2024-09-23 16:39:41
Faced with the current “tsunami” of applications, “big data”, algorithms and “bots” based on artificial intelligence, there is a risk of polarizing their adoption: either everything goes well, or rejection due to saturation.
November 2023. A very high percentage of medical conferences, symposiums, seminars and all types of forums deal with or, at least, include some presentation on artificial intelligence (AI). It is undoubtedly the topic of the moment. “Even those in hydrology talk about artificial intelligence, which displeases hydrologists,” said an expert on the subject in one of these forums; It is called the sin, but not the sinner.
“We are experiencing a wave of AI-based solutions, not particularly economical, so it is necessary to demand, evaluate and prioritize, both on the part of the hospital and, for example, the administration,” said Marcos Hernández, deputy director doctor from the Hospital Severo Ochoa, in Leganés (Madrid), during the last Ideas Incubator at the San Carlos Clinical Hospital, in Madrid, dedicated to Generative Artificial Intelligence in Healthcare.
And do we really know what AI is? What is it for? Also, does it work for everything? “If we could travel back in time and bring a calculator to classical Rome, they would think it was artificial intelligence. Or if we brought the most basic PC to the beginning of the 20th century. Until now, artificial intelligence was every technological challenge that, once achieved, stopped being artificial intelligence,” says Ignacio Hernández Medrano, neurologist, founder and medical director of Savana, a company that has been creating artificial intelligence solutions for almost ten years for hospitals and healthcare services. years.
“AI has been around since the 1950s, but it’s talked about so much now because there is a definition of what it is and because of the rise of generative AI. Furthermore, every time something becomes fashionable, everyone wants to jump on the bandwagon, only to fall under their own weight,” explains Roberto Menéndez, digital CEO of Grupo ADD-Futura Vive, a company specialized in the development of humanoid robots. For this reason he recommends that hospitals and health services “have experts capable of advising and distinguishing what is useful and what is not”.
Another important factor for this boom is the recent past: “We are witnessing a real revolution in the development of medicine at the hands of AI. We have witnessed a before and after with the covid-19 pandemic in relation to the expansion of digital in our lives, which together with the disruption brought by the accessibility of ChatGPT, has formed a historical context that is facilitating these technological changes on course of neck speed,” says José Antonio Trujillo, vice president of the College of Doctors of Malaga and author of the book Basic Guide to Artificial Medical Intelligence.
Learn like a child
Although he acknowledges this somewhat arbitrarily, Hernández Medrano states that today “we have decided that machine learning is considered AI (machine learning)”, made possible by the availability of enormous quantities of data (big dataconcept which, he claims, is becoming obsolete compared to data lakes).
This artificial intelligence, called discriminative, is based “on pattern identification, as opposed to traditional rule-based programming.” Machine learning, in this sense, is like the way a child learns: once he understands a concept (for example the table), he is able to identify it (even if it is in front of a table he has never seen before ).
Therefore, artificial intelligence has, by definition, the ability to learn naturally, excluding from the equation those traditional alarm systems that detect patterns based on fixed rules, i.e. assuming that a table must have a board and four legs.
Its evolution developed in layers, from the simplest to the most complex, depositing itself on lakes of data (data lakes) type onion: images, structured data (such as laboratory data or CIE classifications), natural language processing, telemetry (sensors and wearable), omics and, finally, environmental variables. “We’re all evolving pretty much the same way, trying to create it data lake definitive healthcare,” says Hernández Medrano.
What ChatGPT says
And now that this area appears to be mastered, comes generative artificial intelligence, which “refers to a type of artificial intelligence (AI) that focuses on creating original content, such as images, music, text, or even video. It uses artificial intelligence models to generate data that imitates or resembles real data, although it is not directly based on existing data”, a definition which, as it could not be otherwise, was generated by Chat-GPT, which considers itself to be the same as more than a bot: “I can answer questions, generate text, and perform machine learning-based tasks, but my goal goes beyond just generating content,” he says.
As if it were a Doctor Frankenstein, generative artificial intelligence takes real models – or pieces of real models – to create something new and with the thinking capacity of a supercomputer: according to Trujillo, a team of medical researchers led by the doctor and Harvard University computer scientist Isaac Kohane put ChatGPT-4 to the test in Medical licensing exam in the United Statesthe standardized testing program for medical licensure in the United States. It consists of three phases that cover all topics in doctors’ knowledge, from basic sciences to clinical reasoning, medical management and bioethics. “To the researchers’ surprise, the AI model proved to be better than that of some licensed doctors in many cases. According to their study, GPT-4 was able to answer exam questions correctly more than 90% of the time.”
Fortunately, he was not granted a license to practice medicine. And we cannot forget that the brain of an artificial intelligence thinks in zeros and ones, and that its memory, although almost infinite, is devoid of real reasoning: “These developments can be very useful for repetitive or less valuable jobs, being able interact with users – patients and professionals – with a natural language, as if speaking to a person, but only to help the doctor focus on what he needs to do,” explains Menéndez, whose company has developed a chatbots with generative artificial intelligence that helps triage in the event of an emergency.
As quickly (or almost) as it has penetrated society (teachers have returned to taking exams due to the difficulty of knowing whether the work is done by students or by an artificial intelligence), “generative artificial intelligence will penetrate in hospitals; It is great, for example, for documentation tasks,” says Hernández Medrano.
But he issues a warning: “The risk is thinking that it is as reliable as the discriminative one, to which we are more accustomed.” The problem with generative AI is that because it has the ability to create, its results cannot be validated machine learning they are much more square.
In fact, Savana’s CMO believes that “the current paradigm shift in artificial intelligence in medicine is the validation of algorithms as if they were drugs”. The US FDA already does this and has over 250 validated algorithms for which a clinical trial has been requested. “In Europe, only the CE marking is still required, which is granted with retrospective data, but there are already countries that require clinical trials to validate the effectiveness and safety of the systems.”
And privacy? Hernández Medrano is firm: “Anonymized data is no longer personal data. The immoral thing would be not to use it.” JGS. SJD
#artificial #intelligence #isnt #Health #Medicine