About one million years ago man learned how to control fire. They learned how to create it a couple of hundred thousand years later. Before that, fire was the act of the gods. To mention some of these deities, the Greek god Haphaestus, Roman god Vulcan, Aztec god Chantico, Japanese god Fuji and the Hindu god Agni, all had the power to create and control fire while man lived in fear of the unpredictable consequences.
In the present day, the average man is now facing a new wave of technological wonder: change that is increasingly becoming imminent. This technology is one that can make cars drive by themselves without a human in the driver's seat; call centres that are active 24/7 but deserted, save for some buzzing computers; robots that are performing complex surgeries. Is this all true? What are we to do? Revolt in protest to this world dominated by machines, or silently accept and fade away?
But perhaps the reality is not as grim as all that: on the contrary, it might be beneficial to the progress of man.
Artificial Intelligence is all around, we just didn't take notice
You might wonder how an e-commerce website knew how to suggest to you an item you wanted without any hints or how your navigation app was able to recommend a faster route in very unpredictable traffic situations. You may give yourself the initial reasoning that the people behind it had large amounts of data to analyse and stop at that. How it came to reach solutions so fast and accurate and, most important of all, free, was beyond what you would want to waste your time pondering.
Today's artificial intelligence capabilities have advanced exponentially from those of the past decade due to breakthroughs in computing technology and the use of the cloud, leading to the much broader practice of machine learning and deep learning, where powerful processors spot hidden patterns in enormous amounts of data that would usually take weeks and months to process. The result would be in the form of models which can be used to group things into meaningful categories or predict results and answers. In healthcare, the former would be artificial intelligence to assist in the diagnosis of illnesses, from x-rays or images of tissue samples. The latter, predicting the outbreak of epidemic diseases from key words search by patients in search engines.
At the time of writing, the US Food and Drug Administration has approved through its 501k and De Novo pathways 12 artificial intelligence algorithms to be used to help doctors in diagnosing diseases and monitoring treatment progress.
The value of a tool depends on how it's used
The mere thought of computers replacing doctors may sound apprehensive. However using artificial intelligence as assistance or guide has the potential to augment healthcare practitioners' capabilities. Human experts gather knowledge and experience over time, not just in the science of disease detection and treatment, but also in the care of fellow human beings who are their patients. Nevertheless, doctors are still human, just like us. We get tired, sleepy, emotional, and we have bad days. Computers, on the other hand, don't have any of these short comings. Machines carry on exact operations, every time, all the time. Power supply can be made redundant and maintenance pre-scheduled: No sick leaves or vacations. Specific strengths from computers can be a very good match in filling up some of human nature's organic gaps.
In November 2017, a research team from Stanford University produced and published CheXNet, an algorithm that was used to perform deep learning on 112,120 labelled chest x-ray images. The result was that CheXNet performed better in diagnosing diseases than four Stanford University radiologists who participated in the study. Apart from accuracy, it is important to note that each radiologist reviewed only a subset of 420 x-ray images whereas the machine could take on the full load of a hundred thousand.
Artificial intelligence can be purposely used to assist and enhance the performance of healthcare practitioners with machine properties of accuracy, sensitivity and availability. Keep in mind the problems of rising healthcare costs, inequality and access issue that we face today. Conventional wisdom has not found any sustainable solutions for these concerns. Artificial intelligence has the potential to resolve those challenges.
Creating more creators
If it is so good, why don't we see more artificial intelligence in healthcare?
Unfortunately, this technology is still in the hands of data scientists, computer experts and machine learning enthusiasts. Promising uses in another setting or country are labelled as "science fiction" or "too good to be true" by most clinical practitioners. Patients are still very sceptical. But they do realise that they are hearing about it more often. The knowledge gap to artificial intelligence in healthcare is so huge that it creates fear of uncertainty with even the slightest thought of introducing artificial intelligence in the clinic. It would be no different from the reaction of a hunter-gatherer living in a shack being told that someone is going to bring fire into his house for warmth and light. The fundamental complexity of machine learning does not go along well with the very busy nature of healthcare practitioners in the effort to understand and fully utilise the technology. On the other hand, practitioners who fall in love with machine learning, having clinical data and the vision to create a system that can solve pressing healthcare issues, end up developing the system themselves and leaving their practice behind.
Data scientists that can create artificial intelligence models are not easy to find. But what we need are data scientists who understand healthcare enough to be able to appreciate the value of artificial intelligence in specific case contexts and determine what data is needed for creation.
How healthcare experts can interact with artificial intelligence in practice is an important question of usability and practicality. Software engineers who create systems that house the artificial intelligence systems must be able to master the workflows and processes to come up with systems that can integrate the new technology with existing methods of practice.
We need to produce more of these scarce talents so we can experiment and implement more innovative ideas, hoping that some will mature and help solve the numerous healthcare issues we constantly encounter.
Controlling the machines
As with all things created by man, artificial intelligence is far from perfect, even if we ever get to the age of general artificial intelligence where computer programs can think and address a broad range of issues as humans can. The proper methods to supervise and control artificial intelligence should be planned and put in place before any large scale or life-critical use can be introduced. In the classic dilemma for self-driving cars whether it would choose to turn right and run into a crowd or turn left and run over an old lady, computers may calculate the risks and sum up the value that each person may bring to society if left alive and decide accordingly. We could choose to override the computer heading for a tree.
Author: Dr. Veerachat Petpisit, CEO, Healthcare Ventures, BridgeAsia, Email: veerachat_p@bridgeasiagroup.com
Series Editor: Christopher F. Bruton, Executive Director, Dataconsult Ltd, chris@dataconsult.co.th. Dataconsult's Thailand Regional Forum provides seminars and extensive documentation to update business on future trends in Thailand and in the Mekong Region.