Yael Friedman

• Friedman, Yael & Smajdor, Anna (2026). Synthetic Medical Data vs. The Four Principles: Ethical Trade-Offs and Zombie Data | Health Care Analysis | Springer Nature Link. Health Care Analysis. ISSN 1065-3058. doi: 10.1007/s10728-026-00596-9. Full text in Research Archive Show summary

  Rapid advances in AI for medical applications are accompanied by an increasing demand for vast datasets. Privacy constraints – legal restrictions such as GDPR, as well as more general ethical concerns – pose significant barriers to the acquisition and use of data. Synthetic medical data (SMD) is AI-generated data that mimics real patient information. SMD is being promoted as an ethical solution that preserves privacy while enabling scalable model training. However, as we will show, the use of SMD raises its own set of issues. SMD’s emphasis on privacy and scalability subtly embeds ethical and epistemic trade-offs that undermine the principles commonly regarded as being paramount in medical ethics. These trade-offs lack the kind of ethical justification one would expect in decisions that disrupt commonly accepted values and priorities, such as those captured by Beauchamp and Childress’ four principles. In the discourse on the ethical advantages of SMD, privacy tends to be treated as a value in its own right. We show that this is a problematic assumption. Transparency, privacy, scalability, and fidelity are all linked in complex ways with the principles of medical ethics. We map these relationships and show that when privacy is privileged over other values, it conflicts with autonomy and beneficence. Our aim here is not to establish that this privileging is wrong per se, but to show that it needs careful analysis before we can accept the idea that SMD is indeed ethically advantageous.

• Friedman, Yael (2026). Growing whole: The conceptual expansion of medical holism. Studies in History and Philosophy of Science. ISSN 0039-3681. 117. doi: 10.1016/j.shpsa.2026.102146. Full text in Research Archive
• Friedman, Yael (2024). Conceptual scaffolding for the philosophy of medicine. Medicine, Health care and Philosophy. ISSN 1386-7423. 28, p. 45–64. doi: 10.1007/s11019-024-10231-w. Full text in Research Archive Show summary

  This paper consists of two parts. In the first part, I will introduce a philosophical toolbox that I call ‘conceptual scaffolding,’ which helps to reflect holistically on phenomena and concepts. I situate this framework within the landscape of conceptual analysis and conceptual engineering, exemplified by the debate about the concept of disease. Within the framework of conceptual scaffolding, I develop the main idea of the paper, which is ‘the binocular model of plural medicine’, a holistic framework for analyzing medical concepts and phenomena. In the second part, I demonstrate the use and value of the binocular model by analyzing, through the lenses of the model, the phenomenon of health wearable devices and their effects on the concept of diagnosis

View all works in NVA

• Friedman, Yael & Smajdor, Anna (2026). Synthetic Medical Data. Full text in Research Archive
• Friedman, Yael & Smajdor, Anna (2026). Synthetic medical data: ethical panacea or invasion of the zombies…? Full text in Research Archive
• Friedman, Yael & Smajdor, Anna (2026). On synthetic medical data: Values hierarchies, reification, and zombies. Full text in Research Archive Show summary

  Synthetic medical data (SMD) has been introduced to address data-acquisition challenges. It has been described as a “magical” solution that enables model training without compromising patients’ privacy and is therefore considered ethically desirable (Bellovin et al. 2019; Savage, 2023). However, as we will show in this article, synthetic data is not value-neutral, and it poses both epistemic and ethical trade-offs that cumulatively affect the medical realm. We focus on three primary methodologies: two key deep learning approaches, namely generative models without privacy guarantees and differential privacy (DP) models. In addition, we will use vine copula models as our primary example to demonstrate the statistical approach. We show that prioritizing privacy over transparency undermines explainability, which in turn hinders trust and fairness, pushing these values down the ladder. We also show that prioritizing scalability over fidelity moves medicine away from personalized, participatory, and patient-centered approaches and toward a naturalistic, mechanistic understanding. The collapse of context and experiential knowledge into “zombie data” yields a narrow conception of utility in medicine that ignores the patient as a person. The absence of a direct real-world reference (combined with the use of this data to train algorithms) is where SMD makes a step forward in the reification process relative to “regular” medical data. The data collected to represent reality may gradually subsume the very thing it is intended to represent. As such, SMD also introduces a new degree of reification in medicine, promoting fungibility, which offers clear advantages for data-intensive enterprises, over patients and minorities.

• Friedman, Yael & Smajdor, Anna (2026). DATA, Synthetic DATA, and the spectrum of reification. Full text in Research Archive Show summary

  With rapid advances in AI technology, there is an increasing demand for data to train AI models. These models are used in medicine and health care for clinical research, diagnosis, drug development, and prevention. However, while significant amounts of health-related data are generated, large datasets may not be available due to privacy concerns and regulatory frameworks (e.g., GDPR). To obtain patients’ data, informed consent is required, therefore creating a high threshold for data sharing. In addition, for rare diseases or cases of “no diagnosis”, large data sets are not available. Synthetic medical data (SMD) has been introduced to address data-acquisition challenges; It has been described as a solution that enables model training without compromising patients’ privacy and is therefore considered "ethically desirable" (Bellovin et al. 2019, Savage, 2023). In this talk, we will introduce three main models of SMD: Generative models, Differential Privacy (DP) models, and the C-vine copula-based synthetic data generator. As we will show, these models are not value-neutral, and they pose both epistemic and ethical trade-offs. We will illustrate how SMD use creates new value hierarchies: Privacy over transparency: SMD can obscure the specific characteristics and nuances of real patient populations, which can create explanation gaps that undermine clinicians’ ability to interpret model outputs, erode trust, hinder bias mitigation (Susser et al., 2024), and knowledge development, and risk long-term de-skilling. These risks are especially relevant in the case of rare diseases, where small original datasets and oversimplified synthetic approaches can lose critical dependencies and idiosyncratic features, leading to misleading clinical insights. Scalability over fidelity: developers of SMD aim for low individual fidelity (to protect privacy) but high distributional fidelity, yet, as we will show, noise and selective-distortion methods can break crucial dependencies or entrench biased dependencies from the original data (ibid.), perpetuating the “garbage in, garbage out” paradox (in the process of keeping what seems as important dependencies in C vein copula models), or even introduce new biases (when adding noise in DP models). The preference for scalable, quantified data over individual experience holds risks of alienating patients, erasing contextual and minority experiences (creating what we call “zombie data”), expanding diagnostic disparities, and institutionalizing these trade-offs as a new normal. We will discuss how this prioritization fosters fungibility in medical practice, which we conceptualize as an advanced stage of the medical reification process. We will distinguish between the stage of “datafication”, which is characterize by decontextualization and translation of messy human experiences into transportable data points that creates a partial, curated technical form of objectivity (Leonelli,2016; Nguyen,2024), and SMD, which goes a step further, generating “zombie data” that deepens reification by stripping lived experience, context and agency from those experiences. Finally, we will argue that this new level of reification is a step forward in the process of commodification and, hence, in capitalism. References Bellovin, S. M., Dutta, P. K., & Reitinger, N. (2019). Privacy and synthetic datasets. Stan. Tech. L. Rev., 22, 1. Leonelli, S. (2019). Data-centric biology: A philosophical study. University of Chicago Press. Nguyen, C. T. (2024). Value capture. J. Ethics & Soc. Phil., 27, 469. Savage, N. (2023). Synthetic data could be better than real data. Susser, D., Schiff, D. S., Gerke, S., Cabrera, L. Y., Cohen, I. G., Doerr, M., ... & Wagner, J. K. (2024). Synthetic health data: real ethical promise and peril. Hastings Center Report, 54(5), 8-13.

• Valles, Sean & Friedman, Yael (2025). Philosophy of Biomedicine. Stanford Encyclopedia of Philosophy. doi: https:/plato.stanford.edu/entries/biomedicine/. Full text in Research Archive
• Smajdor, Anna & Friedman, Yael (2025). Epistemological and Ethical Dimensions of Synthetic Data for trustworthy AI. Full text in Research Archive Show summary

  The development of AI systems relies on vast quantities of data, much of which is acquired without meaningful consent from the individuals whose information is harvested. Synthetic data – artificially generated through statistical modelling, generative AI, or data augmentation – has emerged as a proposed solution to both the logistical challenges of data acquisition and the ethical dilemmas of using personal data. However, as we will show, synthetic data does not necessarily offer an ethically superior pathway towards the development of trustworthy AI, and it raises a number of epistemological as well as ethical concerns in its own right. The risks related to AI’s reliance on mass scale data-harvesting are well recognised. They tend to be characterised in three ways: privacy, property and plagiarism. Using people’s data may result in private information being leaked into domains where it cannot easily be controlled or removed. There is also a worry that in using people’s data, AI developers may be regarded as in some senses stealing another person’s property. And in terms of generative AI, it has been suggested that the use of a person’s data to feed the generative model can be viewed as a form of plagiarism. Synthetic data seems to offer a perfect solution to these three problems, in effectively divorcing data-originators from the data itself. Synthetic data is artificially generated using a variety of methods, including rule based systems, statistical modelling based on probabilities, generative machine learning, where the AI learns patterns from real data and then generates its own simulations, and data augmentation, where a ‘real’ dataset is infiltrated with ‘noise’ – random variations in the data – in order to make it harder to trace back data to its original source. (Importantly, it should be noted here that synthetic data still depends on ‘real’ data to begin with; synthetic data cannot be generated out of thin air – it might thus be argued that the challenges of provenance remain, but become harder to identify and track.) In epistemic terms, the push for synthetic data tends to prioritize scalability over fidelity, introducing significant reliability risks. For example, generating synthetic patient records (e.g., for rare diseases where real data is limited) to train diagnostic AI models without exposing real patient data raise an epistemological issue: can synthetic medical data accurately capture the complexity of biological processes, or does it introduce epistemic risks by oversimplifying pathologies? While we want to have more data on rare diseases, generating synthetic data put us at risk of generating misleading clinical insights if this data lacks fidelity. Rare diseases often involve idiosyncratic symptom presentations, genetic variations, and comorbidities that may not be fully captured in small original datasets. Synthetic data can lead to epistemic injustice by abstracting lived realities into quantifiable inputs, privileging dominant frameworks over contextual or subjugated knowledge. Take for example the privilege of genetic information over environmental information or phenomenological inputs. When data originators—particularly marginalized underrepresented groups—are excluded from shaping synthetic datasets, their realities are reduced to raw material for extraction, fostering epistemic alienation. The generalization of synthetic data sets creates a movement away from ideas of knowledge co-creation, solidarity and benefit sharing. Instead, it presents a sanitised product for companies to use as they see fit. In addition, while synthetic datasets can address privacy concerns by severing ties to real individuals, thus circumventing some legal and ethical problems they also reinforce the ‘garbage in, garbage out’ paradox: if the original data reflects biases or underrepresentation of marginalized communities (e.g., representation of mostly white Europeans males), the synthetic output will amplify these flaws under the veneer of neutrality. AI models may then fail to recognize symptoms in marginalized groups, worsening diagnostic disparities. Synthetic data will reproduce and amplify these gaps, leading to AI models that underdiagnose women and non-white patients. In short, rather than repairing knowledge gaps and injustices (e.g., testimonial injustice of minorities in health care), synthetic data can codify them as statistical inevitabilities; data that replicates existing biases naturalizes disparities as inevitable rather than contestable power structures. The generation of synthetic data can be viewed as part of a broader ongoing push towards datafication, and hence, reification. At its core, datafication is the rendering of human experiences, relationships, and biological processes, into fungible units of data. This transformation is neither epistemically nor ethically neutral. The individual is not treated as a subject with agency, but as a means through which valuable goods can be obtained. The term ‘harvesting’ is apt here. The data then becomes as a tool to serve ends that may have no connection with the data originator. In a further step towards reification, the data that is collected in order to represent reality, may gradually subsume the thing it is intended to represent. The attempt to abstract data away from real people facilitates the consumption and commodification of data, without adequately addressing the ethical and epistemological issues that this raises. We conclude our paper by arguing that benefit sharing approaches, which maintain and make explicit the links between data originators and the data they produce, are a better way of approaching these challenges.

• Friedman, Yael (2024). Correction to: Conceptual scaffolding for the philosophy of medicine (Medicine, Health Care and Philosophy, (2024), 10.1007/s11019-024-10231-w). Medicine, Health care and Philosophy. ISSN 1386-7423. doi: 10.1007/s11019-024-10242-7. Full text in Research Archive

View all works in NVA