The healthcare industry is one of the fastest-growing markets, estimated to be worth over $8 trillion USD in 2018. It is unsurprising that tech giants like Google/Alphabet, Microsoft, Apple, and IBM are trying to get a foot in the market as we head into an increasingly digital age of healthcare. Not only has medical technology undergone rapid advancements, but the COVID-19 pandemic has also pressured the acceleration of other healthcare aspects online, such as pushing the concept of telehealth or remote healthcare into the mainstream.
Given the many avenues in which technology has been integrated into our healthcare systems, the sheer wealth of information that is continuously being collected should come as no surprise. However, sifting through overwhelming amounts of data to uncover useful trends requires manpower, a resource in high demand that ultimately becomes the bottleneck to gleaning practical insights from massive datasets.
This is where artificial intelligence comes in.
Artificial intelligence, or AI, refers to computer systems that are capable of simulating human-like intelligence by demonstrating skills such as learning, reasoning, and problem-solving. By leveraging the superior computing powers of machines to analyze data using complex algorithms, AI systems can assimilate and analyze large quantities of data in a much shorter timeframe compared to conventional methods. For comparison, the AI algorithm AlphaFold, developed by Google’s AI subsidiary DeepMind, was able to accurately predict the 3D structure of a protein based on its amino acid sequence in the span of minutes, compared to traditional methods that would have taken researchers six months.
One of the tech industry’s most public forays in AI comes from IBM who made headlines when their supercomputer Deep Blue defeated then-world chess champion Garry Kasparov in 1997, and again in 2011 when their AI system Watson bested Jeopardy! champions Brad Rutter and Ken Jennings on the television quiz show. Immediately following Watson’s Jeopardy! win, IBM announced its intentions to develop Watson for use in the healthcare industry. Within 5 years, Watson Health successfully landed collaborations with institutions and companies such as Johnson & Johnson and Memorial Sloan Kettering Cancer Center (MSK) in New York, with a focus on cancer therapy. In 2015, Google’s DeepMind followed suit, branching into the healthcare industry with a focus on pilot operations in medical diagnostics.
However, as of 2021, IBM has expressed interest in selling Watson Health amidst declining profits and a disappointing performance in the clinic, while Google’s DeepMind Health division has faced public backlash over privacy concerns regarding patient data. These two stories serve to highlight some of the major challenges that still face the integration of AI technology to healthcare practices today.
IBM Watson Health
Watson’s 2011 Jeopardy! win demonstrated its proficiency in one subspecialty of the AI field: natural language processing (NLP). When given a quiz prompt, Watson was able to treat the question as more than just a combination of words, deriving meaning and intent based on contextual clues. This was accomplished by training Watson on a database of past Jeopardy! clues and answers so that the AI learned how to arrive at a correct response. IBM researchers hoped to apply the same methodology towards medical diagnostics.
Partnering with MSK, IBM developed the product Watson for Oncology as a tool for personalized cancer treatment recommendations. Like training for the Jeopardy! competition, Watson was given a database of cancer patient health records and access to the medical literature to learn how to find an answer to the question: which treatment option would best benefit a particular patient?
However, it soon became evident that IBM had bitten off more than it can chew on its diagnostics projects. Even with its advanced NLP algorithms, Watson was unable to consistently untangle meaning from patient health records, nor incorporate findings from the literature if the treatment does not benefit the majority. It seemed like cancer was just too complex a problem to be handled by Watson. Moreover, internal documents from Watson Health showed that the AI generated “unsafe and incorrect” treatment recommendations on more than one occasion, although MSK representatives clarified that these were instances during the testing phase and not actual recommendations given to patients.
Combined with the disappointing failures of other medical collaborations and poor business handling of its AI and support technologies, the ambitious plans of Watson Health were grounded to a halt by 2018.
Google Deepmind Health
Like IBM, Google has leveraged its own AI subsidiary DeepMind towards the business of healthcare. DeepMind’s AI technology revolves around the field of deep learning, and uses algorithms structured like neurons of the human brain to solve given problems.
With its acquisition by Google in 2014, DeepMind also brought to the table a partnership with the National Health Service (NHS) of the United Kingdom, laying the foundation for collaborations with health institutions across the country. For example, DeepMind’s work with the Moorfields Eye Hospital NHS Foundation Trust led to the development of an AI algorithm in 2018 that could accurately identify 50 types of common eye diseases 94.5% of the time based on digital eye scans, an accuracy that was on par with or better than diagnoses made by retina specialists.
Despite preclinical successes by DeepMind, its intimate ties with the big tech giant (and its parent company) Google have given people pause towards its healthcare ventures. Training new AI systems requires big databases as input – in the healthcare industry, this means patient information. In 2015, in collaboration with the Royal Free London NHS Foundation Trust, DeepMind developed a mobile application called Streams that would alert clinicians of patients at risk of acute kidney injury based on their medical history and other health metrics. However, a leaked copy of the data-sharing agreement between DeepMind and Royal Free showed that the AI subsidiary was granted access to detailed information of 1.6 million patients without their consent. This included sensitive information such as HIV status, details of drug overdoses, and other medical tests. A 2017 ruling by the UK’s Information Commissioner’s Office declared that this data-sharing agreement failed to comply with the Data Protection Act. “Patients would not have reasonably expected their information to have been used in this way,” said Information Commissioner Elizabeth Denham.
In 2021, Google has announced its termination of the Streams app, but the worries over who has access to the medical records remains. Compounded with the issues of digital privacy in the age of modern technology, a greater emphasis must be placed on data oversight to ensure transparency in collaborations between private companies and public health institutions. As Information Commissioner Denham puts it, “the price of innovation [doesn’t] need to be the erosion of legally ensured fundamental privacy rights.”
The future of AI in healthcare
Without question, AI has the potential to revolutionize our healthcare systems, particularly as our lives continue to become more entrenched in technology. However, clinical AI use is currently very limited. First, many developers in healthcare AI want to focus on solving complex problems such as medical diagnostics, as that is where there is the greatest need. However, current AI technologies are simply not accurate enough to reliably fill this role, as we have seen with IBM’s Watson Health cancer therapy initiatives. Secondly, there is a general lack of transparency in how these AI systems function, which makes it difficult for both patients and medical professionals to trust in a machine-generated medical opinion. The field refers to this as the black box problem – data goes into the AI system, an answer comes out, and we have no idea what happens in the middle.
To improve the use of AI in healthcare, there needs to be a better integration of AI systems into the existing clinical protocols. For example, AI could be focused into simpler tasks and areas of greatest inefficiency, such as administrative applications related to patient record organization and insurance. By finding ways for AI to augment rather than replace existing roles in the clinic, it can be developed as a tool to support human-focused decision making to build a sustainable framework for future advancements in AI technology.
Latest posts by Karen Yeung (see all)
- The Business of Healthcare - December 20, 2021
- From physical to digital: The technological age of contact tracing - October 4, 2021
- Biomimicry: The Nature of Design - December 23, 2020