Obstacles to AI in Life Sciences: The Imperative of Good Data Hygiene

Reasonably, as AI operates on large sets of data, the accessibility of clean data at scale becomes a fundamental precursor to creating a suitable environment for the growth of AI-based solutions. It is the essence of the process.

FREMONT, CA: With many rough estimates being deliberated around artificial intelligence’s potential to alter healthcare for the better radically, it is easy to lose view of the fact that manifestation of that vision will be far from easy. Admittedly, the task at hand can seem quite tricky. Most pharma companies’ current IT infrastructures are based on legacy systems that were just not designed with Artificial Intelligence (AI) in mind: their data, if stored at all, is usually kept in clumsy free form and their systems lack interoperability.

Reasonably, as AI operates on large sets of data, the accessibility of clean data at scale becomes a fundamental precursor to creating a suitable environment for the growth of AI-based solutions. It is the essence of the process. Without this vital underlying infrastructure of big data, the potential of AI technology will surely fall short. The visceral reality is AI approaches can only be as good as the data applied to them. 

See Also: Top Pharma and Life Sciences Tech Solution Companies

If one puts junk in, they can only expect to get the junk out.

The first thing involved is to learn the importance of having outstanding data actually to base the machine learning on. One has to spend a lot of time and efforts just cleaning the data sets as a prerequisite to being able to run the algorithm as it might take years to clean the datasets. People tend to undervalue only how little clean data is available, and how hard it is to clean and link it all up.

The current abundance of poor data hygiene will probably mean that the practical introduction of AI-based healthcare will be uneven and unstable, at least in the preliminary stages. There are already some indicators of this happening. Healthcare, being a significantly regulated industry, is explicitly a data-rich, but that does not necessarily means that the statistics is balanced. For instance, in Asia to reap the benefits of AI, there is an urgent necessity to increase the availability of Asian-specific data. At present, existing genetics and clinical trial databases are mostly made up of Caucasian data, which means entire geographic regions and ethnic groupings risk being left behind as the technology takes off.