I am writing this article at about 3 am, sitting in the lobby of the clinic and my friend is talking to an oncologist in the room behind me. The air in the lobby is definitely filled with the smell of medicine and there is an undeniable sadness in it that makes me want to leave this building as fast as possible. However, the feeling that has overcome me the most is intense anger and rage.
According to ZoomTech technology news section, quoted from Digital Trends.It was just an hour ago that I received a call from my 24-year-old friend; He complained of severe headache and unbearable heartache. He is currently undergoing treatment breast cancer and after seeing the suffering and hardships he had gone through in the past two years, I immediately went to his house, hoping that the problem was not serious. Without asking any further questions, I searched for the nearest hospital and took him for an emergency check-up.
On the way to the hospital, I asked him when exactly the pain started and if he had any other worrisome symptoms. The answer he gave shocked me greatly; So much so that I wanted to snatch his phone from his hand and break it into pieces. “After coming back from work, I felt a little dizzy and felt that my stomach and intestines were not going well,” he said. That’s why from ChatGPT I asked if I could get some accommodation. And a few hours after taking the medicine, my heartache worsened with a strong burning sensation and I vomited several times.”
Who is responsible for this fatal error?
I was deeply moved by his confession, and his completely cool and casual tone in describing his interaction with artificial intelligence (AI) It was even scarier. This event has many dark dimensions and layers, to the extent that I decided to use artificial intelligence for all my loved ones, especially in the context of conversations related to health and hygieneI forbid completely. For someone who is extremely allergic to drugs and is going through a tough course of cancer treatment, asking an AI to prescribe medication is the last thing you would expect a sane human to do.
I don’t know exactly who is to blame for this dangerous and life-threatening incident. My friend very simply took a picture of a painkiller and imported it ChatGPT and asked if he could take two pills at the same time or not. His attending physician had prescribed only one pill at a time, and only if the pain was unbearable. In response, the chatbot informed him that he should take only one painkiller and that taking two of them is allowed only with a doctor’s prescription.
When I asked him why he went to one Artificial intelligence chatbot He replied, “It’s easier to get an answer from ChatGPT.” “I have no idea what’s real on the internet.” For a person who had to drop out of school after finishing high school and pay for his family by doing various jobs, I understood his situation and hardship. I could not tell him about the concepts Digital literacy and verification of lecture information. There were many mistakes in that chat and both sides were guilty.
Translation crisis and memory error of large language models (LLM)
The problem starts right here. His mother tongue is not Hindi and when he translated his application into Hindi, a grammatical disaster occurred due to changing the meaning of one word. He wanted to ask if he could eat “two numbers” of the medicine in the picture; But his broken translation made the chatbot think he was asking if he could take “both drugs” at the same time!
It was here that Linguistic model of artificial intelligence It was also wrong. In the picture sent, a pill could be seen, the only chemical ingredients of which were “Diclofenac” and “Serraptiopeptidase” visible, and the brand name of the drug was only partially visible. However, the AI correctly identified the housing as something similar to Cipzen-D; But before that, he also mentioned another drug called “Fenak Plus” (Fenak Plus), which has a combination of diclofenac and paracetamol, and my friend had already talked about it with Bott in another conversation.
The two drugs have completely different formulations, although both are available as over-the-counter (OTC) pain relievers and, alarmingly, can even be purchased without a prescription from local pharmacies. So in a general view, we are dealing with one human error (in translation) and two errors caused by extracting memory data by ChatGPT We were facing However, Butt’s final recommendation was sound in theory.
Company chatbot OpenAI It clearly warned that taking both drugs at the same time is not the right decision and can lead to diclofenac overdose. The artificial intelligence described side effects including severe burning sensation in the stomach, stomach ulcers, bleeding and long-term damage to the kidneys.
But remember the photo that to ChatGPT It was given that it only showed one housing. The system suggested that only one of these painkillers should be consumed at a time, not both. But my friend, remembering that both drugs had been prescribed by doctors at some point (though not together), took them both for quick pain relief and soon after suffered terrible consequences.
The anger of the treatment staff and the immediate removal of artificial intelligence chatbots
When the attending physician found out why she had taken two pills at the same time, she was quite rightly angry and scolded her for not consulting a specialist. Next, drug injections were given to relieve her pain and a new medical prescription was prescribed for painkillers that would not harm her weak and fragile body.
As a reporter in the field of technology, I often talk about The dangers of artificial intelligence I had written on sensitive topics such as health and medicine, I felt a complete failure. I felt that I had never seriously discussed the dangers of using these chatbots with my friend, and it was only after this horrific incident that I warned him.
When he admitted that in the past also several times for Medical advice from ChatGPT used, I took his phone and deleted this program. More importantly, I asked him to never go to any other chatbot like that again Gemini or Claude Don’t go for medical advice and basically don’t rely on the internet for these things. The only way to save and do the right thing is to talk directly to a doctor or go to a clinic.
crisis of public perception of artificial intelligence; A giant magic lamp or fake encyclopedia?
I don’t blame him completely for this incident. According to the way ChatGPT As an omniscient person embedded in the daily digital habits of the masses, it is not easy to blame people. People see it as a much easier (and more honest) response engine than the hassle of searching. Google search engine (Google Search)They look at websites, click on blue links and read long texts.
To be honest, they are right. Getting straightforward answers from a natural language chatbot in just a few seconds is much easier than wandering through the maze of Google search links. Google is also well aware of this change in user behavior; An issue that justifies the intense focus on adding conversational AI modes to the search engine.
Currently, conversational responses provided by Gemini are provided, placed at the top of the normal search results. The idea is the same: to provide the user with summarized answers before he has to open dozens of different tabs to search for information in depth.
It should also be acknowledged that ChatGPT (and most leading AI models) typically act cautiously when someone asks for medical advice. In the vast majority of these responses, the chatbots ask the user to consult a licensed professional or contact the relevant helplines. But again, the way these chatbots entered the public opinion as a fast and reliable resource is what makes them so terrifying.
A few days ago, I was watching a heated debate on a television news network. During this tense discussion, one of the experts asked his rival to verify the claims ChatGPT Go! At that moment, I saw that ChatGPT was mentioned just like the modern version of an authoritative encyclopedia; That too in a medium that affects millions of audiences in the world’s second largest artificial intelligence chatbot market.
Linguistic gap and error of understanding non-English texts in artificial intelligence models
The United States leads the way in AI adoption, but also has the highest level of public awareness of the dangers of using chatbots. In countries like India, digital literacy (and awareness of the dangers of AI) is far lower, and thus the risk potential is much higher. This risk is exacerbated when hundreds of thousands of people from ChatGPT and other artificial intelligences use their native languages. This was one of the main factors behind my friend’s story: Language comprehension problem in ChatGPT.
In a research project published by the MIT Center for Constructive Communication, it was found that The accuracy of answers provided by AI chatbots in non-English languages is greatly reduced.. The SAHARA test benchmark also revealed a similar significant difference in AI performance when interacting in non-English languages. Journal The Economist also covered several articles that pointed out the structural weakness of large language models in understanding non-English languages.
Language barriers greatly increase the risk factor. However, translation is not the only challenge. The watchdog Healthwatch England, an independent group that supports the rights of patients in the UK public health system, about the risks of doctors using Artificial intelligence transcription tools (AI Transcription) has warned In one of the alarming reports, it is stated:
“In one alarming case, an artificial intelligence tool reviewed a summary of an MRI scan and recorded that the patient had demyelination (a serious neurological complication associated with MS); While the scan actually showed “Null Demyelination”, meaning there was nothing wrong!
The need for an urgent review of the medical warnings of companies developing artificial intelligence
This is an obvious shortcoming that large language models should address, but there is also a larger issue at play: AI chatbots need to be much smarter and more cautious when dealing with sensitive topics. Platforms like ChatGPT They have created parental control systems and suggest users seek medical help and even provide hotline numbers. You will also see a message saying “ChatGPT is not a doctor”.
But these warnings should be exactly in The highest chat point (big banner) be placed, not buried at the end of paragraphs; Where most users get their short answer in the initial lines and don’t bother to read the rest of the text. This safety protocol should be institutionalized in the deepest behavioral layers of AI models.
In addition, we need a very broad awareness. Tech companies are constantly promoting how Large Language Models (LLM) They solve complex math problems, beat humans in Olympiads, and contribute to medical advances. But all these achievements are made under the direct supervision of experts. For the average person for whom ChatGPT is a quick, conversational source of knowledge, displaying disclaimers and health warnings should be a top priority.
The OpenAI company may think that it has taken a cautious policy, but the public opinion towards ChatGPT It is evolving at an alarming and dangerous speed. Many users look at it like a mastermind and a know-it-all. There have been many reports of very dangerous treatment recommendations from these tools, and research by researchers continues to prove risky AI behaviors in sensitive areas. One can only hope that OpenAI and other tech giants take these warnings seriously and take decisive action before larger human catastrophes occur.
گفتگو در مورد این post