Understanding the AI doctor

By Ajit Narayanan, mfine

It is unlikely, that any reference to the future of healthcare happens today without the mention of Artificial Intelligence (AI).

Although, this field has been in existence from the 50s, it is only now, that this term & the transformative capabilities that it brings to the healthcare industry, has received this much attention. AI in a simple definition is the ability of computers to perform tasks that otherwise takes human intelligence to do.

Much of Artificial Intelligence in healthcare is already in use today, particularly in imaging, echocardiography & screening for neurological conditions. But, there are also outlandish ‘claims’ of what AI is capable of doing in the very near future and among them, the most commonly heard one, is that “AI will replace doctors”.

Events such as the IBM Watson beating Jeopardy! champions and other computer programs winning at chess or Go (an ancient Chinese game), fueled the misrepresentation that if computers can outperform humans on certain specific tasks, they will eventually be able “out think” a human. Hence, the popular question of the Doctor vs AI. Popular, yes, but is it pertinent?

What AI can and cannot do in healthcare

A game like Go, no matter how complex, is still governed by a finite set of rules and possibilities. A computer program is very efficient to process conditions & outcomes of all possibilities by evaluating these rules. It will be able to do this in orders of magnitude faster and more accurately than any human being could.

Identifying patterns from data is another class of problems that computers are extremely good at. This is where you present the algorithm with large datasets (or evidence) and it is able to identify patterns in the data & fit statistical models to it, models that define the data.

In the case of medicine, these algorithms are able to see patterns in these datasets that a human doctor cannot. This is because the changes in data points are often subtle, spatially distributed and complex, escaping detection by visual inspection.

An impossible task for human senses. When repetitive tasks are represented with big data and rules, algorithms can be built and optimized to outperform human doctors at specific tasks. But, is that enough?

A doctor’s intelligence, however, is far more complex than mere rules and pattern recognition. It is obvious that to arrive at decisions and judgements one requires a very different mental process.

Rather than only learning from data, rules and patterns, humans also use pre-formed observations and knowledge from first principles, reasoning, planning, creativity and intuition to arrive at decisions.

These algorithms, however fast and accurate they are at what they do, lack conceptual understanding of fundamental medical concepts and even basic reasoning to evaluate new situations. In some advanced AI implementations, they may be able to form hypotheses, but may still lack the ability to prioritize and test them.

AI techniques in use today are reliable only to the extent that the data used to train them are sufficiently complete and representative of the environment in which they will operate in. When this condition is not met and when faced with a question of judgement, today’s approaches will fail.

A doctor’s intelligence and intuition will, therefore, be required to counterbalance these limitations of AI. This is the only reasonable goal and possibility for AI in healthcare today. Be assistive to human intelligence. It is a partnership.

There are also significant hurdles to overcome to get to this state. Seemingly, data is both the solution and the problem. Machine learning algorithms get better with more data they see, but access to this data, its privacy, inherent biases that may exist in the available set of data, remain points of concern. The more the availability, the better the algorithms work, the better the partnership works and better is the clinical outcome.

This also means that more trust from consumers and healthcare professionals is required to make data more available for research and development.

The future

AI enthusiasts will still argue that advancements are being made even in the realms of such finer elements like creativity, intuition, etc. that constitute a human’s intelligence.

But, we must ask ourselves if this where the focus should be. Instead, doctors and technologists can together build machine agents or a doctor’s personal digital assistant that can help doctors think and perform better. But, this needs both doctors and technologists to be accepting of the future & working together.

It is evident that machines will outperform doctors on specific repetitive tasks. In due course, a doctor’s role will need to morph into a more symbiotic one with the machine.

Doctor’s future tasks will include setting goals for these machine agents, designing them by modeling the foundational knowledge, formulate hypothesis, perform evaluations and be the final authority in decisions and suggestions offered by AI.

AI will do what it is really good at, computationally intensive work, that must be done to prepare the outcomes and suggestions for insights and better decision making in diagnosis and treatment plans.

Today, AI has the ability to pick up water mobility changes in the bone cartilages from an MRI and predict the possibility of someone being diagnosed with Osteoarthritis a couple of years later. The subtle changes in the soft tissue are hard to see and infer for a human doctor. Not so hard a problem for an algorithm.

Artificial agents are able to detect anomalies by continuously monitoring patient’s data from various devices and lab reports and also notify the human doctor for timely intervention.

This is an ideal scenario where we can potentially have a doctor watching over key parameters and vitals of all the patients. Several scenarios like these are unimaginable without AI.

A doctor assisted with this technology, will be able to perform far more effectively and in implausible scenarios than a human doctor or AI can do by themselves.

When built right, AI is sure to become the most indispensable tool in the doctor’s arsenal, assisting them & helping them to get to the right diagnosis and care for patients. Instead of spending time sifting through data and past medical records, doctors must allow their digital assistants do these tasks, refine and approve results, while they spend more time with their patients advising and comforting them with empathy.

Turn a good doctor into a great one using AI and give the doctor infinite reach using mobile and internet technologies. Infinite “super doctors” for everyone - that should be the focus of future research and products using AI in healthcare.

(Ajit Narayanan is the CTO of mfine. Views expressed above are his own)