What surprises me most about artificial intelligence is not what it has achieved so far, or what it will be able to do. What leaves me speechless is how people, companies and institutions They embrace a first-generation AI full of bugs. And while that’s trivial when you ask ChatGPT any nonsense, it baffles its use in critical tasks such as a trial, a medical evaluation, or a police report.
A few days ago a police report in the United States came to light, which stated that an agent had turned into a frog. The AI analyzed what the police camera captured, the movie was being broadcast on TV Tiana and the toadand the AI came to that incomprehensible conclusion.
More serious has been what has happened in the county of West Midlands, in the United Kingdom, where fans of a soccer match in the Europa League were prohibited from entering, because The AI accused them of causing destruction in a match that never existed. Nobody bothered to check it.
AI is a liar by nature
Although it is being corrected with the latest AI models, it is known that artificial intelligence prefers to lie, rather than admit that it does not have the answer. And if it sneaks, it sneaks. Unfortunately, this often happens and we miss it.
On November 6, a Europa League match was played between Aston Villa and Maccabi Tel Aviv of Israel. A police report described the match as “high risk,” so the police prevented fans of the Israeli team from entering the stadium.
After the match it emerged that the report was based on incidents involving Israeli fans in the match between West Ham and Maccabi Tel Aviv… that has never taken place.
How can the police invent a party? All eyes were on AI, but West Midlands Police Chief Craig Guildford He denied that his department uses AI to generate the intelligence reportand blamed the error on “scraping of social networks.”
However, last week the police chief had to change his statement: “On Friday afternoon I realized that the erroneous result regarding the match between West Ham and Maccabi Tel Aviv was due to the use of Microsoft Copilot.” Oh my God.
AI companies cover their backs by saying that their chatbot can make mistakes, and the results need to be checked. But if you use AI to save searches and then have to do a search to check those results… What good is AI?
A police report created with AI has prevented fans of a football team from entering the stadium. It’s not the end of the world, but what if AI generates an arrest report for an innocent person or, on the contrary, hides or confuses facts to eliminate someone who is guilty as a suspect?
Which brings us to the first question: Why are critical tasks entrusted to a first-generation AI that recognizes that it fails, and does not take responsibility for errors?