You search Google for "criticism of artificial intelligence" in Spanish. The first ten results are signed, without exception, by brands that sell AI, consultancies that advise on AI or universities that recruit tuition by selling AI master's degrees. Three categories that live off the very business they claim to criticize. Those who really know —Crawford, Zuboff, Gebru, Hinton, O'Neil, Bostrom— don't write in Spanish, don't do informational SEO, and therefore don't exist for the Spanish-speaking user who opens the browser with an honest question. What that user reads is written by whoever has a direct interest in how they read it.
What you call "debate" is something else
Before going on, a technical word worth clarifying because I'm going to repeat it a fair bit. SERP stands for Search Engine Results Page: the results page Google returns when you search for something. The first ten links are the ones most people open. The rest, almost nobody. Whoever occupies those ten positions occupies, in practice, the entirety of the public debate on that matter for non-specialists. The SERP isn't a shop window: it's the full contour of the mental territory any ordinary person moves through when they wonder "this AI thing, what's wrong with it?"
And that territory, in Spanish, is colonized. You type "risks of artificial intelligence" and the first results are IBM with its listicle of ten dangers, some consultancy with its own version of the same ten dangers, a private university selling a master's with a blog article that also lists ten dangers, and a cybersecurity firm recycling the same material with the word "cybersecurity" stuck to the headline. You type "disadvantages of AI" and the same cast appears with slight chromatic variations. You type "ethical problems of AI" and the protagonists are the same as always. Judge and party. All living off the sector. All with the same format.
There's no debate. There's occupation.
The listicle factory
Maybe it's worth looking at how a "10 dangers of AI" listicle gets manufactured before going on. Because the format isn't innocent; the format is the content.
You take a template. The figure —ten, five, eight, seven— is set by whoever writes it according to what fits without the page getting so long it scares people off. The bullets are interchangeable because they're written in the abstract: "algorithmic bias," "job loss," "privacy," "deepfakes," "disinformation," "Skynet." None lands on a real case. None cites a study with figures. None explains what to do when two of the ten dangers come into contradiction —and they come into contradiction, all the time: privacy vs security, automation vs employment, transparency vs efficacy. The template avoids exactly that question because it would require thinking. And thinking isn't scalable. Listing is.
The most comic part is the pretension of neutrality. The IBM listicle doesn't end by saying "by the way, we sell AI, so take it with a pinch of salt." It ends with a link to "discover our solutions." The consultancy does the same. The university too. The listicle isn't an analysis with a conflict of interest. It's advertising disguised as an analysis with a conflict of interest. The rhetorical operation consists of sowing just enough fear for the reader to need the solution the same sender sells immediately below.
It's marketing. Bad marketing. But it arrives so clean, so well stuffed with bullets, so balanced in its false impartiality, that the reader closes the tab convinced they've read something serious. There the SERP ends. There ends, in many cases, the opinion that person will form about AI for years. Meanwhile, the sector keeps working.
What doesn't appear and what does
Here's the list of people who have produced solid thought about the real problems of artificial intelligence. Kate Crawford, author of Atlas of AI (Yale University Press, 2021), where she describes AI as a material system, not an abstract one: lithium mines, server farms, precarized annotators in Kenya, cooling water in drought-stricken zones. Shoshana Zuboff, author of The Age of Surveillance Capitalism (PublicAffairs, 2019), where she describes how the data economy turned human behavior into raw material for selling predictions. Timnit Gebru, co-author of On the Dangers of Stochastic Parrots (FAccT 2021), the paper Google fired her over and that remains one of the most cited academic critiques of the large-scale model. Geoffrey Hinton, Turing laureate, father of deep learning, who left Google in May 2023 to be able to speak without filters about the risk of the systems he himself helped build. Cathy O'Neil, author of Weapons of Math Destruction (Crown, 2016), where she documents how opaque algorithms systematically harm the people with the fewest resources to defend themselves. Nick Bostrom, author of Superintelligence (Oxford University Press, 2014), required reading for understanding the alignment debate.
Count how many appear on the first page of Google when you search for criticism in Spanish. Zero. Not one mention. Not one link. Not one translation of their key texts in a visible position. Their domains don't compete because they don't do informational SEO. Their books are in English, badly distributed in Spanish, in university libraries almost nobody sets foot in. Their papers are in archives and indexed journals the average user doesn't open. They exist, but they exist in a layer of the world —the Anglo-American academic one— that doesn't cross the layer in which the Spanish-speaking reader looks for answers.
What does appear, we've already seen: IBM, Microsoft, Salesforce, Telefónica, BBVA, Cesce, Pirani, Sentinel, Avast, VIU, UNIR, IEBS, ISDI, ESADE. SaaS companies, insurers, risk consultancies, cybersecurity platforms and private universities with a recruitment department. All with a direct incentive for the reader to leave with a specific mix of fear and dependency: enough fear to feel they need professional help, enough dependency to believe that professional help is given by them.
When methodology is missing, marketing is in excess
There's a difference between criticism and criticism. The criticism worth reading meets some minimal requirements: it starts from concrete data, cites its sources, distinguishes between what's happening, what could happen and what's fantasy, owns its own biases and proposes ways to measure what it claims. It's what in classic terms we call method. Without method there's no knowledge, there's decorated opinion.
What occupies the Spanish SERP has no method. It goes from "AI is going to revolutionize work" to "AI could end humanity" in three lines. It mixes risks at ten years (existential risk) with everyday risks (algorithmic bias in a loan) without distinguishing temporal scale, probability or impact. It uses generic figures ("60% of jobs will be replaced") without linking to the study they come from, because linking it would force an explanation of methodology, sample, assumptions. It cites respectable bodies (OECD, UNESCO, MIT, Stanford) without going into what those bodies say exactly —and often without understanding what they say.
We're not at the center of a scientific conversation here. We're at the center of a marketing stage where enthusiasm and fear are both bought and sold at once. The enthusiasm is sold by whoever wants you to pay for the subscription. The fear is sold, by exactly the same company, in another tab, so you pay for the training course that will "protect" you. Fear marketing and enthusiasm marketing aren't opposing stances. They're the two arms of the same body. If you see it, you stop buying the trick.
Serious criticism exists, but doesn't reach
Stanford has just published, in April 2026, its annual AI Index Report. It's the most complete radiography of the sector we have today. It talks about adoption (53% of the world's population uses generative AI, a speed greater than the PC's and the internet's), about global investment (581.7 billion dollars in 2025, double the previous year), about the closing gap between the United States and China (Anthropic barely 2.7% ahead of the most powerful Chinese model), about documented incidents (362 in 2025, up from 233 in 2024). It's hard information, with methodology, accessible, in a consultable format. How many articles in Spanish have translated that report with serious analysis? Barely a handful in specialist outlets and almost always as a literal translation of the Anglo-American headline. Whoever searches "AI Index 2026" in Spanish finds three-paragraph summaries and back to the listicle.
The operational difference is brutal. A reader who wants to understand what's really happening with AI in 2026 has two options. One: read in English Ben Thompson at Stratechery, Casey Newton at Platformer, the long reports of 404 Media, the technical analyses of MIT Technology Review, the annual reports of Stanford HAI, the papers of FAccT and NeurIPS. The other: read in Spanish the IBM listicle and learn that AI can be dangerous because "there could be biases" and that the solution "involves implementing governance policies," without that sentence meaning anything concrete.
It's not centrism to say there's an editorial problem. It's description.
The fauna of listicles: a taxonomy with no desire to draw blood
Allow me an H2 with a little more sarcasm, because the material deserves it. If you review the sector methodically, the "10 dangers of AI" listicles can be classified into families. There's the bank listicle —BBVA, Santander— that places much emphasis on data privacy because the bank sells products where privacy is a lever of trust, and much less emphasis on corporate concentration because the bank itself is concentrated. There's the consultancy listicle —Deloitte, KPMG, EY— that emphasizes "implementation risks," precisely the risks for which the consultancy sells service hours. There's the private-university listicle —UNIR, VIU, IEBS, ISDI— where the dangers are stated with pedagogical prudence because the article's close is always "find out about our master's in AI and ethics." There's the cybersecurity-firm listicle —Avast, Malwarebytes, Sentinel— where all the dangers turn out to be, what a surprise, cybersecurity dangers solvable with their product. And there's the tech-giant listicle —IBM, Microsoft, Salesforce— where the danger of AI is always something abstract resolved by adopting "responsible policies," which those same giants happen to offer as a service.
There's no conspiracy theory here. None is needed. Each company optimizes its content for its sales funnel, the same way each animal optimizes its behavior for its ecological niche. What there is, is an ecosystem. The ecosystem produces an aggregate image of the sector biased in a concrete direction —the sector's own— because the only animals living there are the ones with an incentive to live there. The rest don't appear. Crawford doesn't appear because nobody pays Crawford to appear. Zuboff doesn't appear because her book has seven hundred pages and no SEO team. Hinton doesn't appear because after leaving Google he gives long interviews, not listicles.
And who's going to read Crawford on a Saturday afternoon?
Here's the final trap, and it's worth naming without moralizing. Serious criticism doesn't reach not only because it doesn't do SEO. Also because it demands of the reader more than the average reader is willing to give. Atlas of AI has 336 pages and a dense academic apparatus. Surveillance Capitalism has 704 pages and two centimeters of footnotes. Weapons of Math Destruction is more accessible but demands patience to understand basic statistics. Superintelligence demands the reader tolerate the vocabulary of analytic philosophy. None of these books reads in fifteen minutes over the mid-morning coffee. None produces reassuring certainties or doses of adrenaline with a striking headline. They produce understanding, which is another thing, and understanding is charged in hours.
The listicle, by contrast, offers the simulacrum of understanding in four minutes. Ten bullets, a circular conclusion, two internal links to related content from the same sender, and you've got an opinion formed. You can go to lunch with the feeling you've thought it through. You haven't thought it through. You took it prefabricated from a site that had an interest in you taking it prefabricated.
That's the trap. And breaking it doesn't require becoming an academic reader. It requires, simply, replacing the first reading. Changing the order. Before searching "10 dangers of AI," read a long interview with Crawford. Before searching "risks of ChatGPT," read a chapter of Zuboff. Before searching "disadvantages of AI," read the whole AI Index —in English, yes, with a translator if needed— and come back with concrete questions. The SERP changes when what you search for changes. And what you search for changes when you stop accepting the frame the SERP itself offers you.
The problem isn't what IBM says, it's who we let speak
The public conversation about AI in Spanish isn't biased. Biased would imply there's a balance to correct. What there is, is something stranger: a market full of vendors where the reader has been convinced they're reading independent press. The operation is elegant. When IBM publishes "10 ai dangers and risks" —that's literally the slug of the original post in English— and positions it at the top of critical searches, it isn't lying about the dangers. Some dangers are real. What IBM does is choose which dangers are visible and which aren't. Which get mentioned and which get omitted. Which need "solutions we offer" and which would require structural reforms of the sector in which IBM would lose.
That selection is the real content. Not the bullets. The selection.
When the entire ecosystem is dominated by vendors doing selection, what comes out aggregated isn't information: it's advertising. Well written, balanced in appearance, full of surface nuances, but advertising. The reader who doesn't see it leaves convinced they already know what there is to know. The reader who sees it discovers they have to go looking elsewhere.
That elsewhere exists. It's in English, it's in long books, it's in academic papers. It's outside Google. And as long as nobody does the work of bringing it into Spanish with the voice and the depth it deserves, the Spanish-speaking sector will keep thinking about AI exactly what the companies that sell it want it to think.
Definiciones rápidas
- SERP: Search Engine Results Page. The results page Google returns for a search. The first ten links concentrate almost all the user's attention and determine, in practice, the public debate on a matter for non-specialists.
- Listicle: an article in numbered-list format ("10 dangers," "5 disadvantages," "7 problems"). A format inherited from easy-traffic online journalism, now adopted en masse by corporate SaaS marketing for its high SEO performance.
- Informational SEO: organic ranking on non-commercial searches ("what is," "how it works," "risks of"). It's the key position for influencing how public opinion forms about a sector.
- Editorial conflict of interest: a situation in which the sender of an analytical piece has a direct economic interest in the direction the reading of the piece steers the reader. It defines the current ecosystem of Spanish-language content about AI.
Referencias
- Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 336 pp. https://yalebooks.yale.edu/book/9780300264630/atlas-of-ai/
- Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs, 704 pp. https://en.wikipedia.org/wiki/The_Age_of_Surveillance_Capitalism
- Bender, E., Gebru, T., McMillan-Major, A. & Mitchell, M. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? FAccT 2021.
- O'Neil, C. (2016). Weapons of Math Destruction. Crown.
- Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press.
- Hinton, G. (May 2023). Interviews in The New York Times and The Guardian after his departure from Google.
- Stanford HAI (2026). AI Index Report 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report
- IBM (2024-2025). 10 AI dangers and risks and how to manage them. https://www.ibm.com/think/insights/10-ai-dangers-and-risks-and-how-to-manage-them
- IBM (2024). AI risk management. https://www.ibm.com/think/insights/ai-risk-management
También te interesa
- How public conversation about AI broke
- Why reading IBM on AI is a waste of time
- The real dangers of AI
No comments yet
No comments yet. Be the first.
Leave a comment