Man shall be overcome. What have you done to overcome him? So asked the German philosopher Nietzsche. Now modern science is answering. Wang Yuke reports.

Fatal accidents caused by medical staff errors involving patient medication may be largely eliminated by a new artificial intelligence (AI) program under development at the Hong Kong Polytechnic University (PolyU).

Sharp increases in the number of patients given the wrong medication or overdoses have set off alarms, bringing concerted efforts to bring those numbers down, and with good reason. A patient given the wrong medication or the wrong dosage could suffer serious, even fatal consequences, cautions Francis Chow Chun-chung, head of Endocrinology and Diabetics at the Prince of Wales Hospital.

He cites allergic reactions to drugs, as in the case of patients suffering from gout who may be allergic to the commonly prescribed drug allopurinol, which reduces uric acid in the system. If the patient is allergic, he may suffer a severe skin reaction, kidney failure, and sometimes death.

Statistics reveal that the incidence of medication errors shot up from 1,399 cases in 2010, to 1, 913 only three years later.

An annual report by the Hospital Authority revealed that there were 49 "sentinel" events and 94 "serious untoward" events between October 2013 and September 2014. Among the 49 sentinel events, five cases resulted in death due to medication errors, while 85 out of the 94 serious untoward events involved medication blunders resulting in death or permanent harm.

Often, medication errors are caused by a lack of information about the patient. For example, says Chow, "it's already understood that allergic reactions are closely associated with abnormalities in a specific gene (HLA-B*5801). If we don't know the patient carries the gene and prescribe medication on the assumption it will work, we'll put the patient's life at risk."

The AI prescription program developed by PolyU is gathering data on over a thousand individual genes and genomic structures, correlating their characteristic reactions to thousands of pharmaceutical products and dosages, all advancing toward a stage where medical professionals likely will turn to AI to prescribe medication. That's the view of professor Keith Chan Chun-chung, program head of PolyU's department of computing, whose research focus lies in AI and computer learning.

Eliminating human errors

People are different. Medications and their dosages cause different reactions in different people. When new patients are identified with specific illnesses, their genotypes will be recorded and their medications prescribed accordingly. What's effective for one person may cause an adverse reaction in another. The program, which Chan says is 70 to 80 percent complete, aims to eliminate medication errors and equally as important, to administer optimal dosages of the correct medication.

The way medications are metabolized and their physical effects are governed by several genes and individual differences in people can trigger a range of reactions, anything from slow or extended response to severe and adverse, explains Chan. Two patients given the same diagnosis and prescribed identical dosages and medications may react completely differently. One patient may experience marked improvement; the other may experience adverse side effects.

Due to improvements in AI and progress in machine learning, it has become possible for machines to process and analyze hefty amounts of data, rapidly to find correlations and patterns from random data, Chan remarked.

Once the sampling is complete, and the individual differences stored in the enormous database, the machine will be able to identify patterns emerging from the multitude of patient experiences or regularities from the data through analysis.

This type of machine learning will provide guidance when new patients with specific ailments are identified. They will be prescribed optimal medication and dosages and that will save lives.

It's all part of dizzying pace of technology that many hard-nosed scientists will mean that humanity will overcome death in this century. Google's chief engineer, author, inventor, and futurist Ray Kurzweil, credits the whirlwind advances to the "law of accelerating returns". Technology, he argued, in his 2006 book, The Singularity is Near, is expanding exponentially, not at a linear rate of 10 + 10 + 10, but at a rate of 10 x 10 x 10. Kurzweil predicts that technology of the 21st century will advance not by 100 years, but by 20,000 years, more than double the entire expanse of human experience, all within the course of a single century.

Case in point, Google's AlphaGo, a DeepMind experiment that defeated 18-time international champion Lee Se-dol, 4 games to 1 in the Chinese board game Go, considered the world's most complex game. Experts had believed it would take at least another 10 years before AI could win so decisively over a human professional player.

The coming of the singularity

Given the advance of cloud computing, accelerated processor speeds and machine learning computers already have surpassed human intelligence in some fields. In his 2006 book, Kurzweil foresees three ongoing technological revolutions, in robotics, genetics and nanotechnology, coming together to engender a "singularity" capable of bringing about the Uebermensch, or superman, envisaged by the German philosopher Friedrich Nietzsche in his 1883 book, Thus Spoke Zarathustra.

Chan and his team of graduate students are working on another device they hope will contribute to the coming of the singularity. The second AI device has its roots in traditional Chinese medicine.

The device is intended to provide immediate diagnosis of a patient's condition based on a patient's pulse, and on characteristic signs reflected in the condition of the tongue. The process owes to deep learning, an offshoot of machine learning, which furnishes the machinery with a keen image and sound recognition ability, so that diagnosis may be rendered remotely from photograph as the machine can assess the patient's tongue color and texture and the waves generated by pulse sensor.

Deep learning puts us closer to strong AI (on a par with human intelligence) that now resides in the realm of science fiction, said Chan. There's much more to come. Microscopic, nanobots, emerging from advances in nanotechnology, and programmed with deep learning software soon will navigate through the blood vessels, removing blood clots and carrying out other repairs on the body, said Chan.

The nanobots will identify characteristics that may appear in clotted blood vessels and send critical data back to the computer screen. Physicians will then be able to identify the patient's condition before instructing the robot in the appropriate remedial action, said adjunct professor Kwong Chung-ping from the Mechanical and Automation Engineering Department at the Chinese University of Hong Kong.

"We built a deep learning algorithm in the nanobot so that it learns to react when it encounters clots in a patient's blood, whether by perceiving the blockage, feeling the blockage, or detecting the chemical component characteristic in clotted blood," explained Kwong.

Kurzweil's predictions are coming to realization, "just on a small scale and with imprecision," says professor Chow King-lau, a life science professor of the Hong Kong University of Science and Technology (HKUST). Although the nanobots can fulfill some functions, they do not perform as efficiently as normal biological protections, he argues. "The question is how long it would take for the efficiency and accuracy to catch up. I assume that it will take very long, more than a century."

As deep learning will equip robots with sophisticated image and sound recognition, robots will be able to read human emotions and thoughts, especially in the cases of physically handicapped people who have difficulty communicating their intentions and mood.

The robot of the future will substitute for many physical disabilities, in part, by reading the disabled person's facial expressions and analyzing their utterances, to determine what should be the appropriate action.

One of the best applications of deep learning would manifest in augmented reality (AR) technology, emphasizes professor Hui Pan from HKUST. Hui is a computer science expert who specializes in AR research. It will make self-driving cars much safer by allowing a motorist to detect potential hazards and road signs far down the road and out of site. The system potentially could render the traditional GPS obsolete, giving drivers greater options for changing lanes or changing routes, explains Hui.

Mobile phones trained through deep learning will assist users with categorizing and retrieving photos in milliseconds because the device can recognize and locate every single object in photographs.

Deep learning promises a more immersive and interactive game play experience in future, where players will become so deeply involved they may lose touch with whether they are having a fictional experience or whether it is reality, he says. "Imagine a fictional creature hops on a tree, then you see the leaves on the tree in real life, shimmering before your eyes."

Deep learning offers much hope. The list of its likely applications is endless, said Hui. It can include prompting more immediate and precise recommendation when using research engines, and assisting users to categorize and retrieve photos in milliseconds and giving automatic translation between text and images. Most scientists agree, many of these changes may be expected in the not too distant future.

Contact the writer at

jenny@chinadailyhk.com

(HK Edition 08/31/2016 page8)