Understanding the A.I. Doctor
It is unlikely, that any reference to the future of healthcare happens today without the mention of Artificial Intelligence ( A.I ). 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. Although lot of A.I is in use today in healthcare, particularly in imaging, echocardiography & screening for neurological conditions, there are equally outlandish ‘claims’ of what it is capable of doing in the very near future. The most commonly heard one, of course, is that “A.I will replace doctors”.
A.I in a simple definition is the ability of computers to perform tasks that otherwise takes human intelligence to do. The key moments of A.I perhaps came with IBM Watson beating the Jeopardy! champions and other publicized events like computer programs beating humans at chess or Go ( an ancient Chinese game ). These events fueled the mis-representation, that, if computers are able to outperform humans on certain specific tasks, they will be able “outthink” a human. Hence, the popular question of Doctor Vs AI. Popular, yes, but is that 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 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. More importantly, 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 arriving at decisions and judgements use a very different mental process. Rather than using laborious ways of learning from data, rules and patterns only, humans also use pre-formed observations/knowledge, 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 commonsense reasoning to evaluate new situations. In some advanced AI implementations, they may be able to form hypotheses, but may still lack ability to prioritize and test them.
AI techniques practically in use today, therefore 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 A.I. 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. Data, it seems 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 algorithms work, the better the partnership works and better the clinical outcomes.
Practically, this means that, more trust from consumers and healthcare professionals is required to make data more available for research and development. The consumers, more importantly, being the owners of this data and therefore become the key to its success.
The future
Many will still argue that advancements are being made in each of those finer elements like creativity, intuition that constitute a human’s intelligence. But, we must question if this is where the focus must lie. By changing focus of AI in healthcare, doctors and technologists can together build machine agents or a doctor’s own digital assistant that can help doctors think and perform better. But this needs both doctors and technologists to be accepting of this future & work together.
It is evident that machines will outperform doctors on specific repetitive tasks. In due course, a doctor’s role will need to morph in to 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 possibility of someone being diagnosed with Osteoarthritis 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 monitor continuous streams of data of patients from devices, lab reports and be able to observe anomalies and are able to bring in the human doctor for the right intervention. This is a state where we can potentially have a doctor watching over key parameters and vitals of all his/her patients. Several scenarios like these are unimaginable without AI. But a doctor assisted with this technology, will be able to perform far more effectively and in implausible scenarios that 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.
Say hello to the doctor of tomorrow.
