Programmed by computer engineers in the late 20th century, AI was born based on a set of instructions (rules) created by humans, allowing technology to solve basic problems.
Editor’s note: There are many industries that are affected by the new technology in the information age. With the impact of automation, computer science, artificial intelligence (AI), subjects such as doctors, hospitals, insurance companies and industries related to health care are no exception. But in particular, in the field of health, AI has a more positive impact than other industries.
First generation
The way AI is trained at this time can be imagined as similar to the approach of medical students, AI systems are also taught hundreds of algorithms to translate patient symptoms into diagnoses. This is considered the first generation of healthcare rules to be incorporated into AI systems.
Decision-making algorithms grow like a tree, starting from the trunk (the patient’s problem) and branching out from there. For example, if a patient complains of a bad cough, the doctor will first ask if there is a fever. There will be two sets of questions, fever/no fever. The initial answers will lead to further questions about the condition. This will lead to further branches. Finally, each branch is a diagnosis, which can range from bacterial, fungal, or viral pneumonia to cancer, heart failure, or dozens of other lung diseases.
In general, the first generation of AI could recognize problems but could not analyze and classify medical records. As a result, early forms of artificial intelligence could not be as accurate as doctors who combined medical science with their intuition and experience. And because of these limitations, rule-based AI was rarely used in clinical practice at other times.
Full automation
By the early 21st century, the second era of AI began with Artificial Narrow Intelligence (ANI), or artificial intelligence that solves specific sets of tasks. The advent of neural networks that mimic the structure of the human brain paved the way for deep learning technology. ANI works very differently from its predecessors. Instead of providing pre-determined rules by researchers, second-generation systems use huge data sets to discern patterns that would take humans a long time to do.
In one example, the researchers fed an ANI system thousands of mammograms, half of which showed malignant cancers and half of which showed benign cancers. The model was able to instantly identify dozens of differences in the size, density, and shading of the mammograms, assigning each difference an impact factor that reflected the likelihood of malignancy. Importantly, this type of AI does not rely on heuristics (rules of thumb) like humans do, but instead relies on subtle variations between malignant and normal exams that are unknown to both the radiologist and the software designer.
Unlike rule-based AI, second-generation AI tools sometimes outperform human intuition in diagnostic accuracy. However, this form of artificial intelligence also presents serious limitations. First, each application is task-specific. That is, a system trained to read mammograms cannot interpret brain scans or chest X-rays. The biggest limitation of ANI is that the system is only as good as the data it was trained on. A clear example of this weakness was when UnitedHealthcare relied on narrow AI to identify the sickest patients and offer them additional medical services. When the researchers sifted through the data, they found that the AI made a harmful assumption. Patients were diagnosed as healthy simply because they had received little medical care in their medical records, while patients who used a lot of medical care were judged to be unhealthy.
Future generations of AI will also enable people to diagnose diseases and plan treatments just like any doctor. Currently, a generative AI tool (Google’s MED-PALM2) has passed the physician licensing exam with an expert score. Many other medical AI tools can now write diagnoses similar to those of doctors. However, these models still require physician supervision and are not likely to replace doctors. But with their current exponential growth rate, these applications are expected to become at least 30 times more powerful in the next 5 years. Future generations of tools like ChatGPT are predicted to put medical expertise in the hands of everyone, fundamentally changing the doctor-patient relationship.
Compiled by VIET LE
Source
Comment (0)