Wilhelm Haverkamp

Background Integration of artificial intelligence (AI) into clinical practice promises diagnostic improvements while raising fundamental questions about physician autonomy and competence development. Objective What are the medical–ethical implications of potentially deskilling medical expertise through AI systems, particularly regarding professional autonomy, responsibility attribution, and intergenerational justice? Methods Conceptual ethical analysis applying principles-based, virtue ethics, and responsibility ethics approaches, supplemented by findings from review and concept papers and prospective pilot studies on AI implementations in clinical practice. Results Unreflective AI integration may lead to gradual erosion of core physician competencies, even with high diagnostic performance. Particularly vulnerable are physicians in training who may never develop full clinical autonomy when foundational competencies are acquired primarily under AI assistance, raising questions of intergenerational justice and reversibility of technological transformations. Conclusion From a medical ethics perspective, protective measures that enable innovation are required while preserving medical expertise and autonomy. As its original contribution, this paper develops a three-phase model for the ethically responsible integration of AI into medical training. The crucial question is not whether AI should be implemented, but how.

This essay examines the emerging phenomenon of recursive AI development, in which artificial intelligence systems increasingly participate in designing, training, and improving their successors. The argument is that a qualitative threshold has been crossed: “AI begets AI”—a self-reinforcing cycle of machine-driven innovation that redefines the boundaries of human agency. Drawing on developments in automated machine learning (AutoML), AIassisted coding, and synthetic data generation, the essay suggests that artificial intelligence is evolving from a human-directed tool to a self-propagating system. This transition marks not merely technological progress but a new stage in technological evolution—one that operates at scales and speeds exceeding human comprehension. Understanding this transformation, and finding epistemic and ethical frameworks to navigate it, is essential for maintaining human responsibility within a world increasingly shaped by machines that design machines.

This paper introduces ‘I-tuning’ - the individualised fine-tuning of large language models on personal data corpora - as a qualitatively new form of cognitive augmentation. Two variants are distinguished. I-Tuning-Lite (training on recent communications, writing, and interaction logs) is near-term feasible and already approximated by current personalised writing tools. I-Tuning-Full (training on a comprehensive longitudinal record spanning medical, developmental, behavioural, and experiential data from early life onward) is a speculative but technically coherent endpoint warranting preemptive governance. Claims in this paper are calibrated to the distinction: risks attributed to I-Tuning-Lite are supported by existing evidence; risks attributed to I-Tuning-Full are advanced as theoretically grounded projections. Grounded in the extended mind tradition (Clark and Chalmers, 1998) and drawing on Clark (2025), who argues in Nature Communications that generative LLMs constitute a qualitatively new form of cognitive extension, both variants are defined as points on a gradient of integration running from the notebook through the smartphone and the generic LLM to a personalised generative system whose outputs become difficult to distinguish from the user’s own thought. The paper argues that this design goal generates constitutional risks - risks that concern not the accuracy or control of information about the person, but the integrity of the cognitive processes through which the person knows, decides, and identifies herself. A tripartite framework identifies three such risks. Epistemic opacity denotes erosion of the ability to distinguish self-generated from model-generated thought. Agential drift describes the unacknowledged transfer of decision-making authority through habituation. Identity confusion - advanced as a philosophical projection rather than an established empirical finding - names first-personal uncertainty about cognitive boundaries when a system trained on one’s reasoning and values generates outputs one can no longer confidently attribute to oneself. A retrospective endorsability criterion provides a principled boundary between harmful integration and successful extension, and grounds a structured response to libertarian transhumanist objections. A tiered governance framework - mandatory constitutional disclosure, architectural transparency features, and periodic professional capability assessment - follows from the structure of the harms.

This essay explores how large language models invert Claude Shannon’s classical distinction between signal and noise in information theory. Where Shannon sought to isolate meaningful signals from noise, LLMs are trained on the entire “noise” of human textual production—errors, contradictions, and redundancies included. The essay argues that these models don’t transmit meaning but approximate it through statistical correlation, transforming noise into seemingly coherent signals. Drawing on Shannon’s mathematical theory of communication and contemporary developments in generative AI, the paper demonstrates that meaning in LLMs emerges not from reference to reality but from pattern recognition within linguistic noise itself. This represents a fundamental philosophical shift in how we understand machine-generated language. The essay concludes that our task is not to silence this noise but to maintain critical distance, recognizing that understanding begins with doubt, not fluency, and that meaning is not what survives the noise but what resists becoming it.