AI in Healthcare: Unpacking the Challenges

Artificial intelligence (AI) is often talked about as the next major step in healthcare, promising everything from faster diagnoses to more personalized treatments. However, as the healthcare industry cautiously explores AI’s potential, the reality is that AI’s potentially transformative impact is still on the horizon, with significant challenges and questions left to address.

I recently attended the WA Data Sciences Innovation Hub seminar on AI in Healthcare, where both healthcare and technology experts gathered to discuss current challenges and future possibilities. The seminar provided an overview of the benefits and potential of AI, the barriers preventing its integration into healthcare, and ethical and practical considerations that must be addressed.

Why Isn’t AI Being Adopted Faster in Healthcare?

A central question raised during the seminar was why AI, despite its rapid development, has not been more widely adopted in healthcare. The answer lies in the complexity of the healthcare system. While there is a clear need for AI particularly given the global shortage of healthcare professionals new technologies must first prove their clinical and economic value, as well as the strict ethical and regulatory standards that must be met.

For instance, AI-powered tools could revolutionize early diagnosis, assist with emergency department (ED) triage, and even perform surgeries. However, concerns regarding reliability, especially in critical environments like emergency rooms, make healthcare a high risk field when it comes to adopting AI. To expand on this point, an AI-powered triage system may not prioritize patients correctly, as it might classify another patient’s symptoms as needing more immediate attention. If the patient who wasn’t prioritized ends up being negatively impacted by this, this would be seen as a major failure and could lead to lengthy legal battles, despite the AI triage system having a high success rate and “saving lives” overall.

Another key point from the seminar is the global shortage of healthcare professionals, which ironically is slowing down the adoption of AI, despite the fact that AI could help address this exact issue. AI requires extensive testing alongside healthcare professionals to validate its reliability, accuracy, and success rate. However, overburdened healthcare professionals will find it challenging to dedicate time and effort to this testing, especially when immediate patient care is a higher priority.

Opportunities for Immediate Impact

While AI’s role in clinical settings often receives the most attention, another crucial area where an immediate difference would be seen is in administrative processes. Healthcare professionals frequently spend significant time on documentation, policy writing, audits, and report generation. AI-powered tools can drastically reduce this administrative burden, improving overall healthcare delivery by allowing staff to focus more on patient care. This would be a huge resource, especially given the ongoing shortage of staff as well as the lack of time for priority tasks.

For example, AI systems are already being used to generate medical reports much faster than humans. A report that once took hours to complete can now be created in just 20 minutes. Similarly, policy documents, clinical notes, and other time-consuming paperwork can be streamlined with AI, allowing more time for healthcare workers.

However, these AI tools, while offering substantial time savings, also introduce ethical challenges that must be carefully managed. For example, digital scribes AI tools that record conversations between doctors and patients and transcribe them are becoming more common but raise concerns. For example, recording consultations could discourage patients from sharing sensitive information, such as cases of abuse, due to privacy and safety concerns. Additionally, there’s a risk that healthcare workers might become overly reliant on these tools, potentially losing critical skills such as report writing and careful clinical note-taking.

Privacy and Accuracy

A recurring theme at the seminar was the balance between privacy and accuracy when using AI in healthcare. The integration of AI systems raises critical questions about data security and patient confidentiality. Many AI applications require vast amounts of data to function, and handling this data—especially in healthcare, where privacy is paramount—is a significant concern.

Moreover, the accuracy of AI systems in healthcare is not yet where it needs to be. While tools like AI-driven diagnostic systems or automated triage are promising, they still have a significant margin of error that can be life-threatening. Nonetheless, AI systems are constantly improving. The key takeaway was that we must continue testing AI tools parallel to traditional methods, ensuring their results are both accurate and reliable before they are widely implemented.

Addressing Bias in AI

One of the critical challenges highlighted during the seminar was bias in AI, which is rooted in the non-representative data that these systems are trained on. However, this bias is not unique to AI, in fact ethnic minorities and women have historically been more likely to be misdiagnosed due to data gaps in healthcare. The introduction of AI, trained on the same non representative data, risks amplifying this problem.

For instance, much of the current data on concussions comes from male athletes, which makes the diagnoses of concussions in women more challenging. When AI models are trained on these male-dominated datasets, the risk of misdiagnosis in female patients increases, perpetuating the same bias that existed prior to the implementation of AI.

Similarly, patients from ethnic minorities—such as Indigenous Australians or African populations—have long been at greater risk of misdiagnosis because the existing data has been collected from Western populations. Again, with AI being trained on these biased datasets, the problem escalates, further reducing the accuracy of diagnoses for non-Western patients.

In addition to gender and ethnicity, genetic diversity plays a role. What may be considered normal in one population could be flagged as abnormal in another due to genetic differences. Without accounting for these factors, AI systems risk delivering false positives or negatives, leading to incorrect diagnoses and therefore treatments.

Addressing this issue requires a commitment to building diverse and representative datasets that account for differences in gender, race, and genetics, ensuring that AI systems can offer equitable care to all patients.

Can AI Be Moral?

One of the more complex discussions at the seminar focused on the ethical use of AI in healthcare. A key question raised was - How do we ensure that AI systems act ethically when there is no universal agreement on what constitutes ethical behavior? This challenge becomes even more complicated when considering the cultural differences that influence what is considered ethical.

Morality, like ethics, is subjective and varies from one person to another. This presents a major hurdle in the development of AI, as morality cannot be programmed. A commonly cited example is the “trolley problem”, which is often used to illustrate moral dilemmas in decision-making. In healthcare, this dilemma mirrors the tough choices around resource allocation. For example, should the AI prioritize patients with a higher chance of survival, or those who are in the most immediate need? These types of decisions are morally complex, and AI, which is driven by data and algorithms, lacks the human empathy and judgment needed to navigate such scenarios.

Adding to the ethical concerns is the issue of accountability. If AI makes a critical mistake—such as misdiagnosing a patient or incorrectly prioritizing treatment—who is held responsible? Is it the healthcare provider, the hospital that implemented the AI, or the developers who built the system?

Current regulatory frameworks have yet to fully address these challenges. While the EU has started to establish laws around AI, Australia is still in the process of developing regulations. This regulatory lag leaves significant gaps in accountability, making it difficult to determine liability when something goes wrong.

Real-World Limitations

Another takeaway from the seminar is the importance of standardizing inputs and ensuring that AI systems are trained on data that reflects the environments in which they will be used. In a clinical setting, for example, AI systems need to account for the variability in human behavior. Something as simple as a nurse holding a device slightly incorrectly could skew the results, hence the importance of training healthcare staff alongside the AI.

AI must work seamlessly across all platforms and devices, from hospital systems to the smartphones and tablets used in rural and remote healthcare settings. This requires a significant amount of testing and refinement to ensure the technology is reliable, regardless of the device being used .

The Path Forward

The WA Life Sciences Innovation Hub seminar made it clear that while AI has the potential to revolutionize healthcare, we must approach this transformation with both enthusiasm and caution. The benefits of AI, particularly in reducing administrative burdens and enhancing patient care, are immense. However, the ethical, regulatory, and practical challenges that AI presents will make the integration of AI into healthcare a complicated and slow process.

The future of AI in healthcare is undoubtedly promising, but realizing its full potential will require a collaborative effort across the industry. A collaborative effort will be required to ensure that AI is not only innovative but also ethical, effective, and reliable.