The AI in Life Sciences Market was valued at USD 902.1 million in 2019 by Mordor Intelligence and is projected to expand at an impressive 20 percent compound annual growth rate in the period 2020 to 2024. Yet despite the massive hype and enthusiasm around artificial intelligence’s propensity to positively disrupt the entire pharma and healthcare spaces, confusion still abounds when it comes to understanding exactly what this technology is and how it functions. Such haziness may be partly down to the fact that there are actually multiple forms of artificial intelligence comprising three distinct types: namely “human-created algorithms,” “machine learning,” and “deep learning.”
“AI, at its most fundamental, can best be described as software platforms comprising complex algorithms to analyse data, reach conclusions and anticipate problems with human-level expertise, but without requiring direct human input,” explains David Crean, managing director of the investment banking firm, Objective Capital Partners. “Currently, most computer-generated solutions emerging in healthcare take this form and do not rely on independent computer intelligence per se.” Indeed, running on human-generated algorithms, they represent evidence-based approaches programmed by data scientists. Once data is embedded into the algorithms, computers are subsequently able to extract information and generate insights of a quality, scale and speed that would hitherto have been impossible. Unlike traditional business intelligence or analytics, both of which typically rely on structured data such as the application of classical statistics such as variances, correlations, and regressions, AI, however, stands apart in its ability to harness diverse and unstructured data sets.
Going Deeper: Machine Learning and Beyond
In contrast to human-created algorithms, “machine learning” advances a step further by leveraging neural networks and multilevel probabilistic analysis. This is essentially a computer system modelled on the human brain that simulates the data processing of the human mind. “Machine-Learning involves continuous and repetitive processes by which machines are capable of integrating new data and improving performance over time without the need for explicitly programmed instructions,” clarifies Crean. Thus, even the programmers cannot tell how the computer derives the final solutions. Meanwhile “deep learning” represents a complex version of machine learning involving multiple layers of abstract variables, for analysis at a scope and speed far beyond human capability.
Unsurprisingly the deployment of such intricate forms in the life science space remains rather limited with clinicians and researchers no doubt somewhat cautious about adopting a black box capability that offers no clearly accessible rationale for its decision-making. “Once an AI system launches, it learns things that shape its future decision-making. With each new tranche of automated learning, the tool’s knowledge base becomes more difficult, if not impossible, for human programmers to discern,” warns Carla Smith, former executive vice President of the Healthcare Information and Management Systems Society (HIMSS). “Frankly, this raises serious ethical questions around who should be held responsible and accountable for decisions made by the machines,” she opines.
The Virtues of Keeping it Narrow
The style of AI which tends to be most widespread and in vogue right now in the life sciences domain is therefore a far less controversial, “narrow” artificial intelligence able to solve only a specific task or a group of tasks and confined to the specifics of the problem for which it is designed such as image recognition. “We are already seeing pockets of solutions such as chat-bots being developed to answer basic patient queries, with the aim to increase medication adherence,” notes Richard Saynor, CEO of generics heavyweight, Sandoz. Meanwhile, IBM Watson Health has won acclaim for supporting health providers by building an algorithm that pre-reads the mammograms and prioritizes all the clearly abnormal ones, redirecting a radiologist’s workload to where it is most needed, thus helping patients requiring treatment to gain access earlier.
“It really doesn’t have to be grandiose at the start,” stresses Adobe’s Senior Solution Consultant, Buck Dossey. “For companies just setting out on their first digital transformation journey, I recommend starting with digital enrolment which can trigger massive results without the need for a heavy lift,” he counsels. “Narrow AI plays a vital role in enabling humans to deliver infinitely more value. Such automated AI strategies can include auto-allocation, automated personalization, and automated targeting of content… while the Healthcare industry certainly boasts a multitude of opportunities to think big, that doesn’t mean that it’s not possible to attain laudable results by starting out on a smaller scale,” he argues.
Transcending Point Solutions
Sticking to the narrow does not mean staying small, however. For many insiders, AI in life sciences will only start to deliver upon its true potential when it can be rolled out at scale. “Right now, we are witnessing a proliferation of digital health start-ups and AI firms offering solutions in the healthcare space. In many instances, AI experts are already providing point solutions in hospitals. But my concern is that many are delivering very discreet insights based on a small portion of the population. Where the deployment of AI will really take off is when a critical mass is attained that allows for systematic and rigorous appraisal of scientific evidence available,” muses Mark O’Herlihy, managing director for EMEA of IBM Watson Health.