were supposed to cure everything, such as decoding the genetic code or the electric car.

The bitterest reading holds that all this is a market niche to move newly printed money and prevent the shanty town from collapsing. The pragmatic reading responds that a bubble and a useful technology are not mutually exclusive: one can inflate stock prices and, at the same time, change the way of working.

In between lies a disconcerting detail. As early as 1994, one could query a database written in Prolog using natural language. Thirty years later, the question remains the same: if it orders words with statistics, is it intelligence or plumbing?