AI’s (Artificial Intelligence) limitations in the medical is rapidly infiltrating health care and playing important roles ranging from automating drudgery and routine tasks in medical practice to managing patients and medical resources. As developers create AI systems to perform these tasks, several risks and challenges emerge, including the risk of patient injury due to AI system errors, the risk of data acquisition and AI inference compromising patient privacy, and more.
While AI has a number of potential benefits, it also has a number of risks:
Injuries and human error
The most prominent of AI’s limitations in the medical field is the obvious risk that AI systems will occasionally be incorrect, resulting in patient injury or other health-care issues. If an AI system prescribes the wrong drug to a patient, fails to detect a tumor on a radiological scan, or assigns a hospital bed to one patient over another because it incorrectly predicted which patient would benefit more, the patient may suffer harm. Of course, many injuries occur in the health-care system today due to medical error, even without the involvement of AI. AI errors may differ from one another for at least two reasons.
Initially, patients and providers may react differently to injuries caused by software than to injuries caused by human error. Second, as AI systems become more widely used, an underlying problem in one AI system could cause injuries to thousands of patients, rather than the small number of patients injured by any single provider’s error.
Data accessibility
To train AI systems, you need large amounts of data. You can get this data from electronic health records, pharmacy archives, insurance claims records, or consumer-generated information such as fitness trackers or purchasing history.
However, health data have been frequently problematic. Data is typically dispersed across numerous systems. Aside from the variety mentioned above, patients frequently see different suppliers and shift insurance companies, resulting in data fragmentation across multiple systems and formats. This fragmentation raises the risk of error, reduces the comprehensiveness of datasets, and raises the cost of data collection, limiting the types of entities that can develop effective health-care AI.
Concerns about privacy
Another set of AI’s limitations in the medical arises in the context of privacy. The requirement for large datasets incentivizes developers to collect such data from a large number of patients. Health insurance may have to share patients’ information with companies in order to collect premiums As a result of data-sharing among health systems and AI developers. AI may also jeopardize the privacy in another way: AI can foresee private patient data even if the algorithm has never seen that information. For example, an AI system may be able to detect Parkinson’s disease based on the trembling of a computer mouse, even if the person has never disclosed this information to anyone else (or did not know). Patients may regard this as a violation of their privacy. This is particularly true if the AI system’s inference is made available to third parties, such as banks or life insurance companies.
Inequality and bias
In health-care AI, there are risks of bias and inequality. AI systems can learn from the data that they receive, these data include biases that they can learn from.For example, social media companies have access to regional data with little advertising, therefore this data increase Facebook’s advertising revenue with little loss to its customers. Similarly, when using speech-recognition AI systems to transcribe encounter notes, such AI may operate worse when the provider is from a background or gender. Even if AI systems are accurate, representative data, there may be issues if that data reflect underlying biases and inequalities in the health-care system. For example, African-American patients receive less pain treatment than white patients on average; an AI system learning from health-care records may learn to recommend lower doses of painkillers to African-American patients, despite the fact that this decision reflects systemic bias rather than biological reality.
Professional repositioning:
The long-term dangers include changes in the medical field. Some medical specialties, such as radiology, are prone to dramatically alter as more of their work is automated. Some academics are concerned that the widespread use of AI will lead to a loss of human knowledge and capacity over time, such that providers will lose the ability to detect and correct AI errors while also developing medical knowledge.
Solutions
Potential solutions are complex, but they include investments in infrastructure for high-quality, representative data, collaborative oversight by the FDA and other health-care actors, and changes to medical education to prepare providers for shifting roles in an evolving system.
ALSO READ: Phasic Policy Gradient: A Game-Changing Approach To AI
Conclusion
AI has a wide range of applications in the healthcare industry. However, there are some limitations of AI in the medical field. It is impossible to make AI work with the current health care systems. Lack of quality medical data, inapplicable performance metrics and flaws in research are some of the limitations. The solutions would be to work on them.
ALSO READ: What Are The Challenges Of AI in Health Care Services?
