It’s Klippel–Trénaunay–Weber syndrome,” a young doctor Ross Crawford suggested immediately on seeing a patient. While his boss was in awe at his feat of picking out the rare condition, characterised by port-wine stain, varicose veins, and bony and soft tissue hypertrophy involving an extremity, it was a fluke. It just so happened that in […]
It’s Klippel–Trénaunay–Weber syndrome,” a young doctor Ross Crawford suggested immediately on seeing a patient.
While his boss was in awe at his feat of picking out the rare condition, characterised by port-wine stain, varicose veins, and bony and soft tissue hypertrophy involving an extremity, it was a fluke.
It just so happened that in an earlier rotation on a paediatric round, he had seen the same condition in another woman. But most medical practitioners may go their entire careers without coming across a single case like this, let alone two.
Each doctor is an individual, and their expertise is largely drawn from the experiences they’ve accumulated over a lifetime. When that person retires, the knowledge goes with them, Professor Crawford muses.
Instead, imagine being hooked into a system that could instantaneously access information and compare your patients’ symptoms from anywhere in the world.
If we want to transcend the limitations of our own learning capabilities, artificial intelligence (AI) may be the way forward.
When faced with a curious set of symptoms or an unusual scan, an AI program could browse through thousands or millions of documents to find patients with similar conditions in a second, and provide a doctor with information on how those patients were diagnosed and treated.
Nowadays Ross Crawford is an orthopaedic surgeon, and chair of orthopaedic research at the Queensland University of Technology. As an internationally recognised expert in the field of hip and knee replacement surgery, he is eager for the introduction of artificial intelligence into healthcare.
“It’s going to make me a better surgeon,” he says. “It’s going to revolutionise medical care.”
Movie and TV directors and writers have already tried to imagine what AI might look like in a day-to-day setting. Such creations range from HAL 9000, the sentient computer in charge of the spaceship in 2001: A Space Odyssey, to Red Dwarf’s Arnold Rimmer, a hologram of a dead crewmate created from the memories, thoughts and feelings stored on the ship’s computer.
Spike Jones’ 2013 movie, Her, cleverly shows the ways in which we might use artificial intelligence to organise our lives. In the movie, a computer operating system, voiced by Scarlett Johansson, acts as a personal assistant, confidante, source of information, and eventually, even, as a lover.
While the latter probably won’t be a feature of AI used by doctors, the ability to have real time conversations with a machine that could arrange emails, or manage the administration of an office, or remind us of important events or chores, would be a game changer.
But AI doesn’t necessarily have to look like the “strong” AI in those examples. Rather than having the same cognitive abilities as humans, “weak” AI can also help with cognitive work in narrower fields. For example, Siri-like information retrieval, playing chess or pattern recognition.
On a really basic level, we have an instance of robotic AI in the Roomba vacuum cleaner that can navigate a room or a house and respond to obstacles in real time.
Machine learning – the way computers learn without being explicitly programmed to learn – is now being used to turn huge amounts of medical data into useful outcomes.
One company to turn the power of machine learning to diagnostics and clinical decision support is Enlitic.
Jeremy Howard, Australian CEO and founder of the company, describes the superior efficiency of applying an intelligent algorithm to diagnostics, and stresses the cost savings of the technology that allows radiologists to be more efficient.
“It is 50% more accurate than a panel of four top radiologists at lung cancer classification,” he told The Medical Republic.
“On the Lung Imaging Data Consortium dataset, there are 0% false negatives for the algorithm versus 7% for the radiologists, and false negatives are 45% for us versus 65% for the radiologists.”
With this kind of technology, theoretically all radiologists would be better than the best radiologist.
So when computers are able to exceed the skill of expert radiologists using advanced pattern recognition, would and should anyone choose a human over a robot?
For Professor Crawford, it’s not even a question of whether that will happen, only when.
We are seeing the fruits of AI in healthcare already. The first time most people heard of IBM’s artificial intelligence project, called Watson, was when it achieved a crushing victory over its human rivals on the game show Jeopardy.
Now Watson is turning its immense cognitive power to the big analytics problems in health.
Over the last few years, Watson has been training to be an oncologist. The aim is to provide personalised treatment plans for every patient, and to do so more quickly and cheaply than human doctors.
IBM describes how the process of cognitive computing can help in the field. After talking to a patient and gathering information on the problem, the doctor could turn to Watson and describe the symptoms and related information.
Watson can then identify the significant factors and search through information on the patient’s history, current conditions and medications the patient is taking.
Tying this together with the latest research, guidelines and the patient’s test results, Watson then presents the doctor with hypotheses and provides a confidence weighting.
A computer can incorporate electronic medical records, the potential side effects from the hundreds of drug combinations and patient notes with relative ease – and spit the results out quickly and repeatedly without tiring.
Already Watson has teamed up with dozens of cancer institutes around the United States to improve treatment for cancer patients.
Here, the proposed interface looks like a decision tree. It offers possible diagnoses, suggests tests that could confirm or rule out a diagnosis and then presents treatment options.
So potentially it will do the heavy lifting of clinical practice, and allow doctors to spend more time building relationships, counselling and comforting patients.
It may provide a better interface for learning about new research, instead of trawling through the tsunami of papers that are published every week.
Professor Crawford stresses how much more efficient a robotic surgeon will be, pointing out that one robot could perform 100 operations in a day, without tiring.
The question lingering on everyone’s minds, is whether the robots will replace humans altogether?
Dr Bertalan Meskó, a medical doctor and self-described “geek medical futurist”, says he is often asked this question by doctors.
“Cognitive computers will help physicians diagnose the same way the stethoscope changed the medical profession from the early 19th century,” he said, pointing to the difference between the very best clinician and a computer that can process the equivalent of a million books in a second.
While it will transform the way healthcare is provided, computers can never replace the human touch, he says.
But it’s not only going to be a question for doctors. Robots and artificial intelligence will be responsible for 5.1 million net job losses over the next five years, according to new research from the World Economic Forum.
Already we’re seeing humanoid robots taking orders and serving food in restaurants in China, and working as receptionists in Japan and Canada.
It’s inevitable that countries with ageing populations will turn to technology for help, says Jonathan Roberts, Professor of Robotics at Queensland University of Technology.
And with an ageing population, so too will we be facing a greater number of people living with dementia. Something will be needed to help people stay at home longer, and then in managed care after that, Professor Roberts notes.
Japan has led the way here, already introducing robots such as Paro into aged care facilities. Paro is a fluffy, robotic seal that acts as an alternative to companion animals. Instead of real life animals however, the 3kg baby harp seals don’t come with the burden of keeping them alive, nor the risk of allergies or biting.
In contrast to other so-called “dumb” robots the robotic seals are specifically programmed to perform tasks. For example, Paro has artificial intelligence that allows it to learn what to do to get stroked and to repeat that. It seeks out eye contact, can learn whether its carer is happy, sad or angry and respond accordingly. It knows its own name, and learns words that it commonly hears and blinks and opens its eyes in different lighting.
Paro even cries when it is upside down or is treated roughly.
These robots have already been on the market for the better part of a decade. And in that time, questions have been raised about the ethics of transferring the burden of caring for the elderly and those with dementia to robots.
While it is difficult for carers to manage with these types of individuals, some might feel uneasy at the prospect of encouraging the emotional needs of our parents and grandparents to be met by something that has the emotional expressiveness of a mewling kitten.
Other influential tools may not be traditional robots as we think of them, Professor Roberts notes.
“There’s already some of these kinds of robots that remind people to take their medication at certain times of day, that assist them and that help them [to keep doing these tasks],” he says.
While AI has a huge potential to benefit the health system in countries such as Australia, the impact will really be felt in developing nations, according to Toby Walsh, a professor of AI at the University of NSW.
Every day people in the Third World are dying needlessly from diseases that take “pennies to fix”, says the research leader at Data61, Australia’s Centre of Excellence for ICT Research.
“I like to tell people to think of a time in the very near future when we can give people in the Third World the sort of diagnostic advice that we expect in First World health care with simply a smart phone app,” Professor Walsh says.
And in developed countries like our own, AI holds the key to truly personalised healthcare.
“If we are to tackle the increasing costs of health care, technology must be part of the solution. We cannot live longer and society cannot support the ever increasing costs of health care unless we embrace technologies like AI,” he argues.
A new analysis by research firm, Frost & Sullivan, estimates that AI in healthcare will see a 10-fold increase in revenues in the next five years, from $634 million in 2014 to $6.7 billion in 2021.
“Clinical support from AI will strengthen medical imaging diagnosis processes. In addition, the use of AI solutions for hospital workflows will enhance care delivery,” the report states. “Overall, AI has the potential to improve outcomes by 30% to 40% while cutting treatment costs by as much as 50%.”
The firm predicts that AI could become a part of everything from population health management, to digital avatars capable of answering specific patient queries over the next decade.
“On a global scale, in regions with high under-served patient populations, AI is expected to play a significant role in democratisation of information and mitigating resource burdens,” said Frost & Sullivan Transformational Health Industry Analyst Harpreet Singh Buttar.
So far hundreds of companies have formed partnerships with Watson Health, subscribing with a monthly fee to access the cognitive power of the cloud where Watson lives.
Novo Nordisk and Medtronic are now attempting to bring Watson into the management of diabetes, with the companies claiming it could warn patients up to three hours before a hypoglycaemic episode by analysing data from insulin pumps and glucose meters.
Then we have the athletics wear company, Under Armour, who are hoping to turn the tool into a more encompassing “cognitive coaching system” in an app. They capture the user’s activity, nutrition, fitness, sleep and how they are feeling mentally, with the hopes of providing personalised health advice throughout the day.
Genomics companies are also vying to bring personalised health advice to customers that depends on each person’s specific genetic makeup.
“Of course, like any technology, there are also many considerable ethical concerns,” Professor Walsh warns.
“Will insurance companies start to misuse this mass of personal data? Will AI enlarge inequalities within our society in terms of our wealth, health and happiness? Will robot carers dehumanise the care of the elderly?”
Professor Roberts agrees that privacy and confidentiality issues are a big potential concern.
According to IBM, the medical community has been one of the earliest adopters of the technology.
“Luckily the medical world is good at knowing the pitfalls of technology and preparing for that,” says Professor Roberts. “There’s a strong ethics bent in this community, and they have been advanced compared to other industries when it comes to knowing what to do with these issues.”
| Panacea? Or harbinger of the apocalypse? |
| Just as some sectors are racing toward the development of a human-level artificial intelligence, there is a growing movement urging caution.
Rather than luddites, it’s experts in the field of AI who are worried about what will happen when we create intelligence that is as good as, or better than, human intelligence. “When it eventually does occur, it’s likely to be either the best or worst thing ever to happen to humanity, so there’s huge value in getting it right,” astrophysicist Stephen Hawking said in a Reddit Ask Me Anything. While it’s all up in the air, people’s perspective on the potential of AI falls on a spectrum ranging from “panacea for all the world’s ills” to “harbinger of the apocalypse”. However, according to Professor Hawking, the danger will not come in the form of malevolence. Rather it is that AI could be so powerful that a bug – or even the unintended consequences of doing its job properly – could end drastically for us. The most incendiary example at the moment is Right now the rationale is that these kinds of robots will be used to save lives both on the battlefield and in emergency situations that are unsafe for humans. Both Professor Walsh and Professor Roberts have been among the experts in the field to have added their names alongside those of Hawking, Tesla founder Elon Musk and Microsoft’s Bill Gates in a petition to ban killer robots. They have called for more research on the societal impacts and ways to prevent potential hazards before we rush headlong into creating something that we don’t know how to control. Disaster or combat zones would be hugely benefited by an autonomous robot that can make decisions on its own and lift injured soldiers or civilians out of harm’s way. But the worry is that autonomous robots could go the way that just about every other military technology has gone – from planes to lasers to drones – which is to be repurposed by the military as a weapon. The military artificial intelligence arms race must be fought against, say the signatories of the petition. And more broadly we must be careful of unfettered movement towards “strong” AI, Professor Hawking warns. Sooner or later, we may be facing a scenario in which we have created an AI that is more intelligent than us. There’s no reason to think that this super-intelligent machine won’t be able to create something more intelligent than itself, and so create a sort of intelligence explosion. This is often known as the singularity. “A super-intelligent AI will be extremely good at accomplishing its goals, and if those goals aren’t aligned with ours, we’re in trouble,” he said. “You’re probably not an evil ant-hater who steps on ants out of malice, but if you’re in charge of a hydroelectric green energy project and there’s an anthill in the region to be flooded, too bad for the ants,” he said. “Let’s not place humanity in the position of those ants.” |
