ML/AI in the Healthcare Industry: Near Term Applications and Long-Term Challenges - Part I
We recently had the chance to present a seminar on the recent advances in machine learning and artificial intelligence. Advances in machine learning and artificial intelligence, in general, are progressing very fast. However, introducing these models into marketable or clinically usable products is not going that fast. The healthcare industry faces multiple challenges that will delay the application of intelligent, data-based systems to medical diagnosis and treatment in particular.
Artificial intelligence talks or seminars usually try to do two things: either scare you to death with bleak prospects of a mighty AI that makes humans obsolete or a fantastic Starwarsdesque universe with hyper-friendly and very supportive AI robots. Humanity is in the middle, waiting to be happy ever after or annihilated.
The situation, in our opinion, is that the implications of early adoption of AI, in general, are more similar to the early stages of the industrial revolution.
The industrial revolution, the substitution of human muscular power by steam-powered machines, brought, eventually, a higher standard of living globally. The early stages brought periods of meager standards of living, specifically in booming urban locations. The situation is similar now; human mental power is going to be replaced by digitally powered machines. The game is the same, but the stakes are one notch higher: we are staking the last thing humans have, our mind. AI-powered systems are and will be self-efficient. No moral or societal norm can stop a group of “renegades” using robots, be it healthcare, construction, or, in the worst possible scenario, war. This landscape has to be navigated, not avoided, as it is sure that the machines will eventually overwhelm us as the steam engines did in the past.
The healthcare industry can claim the absolute necessity for human contact across most of its value chain. This is what every other sector looking to turn their heads from automation was arguing about before the COVID19 pandemic hit. This very informative piece from MIT Sloan presents a harsh reality against the excuses for non-digitalization that many sectors use as shields:
Assumption: Customers value the human touch.
Reality: COVID-19 proved that a well-architected digital experience could offer an equivalent or even a more personalized transaction than an in-person engagement.
Assumption: Regulation inhibits digital transformation.
Reality: During the pandemic, highly regulated industries like health care were open to addressing barriers like privacy concerns for much-needed services like telehealth visits.
Assumption: It’s prudent to be a “fast follower.”
Reality: Evaluating others’ innovation efforts before taking action on your own wasn’t an option during COVID-19 and isn’t a good idea as we advance. Companies operating as fast followers during COVID-19 were more likely to lag behind competitors and miss opportunities, putting them at higher risk for business closures.
Assumption: IT can’t keep pace with digital transformation efforts.
Reality: Tech organizations across every industry stepped up to keep operations going and revenue coming through the door.
Assumption: People won’t pay the total price for digital-only.
Reality: Consumers paid for digital products and services — and will continue to do so post-COVID.
Companies, even whole sectors, actively hiding from digitalization and AI are running out of excuses. As a result, the introduction of these technologies is becoming a competitive necessity. Critically so in most cases.
The Healthcare industry can apply AI in two distinct areas:
1. Business Process Automation / Robotization (AI-BPA) as everyone else.
2. Artificial Intelligence Powered (AIP) healthcare products and services with truly unique implications.
We will discuss the first point in this post and the second point in a future post.
AI-BPA is possibly a minimum requirement for any company now. The competitive disadvantage of traditionally managing low and mid-value business processes is too high to be ignored, as they say in the logistics industry. If you are not automating, you are being automated. Healthcare industry companies, healthcare equipment providers, in particular, should be already engaged in or planning to engage in at least these areas of process automation:
Sale and upsale automated opportunity discovery. In this area, applying automated tender discovery and management is of great importance.
Automated customer service systems. The use of adequately working computerized claims management, at least partially, with a sufficiently efficient robot and human support. This is an area in which customer dissatisfaction with a lousy service AI can do more harm than good.
Sales forecasting and sales effort budgeting. Artificial intelligence can potentially forecast sales using a very high volume of data and determine where marketing and sales efforts are most effective.
The application of these AI tools is not free from challenges and caveats. To name a few, the main requirements that a company in the healthcare sector, or any other industry, in reality, must meet to start robotizing processes intelligently are:
Identified Decision Points:
The company must have clear business procedures, algorithmic in nature, with defined data inputs, data outputs, and actions. Even if machines do not run the company, even if it is humans following a set procedure, these procedures must be unambiguous, effective, and able to be followed by a computer system. Every management procedure left to ambiguous interpretation constitutes a source of bad data for an artificial intelligence trying to learn the business.
Existence of Process Data Mine:
The company needs to have process logs, conformance checks, and performance indicators of the existing processes. The company must be logging the operations in a machine-readable manner, database format, for example, to automate business processes with AI. Every step requiring a human-readable pdf or word file is ultimately machine-readable. However, the cost of reading and mining the data is much higher in terms of process decoding. The closer the data mine is to a database, the easier it will be for the AI to learn.
Existence of Sufficient Data Quantity:
There should be enough historical data or singular process examples for the AI to be able to learn. Then, even if the task or process is complex, AI models can understand it. The problem is that lack of examples can prevent the model from generalizing the task.
Sufficient Data Quality:
The quantity of data has to have enough fidelity. Business process steps that happen offline or are not correctly logged data may become poisonous for artificial intelligence. Undocumented process overrulings and offline exceptions will confuse any model. Documented overrulings are acceptable as the machine can learn these. Every (unrecorded) phone call or (unlogged) WhatsApp message slowly erodes your data quality.
Most companies, in general, not only in the healthcare sector, fail at one or more of these requirements and thus fail in their large-scale artificial intelligence business process automation projects. The sooner enterprise data is correctly managed, the earlier an artificial intelligence could provide management help and reduce the human load and costs.
The second part of this post series will present some of the currently existing healthcare-related AI models and the sector’s challenges in effectively applying them and placing them into the market.
If you require data model development, deployment, verification, or validation, do not hesitate and contact us. We will also be glad to help you with your machine learning or artificial intelligence challenges when applied to asset management, trading, or risk evaluations.