Does the intelligence of AI make waters? No, we just believed it more list than it really is

In recent months, multiple news emerged to the surface that would be realizing an alleged cognitive impairment of AIthat would be supposedly at the mercy of increasingly hallucinations. Only a few weeks ago it was, for example, news that target AI, the IA imbricated in the entrails of WhatsApp, would have inadvertently shared a user’s telephone number with a complete stranger.

Similarly, «AI Overviews»the summaries generated with AI that open their way in the Google search engine, they also monopolized holders some time ago for committing striking dislates such as recommending the addition of glue to pizza.

The courts have also witnessed the folly of AI. In a trial in which a client made the determination to sue the Avianca airline for being injured during a company’s flight their lawyers decided to rely on Chatgpt to shape the complaint. However, the lawyers did not bother to check whether the information on the table by the OpenAi chatbot was really real, the judge was aware of the error and ordered them to pay a mule of 5,000 dollars.

Just a couple of months ago the newspaper The Chicago Sun-Times published a list of books recommended for the summer that had clearly prepared with the chatgpt support, then In the list of marras there were fully invented book titlesa yerro who managed to put the American newspaper.

Even when AI evolves by leaps and bounds, the truth is that The number of users who lament the errors (often monumental) committed by chatgpt and company seems to increase at times. This is evident at least numerous articles and publications on social networks.

And user complaints would not be unfounded. According to tests undertaken by OpenAi, O3 and O4-mini, the latest AI models of the Chatgpt matrix, are victims of hallucinations in 50% of cases. And a neighborhood study also concludes Some AI models seem effectively hallucinating more than beforeAlthough it suggests that their errors obey training and not so much in the system of reasoning of such models. But are Ia models be shared for a kind of cognitive impairment? Is this type of deterioration in AI possible?

To the AI ​​we have arrived some “super powers” that do not actually correspond to it

Experts suggest that the cognitive deterioration that AI models would currently be suffering is not reality And that their hallucinations are actually led by the data with which such models are trained, which are sometimes outdated and can be equally biased and false. “Hallucinations in AI will become less frequent as their reasoning capabilities with spoiled training systems progress for more precise and open source information,” explains Daniel Keller, CEO and co -founder of Influex Technologies, in statements to HuffPost.

Raj Dandage, CEO and founder of Codespy AI, admits that AI is suffering a slight deterioration in the cognitive plane, but believes that This deterioration is more conspicuous in models widely used as chatgpt, which are being orphaned with real quality data for their proper training.

For its part, Binny Gill, CEO of Kognitos, considers that The failures of AI are to a greater extent debtors of the humans of flesh and blood than of AI itself. After all, AI does nothing but nourish data that carry the rubric of people and that in this sense are not in any way immune to errors.

“The AI ​​does not suffer any cognitive deterioration because if it really suffered it, we would also be assuming that at some point it was intelligent and not the case”asevera Gill.

From Gill’s words it is inferred, therefore, that we have endorsed some “super powers” that he doesn’t really have. And although in the eyes of many AI tools they are a boat soon the epitome of intelligencewe often forget that this intelligence is an orphan of reasoning (and cannot, therefore, discern the truth of the lie).

It also seems that AI not only is not as intelligent as it seems, but its use is also making more silly to those who have it well to use it. A study recently published by the Massachusetts Institute (MIT) concluded, without going any further, that the use of AI could translate into a decrease in cognitive activity and be harmful to the dessert for the development of truly critical thinking.