Introduction

The use of artificial intelligence (AI) in professional settings has spread far and wide over the past few years in hopes of automating monotonous tasks, minimising human error, aiding in optimising workflow, and improving predictive trends by quickly accumulating reliable insights.¹ AI has already been adopted in sectors such as finance, transport, and data security² but one area that is broadening its horizons on how to adopt the use of AI in its field is medicine. 

Many studies have already been conducted to elucidate promising areas for AI integration,³ but there are still a number of concerns and considerations at play before AI can be safely and successfully adopted in healthcare settings. These challenges can be broadly broken down into the following areas:

• A scarcity of standardised processes
• A lack of ethical guidelines
• Inadequate legal supervision
• Biassed data collection and subsequent training
• Social resistance

Understanding AI in healthcare

The word AI gets thrown around quite a lot, but what actually is it and how is it being used in healthcare? Simply put, AI is a machine that is given a huge amount of input (i.e. real-world data) and is then trained to take all that knowledge and perform human-like tasks with it such as looking for anomalies, picking out patterns or trends, predict future outcomes, and learn how to operate in the future from the information it is given - much like how we learn to converse in a party by observing and participating (i.e., getting real-world data) in social gatherings from young. 

There are a number of applications of AI in healthcare and we will go into detail on some of the areas, but the overall aim for incorporating AI in clinical settings is to exceed human capabilities in analysing, understanding, and presenting complex medical data in order to synthesise new ways to diagnose, manage, and prevent ailments.

Medical education is one of many areas in healthcare that have potential to benefit from the use of AI technology. For example, Nykan and colleagues⁴ created an AI-powered simulation training tool for certain surgery procedures as well as a virtual assistant that provided constructive feedback using the data collected in real-time while participants completed the training simulation. This showcases the potential power AI has to help educate future practitioners on medical or surgical procedures. 

Additionally, there are several studies and reviews that revolve around using AI to optimise hospital resources such as operating rooms and recovery rooms, as well as scheduling and managing appointments. 

With the data provided, AI technology has been able to identify which surgeries have a high risk of cancellation, predict the duration of patient stay depending on the type of surgery they undergo, and increase the efficiency of administrative tasks by simultaneously coordinating multiple hospital spaces.⁵ 

Major challenges in AI adoption

Despite the growing benefits of AI within healthcare, the public and professionals in the field are still hesitant to move forward in this direction - and rightly so. When incorporating something as new and powerful as artificial intelligence, we collectively need to proceed with due diligence on how to incorporate this new technology and to what extent we should rely on AI when it comes to matters of our health. 

Technical challenges

Data surrounding people's health is undoubtedly important and incredibly private, notions that are reflected by the confidentiality surrounding what is documented and how it is stored as well as the idea that during a data breach health records are a common target. Because of this, medical institutions are very hesitant to exchange patient data. 

AI technology needs very big datasets in order to properly learn to classify or predict future trends and while there are huge amounts of data accessible to the public, it has usually been stripped down and averaged before being reported. 

The issue extends a little deeper when we realise that these datasets need to come from medical institutions around the world in order for the classifications and predictions to be universally applicable. However, different countries have different practices when it comes to sharing and collecting health data, which adds another level of complexity to this specific issue.

Touching on the need for AI integration in healthcare to be universally accessible, another challenge is put forth when looking to obtain data that ensures inclusivity on several defining traits (i.e. gender, age, socioeconomic status, ethnicity) in order to avoid bias and unfair predictions or classifications.

Another technical challenge the healthcare industry faces in adopting AI stems from reports of medical data rarely being organised logically and thus can produce gaps or inconsistent data when compiled together. Although this issue may be minimised by the evolution of storing data in digital records instead of physical ones, the data that AI technology will be trained on will still be limited to some degree.^6,7 

Legal challenges

Compared to the rigorous assessments that healthcare workers go through prior to their employment and the strict codes of conduct they need to uphold daily (think GDPR), there is currently no unified consensus for global regulations regarding the use of AI in healthcare. 

These global laws are imperative to finalise before incorporating AI technology in order to prevent a potential new wave of crime (AI-crime). 

Although a seemingly straightforward issue, the legal challenges surrounding the use of AI currently have more questions attached to it than answers. For example, who else aside from legal experts should help create these laws? Who holds responsibility on the off-chance that AI-related infringement occurs? What is the extent of accountability of each stakeholder? What systems should be in place in order for these regulations to be reliably updated as time goes on?^6,7

Social challenges

As with the public opinion of AI integration in society in general, there is a divide of opinion in healthcare workers when it comes to AI adoption. The concern echoed the most is the fear that many healthcare jobs will eventually be taken over by AI, although there is still a substantial amount of time to go before AI technology is at the stage where it can carry out healthcare-based jobs devoid of human supervision completely. 

A broader social concern is a general mistrust or scepticism about AI technology and this mistrust will have a direct effect on the acceptance and integration of AI tools. The social challenges surrounding AI integration can be mitigated by improving education around AI and machine learning. This includes base-level knowledge as well as higher-level terminologies - especially when it comes to training future healthcare practitioners.^6,7

Ethical challenges

Since its early inception, a prevailing challenge surrounding AI has been to do with accountability. Similar to a point that was mentioned above in the Legal challenges section, a question that needs to be asked before the adoption of AI in assisting healthcare professionals is who is to blame if there is an erroneous decision. 

This question holds a lot of weight specifically in clinical settings as the AI technology used would be in high-stakes applications. Should the blame fall on the healthcare professional, even if they are not fully informed on how the AI tool is developed? Should the blame fall on the developer of the AI technology, even if they have no ties to the clinical setting? There is no black-and-white answer to these questions, so these issues will need time and varying input from different professionals to be resolved.

Similarly, we must be careful to not assume that an AI tool that is ‘rule-oriented’ and grounded in logic may not always present the most ethical suggestion to a situation. This is because, ultimately, the AI tools are only as good and robust as the (un) biased data that they are trained on. One practical example of this issue would be an AI that categorised pneumonia patients as high-risk patients but then proceeded to wrongly classify pneumonia patients who also had asthma as low-risk patients.^6,7

Summary

While the use of AI in personal and professional settings is an incredibly exciting prospect in terms of increasing our own time management and allowing us to think of what else we can give our attention to, there are many challenges that need to be overcome before we can safely rely on adopting AI into our workforce. 

Namely, there are several legal, ethical, social, and technical difficulties that have yet to be resolved. A lot of these issues can be improved through reliable and accessible education on the development and applications of AI. However, there also needs to be a bigger initiative from various stakeholders to address these challenges in a collaborative and moral setting.