AI algorithms can sometimes make mistakes in their predictions, forecasts, or decisions. Indeed, the very principle of such models’ construction and operation is fallible due to the theory of complexity [102].
If AI is inherently fallible because of the complexity in its design, how much trust should we really place in its predictions when it comes to critical areas like healthcare?
One study reports, for example, that “the use of medical cost as a proxy for patients’ overall health needs led to inappropriate racial bias in the allocation of healthcare resources, as black patients were erroneously considered to be lower risk than white patients because their incurred costs were lower for a given health risk state” [95]
How can we ensure that AI systems in healthcare don’t reinforce existing racial inequalities when the data they rely on already reflect those biases?
Tools ostensibly sold for healthcare or fitness (e.g., smart watches) become monitoring and information-gathering tools for the firms that collect these data [80].
This point really makes me question the true purpose behind popular health technologies. While smartwatches and fitness apps are marketed as tools to improve ourselves, it’s unsettling to realize they also serve as data collection devices for large companies. I’ve noticed how these devices constantly encourage users to share more information, which makes me wonder if health improvement is just a cover for a way to make profit. It’s a clear example of how convenience and self-tracking can blur into surveillance, raising important ethical questions about consent and corporate transparency.
Each of these three elements, however, differs depending on the individual’s level of AI literacy and other subjective characteristics (i.e., psychological, cognitive, or contextual), the interpretability of the algorithm used, and the amount and accuracy of information given to the patient.
I think this point really emphasizes how personal and situational our interactions with AI can be. It makes sense that someone’s level of AI literacy or their psychological traits would affect how much they trust or understand an algorithm’s decision. I’ve noticed this myself when using health or fitness apps, if I don’t fully understand how the technology works, I tend to question its accuracy more. This idea also raises an ethical concern: if people have unequal access to AI education or information, then their ability to make informed choices could be unfairly limited.
From an ethical point of view, issues of privacy are rooted in conflicting moral values or duties. The very concept of privacy has been defined in many ways in the ethics literature, with its origine intertwined with its legal protection [45], so it can hardly be summarized through a single definition.
I find it interesting that the authors highlight how privacy can’t be defined by a single, universal meaning. From my perspective, that really shows how complex the issue is. What one culture or generation views as a right to privacy, another might see as unnecessary secrecy. It also makes me think about how technology has blurred these moral boundaries even more, especially with social media encouraging people to share so much of their personal lives. The idea that privacy is tied to both ethics and law suggests that our understanding of it changes depending on social norms and legal systems, which I think explains why it’s such a difficult issue to regulate fairly.
normativity
Normativity means to establish, relate to, or deriving from a standard or norm, especially of behavior.
computational
Computational means relating to or using computers.
implemented
The term implemented means that something like a plan, policy, or idea has been put into practice or carried out.
autonomous
The term autonomous when it comes to learning means a process where individuals take responsibility for their own learning by setting goals.