AI and machine learning are heralded as technologies of great hope for the future of healthcare, but an industry expert has pushed back against the hype, predicting that they will not be fully seeded into industry practice anytime soon. 
Speaking at the recent Wild Health Summit in Sydney, Macquarie University Australian Institute of Health Innovation Centre of Health Informatics Director Professor Enrico Coiera said the full blown effect of AI and machine learning on the healthcare industry is not near.   
“Amara’s Law states that we overestimate the effect of technology in the short run and underestimate it in the long run. This idea of an AGI – artificial general intelligence – is quite a way away. I’ve yet to see any inkling of the class of technology needed to bring that world upon us,” he said.  
According to Coiera, the speed of progress will only see AI and machine learning integrated into healthcare in some 10 years or so.
“There won’t be big changes within the next five years. The healthcare industry is quite slow in its uptake as compared to other industries and trying to make changes within that period is hard for us. Our opportunities to innovate in models of care is to start embracing tech disruptors and doing things differently,” he said. 
“We will need to engage in the process. The other important piece for our industry is to change the way we practice healthcare. This involves education around automation bias. We need to train people to handle technology and not be over dependant on it.” 
Coiera said AI and machine learning will not take over human jobs, but instead, support human capabilities focusing on patient care. 
“There will still be work for humans. We will be able to expect transformation of the industry, not human redundancies. There will be more of human and machine interactions and developments. AI and machine learning will be doing very specific tasks and there’s no reason why those tasks will cross paths with what humans do.”
Professor Coiera also warned of the potential issues around safety, with the possibility of AI and machine learning becoming autonomous to some degree and making its own decisions. 
“This will pose a lot of issues for certification and accreditation of safety. The disruption that does arise, will happen in unexpected ways. So, there’s a lot of things that we need to have the foresight of and change before this disruption occurs,” he added.  
Tools of change 
Google Brain USA Product Manager Eyal Oren, who also spoke at the summit, identified some ways that the industry can start roping in AI and machine learning into practices.  
He said healthcare providers should be able to leverage millions of data points to improve patient care, and do that at scale. To perform that task, he said machine learning and AI is necessary. 
“As we move through clinical domains, providers, IT systems, etc. records should too. A patient or healthcare provider should be able to access and share end user data,” he said. 
However, Oren said that an abundance of data currently available is blocking the potential of AI and machine learning. 
He claimed that for a number of information flow, business or privacy reasons, many healthcare providers don’t exchange data. They also refrain from exchanging data as a result of not knowing how to clean and extract value from it. 
“The problem is, there’s an abundance of data that’s siloed, messy and is usually kept away. So, how can healthcare professionals get value from that? Machine learning is math, not magic. Messy data beats no data and you can actually do quite a lot with it,” he said. 
In addition to improving cost and quality of healthcare by learning from millions of patient-years of messy electronic health data, AI and machine learning also enables real-time clinical predictions that outperform traditional risk scores by harnessing patients’ longitudinal data, according to Oren.  
“Health is longitudinal. For example, in the US, one in four patients usually has an adverse reaction in patient care during their hospital stay. Very often, there is data that could have helped predict that. And with a lot of the hospitals in the US being digital, there are some 150,000 data points for a clinician to use.”
But not much of the data is being put to practice as a human can’t read all of those data points. 
“So, the aim should be to use AI and machine learning to go through this data and make predictions. With predictions, you’ll be able to do a lot of other things. In Australia, this should involve the use of electronic health records to make substantially better predictions. Machine learning is not the next big thing, it’s the current big thing,” he said. 



White papers