AI Enterprise Analysis

AI will affect a vast array of economic (both efficiency and distributional effects), environmental, social, equity, and political domains that need to be considered now (Table 2). Efficiency improvements driven by AI will involve a reduction of human effort to deliver services like radiological image interpretation, chart review, and care coordination. However, these gains also come with costs that may erode those gains. These erosions include the cost of generating health care services that are of no benefit, the costs of creating and maintaining AI infrastructure, and the environmental impact of storing and processing large volumes of data. Frontier models can greatly enhance efficiency in radiology workflows for cancer screening programs by rapidly reviewing images and automatically identifying lesions. While this increased speed and accuracy could improve care delivery, it might also lead to overdiagnosis, resulting in more frequent biopsies and follow-up procedures. Many of these identified lesions may not become malignant (or are cancers that will not affect health at all), meaning that patients could undergo services that offer no benefit and may even cause harm. Creating usable and scalable AI infrastructure will also be costly. Deploying these models in specific contexts poses significant challenges, requiring separate and standardized validation, making the process laborious and costly and out of reach for small independent health centers. The environmental impact of AI, requiring enormous amounts of electricity and water for cooling, raises the possibility of externalities (i.e., costs not borne by the parties that generate them) imposed on others. Typically, these kinds of externalities disproportionately affect low-income countries. All these consequences may potentially wipe out the cost savings derived from AI's efficiency improvements in diagnosis and care management. Moreover, the substantial investment needed to implement this technology could result in market concentration among a few firms, and the resulting exertion of monopoly power. As such, even if actual costs may go down, the cost to the consumer may increase substantially, generating high profits. This income redistribution may produce highly undesirable equity effects, with yet another small set of people generating massive personal wealth and political power, as has happened with many other industrial disruptors (e.g., railroad barons, telecommunication companies, social media creators) a pattern President Biden termed the 'tech-industrial complex' in his farewell speech, echoing Eisenhower's warnings from 60 years ago. The disruption brought about by AI could also threaten physician incomes as payers realize that AI allows them to spend less time on certain tasks. This realization may lead to a shift in payment structures, reducing compensation for physicians and potentially altering the traditional dynamics of healthcare delivery. Using radiology as an example again, payers may observe a significant reduction in the time required to interpret an image, prompting attempts to adjust reimbursement rates—a process often hindered by political interference from powerful lobbies, such as those involved in debates over the resource-based relative value scale in the USA. Initially, clinicians may welcome AI as an assistant, only to later realize it is supplanting their roles. Finally, it is important to note that major advancements in healthcare AI are not emerging from health systems but rather from corporate tech firms. Currently, the most effective Frontier models are developed by Google, Meta, and OpenAI—companies with a keen interest in entering health-care. However, this signals a noteworthy change in players, where traditional healthcare entities (such as hospitals, physicians, and pharmaceutical companies) are being gradually displaced by Big Tech. With the recent Google Antitrust lawsuit highlighting how monopolistic practices can stifle competition, we are edging towards a future where essential components of care deliv-ery could be concentrated in the hands of a few companies, potentially shutting out competitors. This shift is further compounded by the challenge of validating outputs from opaque AI models developed by Big Tech—such as chart summarization and clinical notes—which require continu-ous post-market surveillance, blending technical and manual oversight to ensure accuracy. Yet, in the rush to deploy, Big Tech may prioritize speed over safety, relying on a reactive approach where errors are addressed only after harm occurs. This change in players can be especially problematic, as current care delivery entities, such as physicians, pharma-ceutical companies, and hospitals, prioritize patient safety and well-being (e.g., the Hippocratic Oath); improving health is in their mission statement. AI can shift sensitive personal health data and the future of healthcare delivery into the hands of technology firms, which may not have the same ethical and equity principles foundational to health-care. These firms may view healthcare as one of several revenue-generating avenues, potentially compromising the ethical standards of care.