As keepers of the world’s ‘big data’, many of the major tech companies are naturally well placed to turn to the data-hungry development of artificial intelligence, or AI. And while AI is a hot topic in almost every industry, its presence and potential application in healthcare is exploding.

AI – what is it exactly?

In the simplest sense, AI is a computer system that can reason and perform a task in a human-like way. This can mean computer systems that are able to learn, communicate, make decisions and solve complex problems on their own.

Traditionally, AI’s application in medicine has been limited to the use of algorithms to analyse data and suggest diagnoses or recommend treatments. An example is the treatment of cancer. By feeding patient data into algorithms, a computer can review treatment alternatives and recommend a combination of chemotherapy drugs for a particular patient.

But these algorithms have, until recently, required intensive human effort to define and label data before subjecting it to analysis – necessarily, therefore exposing it to at least some level of human bias and error.

A more advanced step in the AI chain is “machine learning” – a type of computer data analysis that finds patterns in data and, in doing so, learns from the experience. Machine learning is, in other words, a kind of “smart” algorithm. It mimics the way the human brain processes data through probabilistic analyses.

Another form of AI is “deep learning”, a variant of machine learning which simulates neural “layers” that can each be trained to identify and integrate visual, aural and other variants to generate an outcome. The layers of software neurons each learn patterns in different features, like colour and shape. This enables deep learning systems to solve problems of far greater complexity than other forms of machine learning.

The promises and challenges of AI in medicine

AI has the potential to contribute to the diagnosis, treatment and prevention of disease, and provide more targeted healthcare to more people at lower cost. Through the use of predictive analytics, AI has potential applications across almost every facet of healthcare, spanning from administrative through to clinical contexts.

One example of an application of deep learning comes from experimentation with algorithms that scan the retina to assess blood pressure and risk for heart attack or stroke. The research, conducted by Google and its life sciences subsidiary Verily, relies on the fact that diseases like diabetes and high blood pressure can cause visible changes in the retina. The research involved scanning hundreds of thousands of patients’ eyes and embedding this data into the algorithms for review. The deep learning systems then established and learned patterns that reveal signs of pathology. The potential for this technology in risk prediction is, demonstrably, enormous.

AI might also have a role to play on the hardware side of healthcare through applications like robot-assisted surgery – another project Verily is undertaking in partnership with Johnson & Johnson.

But AI still has vulnerabilities. One weakness, in particular with deep learning systems, is a cultural one. Computer-based diagnostic tools, for instance, analyse data and produce predictions based on a series of algorithms – but do not provide the patient with how or why the conclusion was reached. Widespread implementation is dependent on significant demand and an enormous level of patient trust as well as an acceptance by clinicians and patients of digital monitoring and treatments.

Where does this leave pharmaceutical companies?

It’s clear that the development of AI applications in healthcare is changing the landscape of the healthcare industry, shifting dominance from traditional healthcare providers to big tech firms. This trend is a result of the fact that giving computers enough understanding of the world to do useful things requires quantities of data, processing power and computing infrastructure that only big tech has the means to mobilise and develop.

But AI in health does present opportunities for collaboration between tech and pharmaceutical companies. One example is in the field of digital therapeutics – or ‘digiceuticals’ – essentially, apps that are subject to regulatory approval processes and prescribed by doctors. Take diabetes, for instance. Diabetes digiceuticals essentially monitor markers through sensors or through user input, collate this information, and use it to more effectively manage the disease. The development of these apps is dominated by tech startups and they are being developed to treat a range of conditions such as addiction, depression and schizophrenia.

AI applications can also be adapted to meet pharmaceutical companies’ research needs, increasing efficiency and automation to drive R&D. Pharmaceutical companies have been partnering with AI-driven startups to apply AI to drug discovery by synthesising vast amounts of data relating to, for instance, toxicity, absorption, metabolism and excretion, to find promising drug candidates and even repurpose existing drugs to other indications.

What’s next

Stakeholders have been trumpeting the potential of AI to reinvent healthcare for decades. It seems we are finally at the point where AI can start to make good on promises to provide wider access to better health outcomes at lower cost. Some applications of AI, such as in the field of drug discovery, might take years to have a tangible effect on the industry. Others, like digiceuticals, are already empowering patients and changing the model of disease diagnostics and management.

Despite its challenges, AI projects are driving change in the healthcare industry. Industry players, healthcare providers, and patients – take note!