Artificial Intelligence and Human Upskilling

When Is AI Good for Us?

Editor’s Note: This post is a part of a series of responses to Reclaiming Human Agency in the Age of Artificial Intelligence curated by Alessandro Rovati, the Associate Editor of the Journal of Moral Theology

Photo by Gaurav Tiwari on Unsplash

It was recently reported by the Pew Research Center that Americans are now “more concerned than excited” about the increased use of AI in daily life, with 82% of respondents indicating extreme or moderate concern that “people’s ability to do things on their own will get worse because of AI use,” and a majority of adults under thirty saying that “the increased use of AI in society will make people worse at thinking creatively (61%) and forming meaningful relationships with other people (58%).” While positive views of AI remain prevalent among technology experts, only 17% of the American public believes that it will have a “positive impact on the United States over the next 20 years.” It is against this backdrop of growing unease about the effects of AI on our common life that the A.I. Research Group for the Vatican’s Centre for Digital Culture has released its second volume, Reclaiming Human Agency in an Age of Artificial Intelligence. Building on its previous study, which explored a variety of theological, anthropological, and ethical issues related to AI, Reclaiming focuses more narrowly on questions related to human freedom, practical judgment, and moral action. In particular, it considers how AI can endanger human dignity by undermining or replacing human agency. The volume understandably emphasizes the growing number of threats that AI poses to the responsible exercise of human agency. Yet despite its predominantly critical posture, it admirably resists the temptation to adopt a wholly negative assessment of the technology.

In this respect, Reclaiming reflects the vision of Pope Francis (and of Catholic social thought more broadly), who observed in his 2024 address to the G7 that AI, like any other technology, “arises precisely from the use of [our] God-given creative potential.” Drawing on this insight, Reclaiming argues that our attentiveness to the unique risks of such a powerful technology must be balanced by a recognition of its potential to benefit society. When properly designed, AI can serve as “a tool that genuinely supports human flourishing and responsible action rather than eroding it” (11). By highlighting the challenges involved in preventing harmful uses of AI technologies, Reclaiming ultimately seeks to promote a human-centered design process, where AI systems are intended to “complement rather than replace human skills” (184).

Reclaiming discusses several ways in which AI can be used to promote human flourishing by supporting and enhancing human agency. Some of these applications have to do with the ability of AI systems to process data at a scale that far exceeds and outpaces the capacities of the human mind. For example, it is mentioned how AI can be used to quickly analyze vast amounts of satellite data in order to more effectively track the effects of climate change and land use (89). Similarly, the book highlights the Nobel Prize-winning program Alphafold, which uses deep learning to predict unknown protein structures, as an exemplary instance of AI augmenting human agency by “[performing] a task at which humans do not and will never excel” (185). In each case, AI “solves a problem that otherwise cannot be solved” (89), since the data is either so complex or so extensive that we lack the ability, let alone the human resources, to analyze all of it. With these applications, it is clear that AI is being used to enhance and augment, rather than replace or weaken human agency.

Other applications, by contrast, involve tasks that human beings could in principle perform but often do not, due to various practical constraints. Reclaiming notes, for example, that “AI applications have provided support for people with many kinds of disabilities,” such as text-to-speech apps that give people with blindness or dyslexia greater access to written content (2). The book also cautiously endorses tools such as Ambient AI, which uses speech recognition technology to automatically document medical appointments in electronic health records (185). While this task could just as easily be performed by a human scribe, the cost of employing such scribes has led most medical practices to require their practitioners to fill out this record themselves during patient visits, often at the cost of sustained attention to the patient. By transcribing and summarizing clinical conversations, an AI scribe frees up the practitioner’s attention and thus “supports rather than undermines the practitioner’s agency toward intersubjective engagement with the patient” (186).

This last example raises further questions precisely because it suggests that AI can support the agency of a medical practitioner in performing the (primary) task of diagnosing and treating the patient by eliminating—or at least reducing—the need for his or her involvement in the (secondary) task of keeping a medical record. As Reclaiming makes clear, however, this is only the case because “[t]here is no particular human excellence to the bureaucratic task of filling out a chart that Ambient AI replaces” (186). In other words, charting does not appear to be a task that involves “higher-order levels of thinking” or “offers valuable challenges for human development,” unlike activities more closely connected to the ends of medicine (86). While one might reasonably question whether documenting a medical appointment is as mechanical an activity as Reclaiming suggests, one can readily understand how it differs from, say, a student using an AI scribe to take notes during class. In this case, the task AI replaces is one that is not merely secondary and instrumental, but intrinsic to the practice and aims of education. By eliminating or reducing human involvement in an activity through which certain intellectual virtues (e.g., attention, interpretive judgment, and critical thinking) might otherwise be cultivated, an educational AI scribe would have the effect of disempowering rather than supporting the agency of the student.

Thus, drawing on a distinction popularized by the philosopher Shannon Vallor, Reclaiming contends that AI can support human agency not only through the “upskilling” of human agents, but also through certain forms of “deskilling” that allow them to redirect their energies elsewhere, while nonetheless preserving the virtues necessary for excellence in a given activity. On this view, deskilling is “neither good nor bad in itself” (8), but depends on the moral salience of the task being replaced. If it is a task that “[offers] great scope to judgment, creativity, empathy, practical wisdom, and critical thinking,” we should expect that deskilling will also result in a certain amount of “de-virtuing” (87). The authors recognize, however, that because the moral salience of particular skills is not always obvious, judgments regarding whether a given form of deskilling will ultimately promote or impede human flourishing can often be made only in hindsight (80). Indeed, the question of whether a given technology should be viewed primarily as an instance of upskilling or deskilling may be difficult to answer with any degree of certainty prior to its widespread adoption.

Consider how AI is now being used by radiologists to interpret X-rays, CT scans, and MRIs. There are certain ways in which AI unambiguously “upskills” the practice of radiology; for example, by enhancing the quality of images and flagging which scans need immediate attention. AI has also helped uncover “radiomic” biomarkers in diagnostic images that correlate with disease in ways that are imperceptible to human radiologists. It remains to be seen, however, how AI will reshape how radiologists approach the core task of identifying abnormalities in diagnostic scans. Technology experts have long predicted that AI will eventually eliminate the need for a human radiologist to be involved at all in this task. Others argue that AI will not replace but rather enhance the performance of radiologists by improving their efficiency and helping them attain a consistently higher level of accuracy in their scan analysis. We now have a growing body of evidence that supports this latter view.

However, a recent study published in Nature complicates the picture somewhat. While AI assistance may improve performance on average, its effects are uneven, with some radiologists even performing worse while using AI. A key finding of the study is that “radiologists struggle to consistently distinguish between accurate and inaccurate AI predictions and can be misled by inaccurate AI predictions.” Moreover, it showed that “[c]linicians who had low performance at baseline did not benefit consistently from AI assistance.” These findings raise the question of whether the full integration of AI with the practice of radiology may lead not only to beneficial upskilling, but also to a degree of deskilling that risks compromising the effectiveness of clinical medicine. Precisely for this reason, some experts have recommended using AI only as a “second set of eyes” after a radiologist has already interpreted a scan without assistance.

This mixed assessment of AI-assisted radiology reflects the broader ethical and professional challenges that accompany the integration of AI into any complex human activity that engages higher-order levels of thinking. Reclaiming rightly observes that we must consider how AI “can be used to assist and enhance the essential human skills and virtues in these tasks and not replace them” (87). Yet, as we have seen, it may not only be difficult to determine whether a certain AI application is primarily an instance of upskilling or deskilling; it may also be the case that the same technology can have different effects on different users, improving the performance of some while diminishing that of others. The authors of this excellent volume have helpfully framed the contemporary debate about AI’s effects on human agency in light of a Catholic vision of human flourishing. Yet the task of discerning how this vision should inform our judgments about specific AI applications remains in its early stages. One hopes that the members of this research group will continue to explore how the various AI technologies that are rapidly reshaping our economy and society might be properly directed toward a genuine vision of human flourishing that supports rather than undermines human agency.

Editor’s Note: For more conversations with the Journal of Moral Theology’s authors and responses to their essays, check out HERE.