A Different Kind of AI Story
There is so much hype around AI. Anything remotely interesting gets labeled revolutionary when it is little more than a marginal improvement over what you could already do. Much of the AI news is really just branding.
However, there is one field wherein the advances are indeed life-changing and saving and not just “showy”-healthcare.
Here’s what is really going on.
Reading Medical Images with Precision
The most well established application to date is in the diagnostic imaging of medicine. When you think about the millions of gigabytes of information stored in a simple x-ray, an MRI or a CT scan-an expert radiologist can look at them for issues. But they are still human. They get fatigued.
AI systems that have been trained on millions of medical images can scan through a scan and highlight areas that ‘seem’ abnormal. They won’t replace the radiologist – the doctor will still decide on the diagnosis. But AI becomes that additional eye that doesn’t tire, doesn’t slack off.
Indeed, research has demonstrated that AI-supported reading will detect cancers and other lesions that the initial reading missed. This is not just an intangible benefit. Those are real people receiving a diagnosis that otherwise might not have happened.
In the U.K., the National Health Service has been experimenting with AI programs that interpret mammograms, a mammogram is the picture used to diagnosis breast cancer. On certain test run the AI had been nearly as accurate or more than an experienced radiologist. The NHS is now deploying these programs more widely.
DeepMind, Google have developed an AI that will scan an eye and identify over fifty eye diseases. The AI has been developed and trained using data from Moorfields Eye Hospital London and will be able to diagnosis diabetc retinopathy (a leading cause of blindness) with the same accuracy as experts.
What is exciting about this is the magnitude of difference it makes. An eye specialist can only see a limited number of patients in a day. An A1 can scrutinise thousands of scans in an hour. Countries where specialists are not in abundance will find this truly revolutionary.
Accelerating Drug Discovery
And then there’s drug discovery. Coming up with a new drug is one of the hardest problems in science. It takes between 10 and 15 years and over a billion dollars, on average, and most candidates fail. That’s because the universe of all possible molecules is huge beyond belief-you have to examine a vast number of compounds to find one that hits the mark.
This is not the way it’s going to stay. Already solutions available from companies like DeepMind, Insilico Medicine and many others, are using AI to identify potential promising compounds, narrowing down the search significantly. DeepMind’s AlphaFold system were able to breach a puzzle in the field of protein structure prediction, which baffled scientist for 50years. And this is the key to drug discovering!
A handful of AIgenerated drug candidates are now in human clinical trials. None have yet been approved-clinical trials take years. The pipeline is happening.
Supporting Doctors with Clinical Decisions
And there is the matter of clinical decision support. This is AI that will aid doctors in making better decisions for their patients. Not by usurping the doctor’s decision, but by providing them with more information, more quickly.
Picture a doctor meets with a patient, and an AI system has just read through all of his-or her-full decade plus of medical records, identified the three most pertinent issues for this office visit, and brought to the doctor’s attention that the patient might be taking the wrong dosage of his medicine, which the doctor might not have known to look for otherwise. That’s not science fiction. And that’s the kind of things that Epic’s AI products in the hospital are beginning to do.
Reducing Paperwork and Burnout
The bureaucracy of the practice of medicine is also being revolutionised. Doctors today spend a huge proportion of their day doing the paperwork: notes, billing codes, documentation, discharge summaries, etc. Now AI can produce a clinical note for the doctor based on what was said to the patient.
The doctor talks to the patient.
AI records what is said, creates a note and the doctor reads it through, modifies it and signs it off. The note now only takes two minutes to produce rather than twenty.
This is important not just to save our time. This is so important because the less time our physicians we have to spend on maintenance and administrative work, the extra time those physicians will spend in patient care. And the less burned ou the physician the greater his decision’s quality.
Real Concerns to Consider
Certainly there are real issues at stake as well.
AI in health care is not perfect. And in medicine, failures can cost lives. There have been situations where one particular AI system was discovered to fall behind on some types of patients, usually because of lack of representativeness in the training dataset.
A system trained primarily with data on one group of patients may just not be as accurate in diagnosing some other group.
This is a known issue and people are working on it.
Additionally, who is to be held responsible?
When a human doctor errs, whose fault is it?
The doctor?
The hospital?
What if an AI system helps to produce a false diagnosis?
The doctor?
The hospital?
The AI company?
Those questions have yet to be answered in court.
And then there is access: The most cutting-edge AI-based medical technologies are arriving first in the best funded hospitals in rich countries. It’s the same pattern as with every new technology: those who get access first are those who have the most. Expanding access to low-income countries and rural areas will be extremely difficult.
A Genuinely Good Story
However, take all of those reservations at face value-at heart, the story here is a good one. AI is enabling us to diagnose illness sooner, develop medicines more quickly, combat burnout among doctors, and bring specialist skills to locations they previously could not reach.
They are not insignificant. These are matters that decide lives and deaths.
And that, more than any chatbot, is the AI narrative that merits the focus.
Leave a Reply