09-03-2021 | | By Sam Brown
Recently, reports of IBM considering the sale of Watson Health have been emerging. What is the Watson Health division of IBM, why is IBM considering its sale, and why is the medical field showing signs of hesitation with medical AI systems?
IBM is famous for a wide range of different technological achievements and inventions, but the Watson system is arguably one of their most famous creations. The IBM Watson system is a deep learning computer system that can answer questions written in natural language, and provide an answer in natural language.
Watson was able to show off its capabilities against other Jeopardy! players back in 2011 by winning the first place prize of $1 million. However, the Watson computer system applications go beyond winning game shows, and as such, the developers of the Watson system created a new division of IBM, IBM Watson Health.
IBM Watson Health is a business that utilises Watson’s capabilities to analyse medical data, records, and patient symptoms to develop healthcare solutions. The platform, which charges for its services, uses cloud-based processing to allow medical professionals from all around the world to use the Watson system.
The use of such AI systems is proving to be an invaluable tool for the medical diagnosis of patients thanks to the learning capabilities of AI and the ability to look at minute details. For example, radiology departments are responsible for analysing images taken by scanning systems (such as MRI and PET). Still, while the human eye can spot oddities, it is not as good as seeing changes. AI, however, is very good at looking for changes in images, and then parsing such changes through neural nets (which have been trained by millions of other examples), to decide on what the change is and if it brings cause for concern.
The same can be said for disease diagnosis. Before tests can be ordered, a physician needs first to determine the likely cause of illness, and this is done by gathering symptoms from the patient. In particular, the Watson system is extremely efficient at being given an answer (i.e. the symptom), to guess what the cause is. Doctors tend to show bias in diagnosis whether it’s due to a lack of care or other held prejudice. Still, computer systems merely take inputs and provide outputs with no prejudice involved.
According to recent reports, IBM is in the process of considering the sale of IBM Watson Health. The recent report shows that IBM Watson Health is unprofitable for IBM, and last year saw a decrease in revenue of $1.5 billion.
The exact cause of IBM Watson Health’s poor performance is not clear. Still, it may be several contributing factors including the difficulty in implementing Watson into solutions as well as a general distrust of AI in the medical field. WSJ reported that IBM could be considering the sale of IBM Watson Health to a private-equity firm or industry player.
If it’s one thing that AI is excellent at, its detecting deviations over time in systems. In fact, AI is so good at this task that many sensor systems used in industrial applications now have predictive functions that can determine when an industrial process is deviating, and when maintenance is required before problems occur.
While far more complex, the human body is still a machine (albeit a biological one). As such, it only makes sense that AI systems are ideal for making a medical diagnosis, determining what medications should be prescribed, and what tests should be done. From being able to image all parts of the body to the ability to read through all medical records, AI clearly poses an advantage over human doctors, so why is its implementation taking so long?
The answer is that it depends on who you ask...
Some will say that it relates to privacy. The ability for medical records to be made freely available to AI systems could leave patients at risk. For example, those who operate AI systems could potentially have access to private information regarding patients (from test results to pictures of the patient). Another example would be the same information being obtained by constant third-parties who may eventually sell the information to untrustworthy parties.
Others will say that doctors don’t trust AI to get the right diagnoses. The field of medicine is rather unique, somewhat antiquated, and most likely corrupt. In countries such as the UK, the organisation that determines the wage of doctors (BMA), is also the same organisation that determines how many students at University are allowed to become doctors. It would make sense that such doctors would want to protect their own job security, which is helped by rejecting AI.
The claim of doctors is that AI cannot understand lies from patients, be able to ignore irrelevant data, and make outrageous diagnoses. Simultaneously, doctors also suggest that AI cannot be empathetic and show care to patients. Of course, anyone who has seen a doctor knows that doctors often use “Gpedia” to diagnose patients (i.e. Wiki version for GPs), often disregard patients concerns, show irritation when a patient attempts to self-diagnose, cast judgement on patients for lifestyle choices, and disregard certain diagnosis because “the chance of that happening is very low”.
As such, those in the medical field may see their jobs at serious risk (especially for General Practitioners and data analysis's), or believe that their diagnosis capabilities are simply beyond the capability of a machine mind. But when AI often proves its capabilities, and that some 90% of nurses who use Watson follow the guidance provided, the question as to why doctors are not using AI is raised, and could be a real cause for concern.