CHAPTER 7
Ethical, Legal, and Social Implications of AI in Life Sciences
114
7.1 DATA PRIVACY AND SECURITY IN AI SYSTEMS
Artificial Intelligence (AI) has become increasingly important in life sciences, supporting applications such as medical diagnosis, drug discovery, genomic analysis, personalized medicine, clinical research, disease surveillance, and biomedical data analysis. These applications depend heavily on large and diverse datasets, including electronic health records, medical images, genomic sequences, laboratory results, wearable-device data, clinical-trial information, and patient-reported outcomes. Although access to such data can accelerate scientific discovery and improve healthcare delivery, it also creates significant challenges related to privacy, confidentiality, cybersecurity, and responsible data governance. The World Health Organization (WHO) emphasizes that AI in health should place ethics, human rights, accountability, and protection of individuals at the center of system design and deployment (WHO, 2021). (Figure 7.1)
Figure 7.1: Data Privacy and Security in AI Systems
7.1.1 Nature of Privacy Risks in Life-Science AI
Health and biological data are particularly sensitive because they can reveal information about an individual's physical condition, genetic characteristics, 115
reproductive history, lifestyle, disease risks, and family relationships. Unlike conventional identifiers such as names or addresses, certain biological characteristics cannot simply be changed after disclosure. Genetic information, for example, may provide information not only about an individual but also about biologically related family members.
AI introduces additional privacy risks because machine-learning systems can identify patterns that were not necessarily obvious when the data were originally collected. Even when datasets are de-identified, combinations of apparently harmless attributes may sometimes allow individuals to be re- identified. NIST notes that AI systems can create new privacy risks by enabling inference about individuals or previously private information. It therefore recommends privacy-enhancing approaches such as data minimization, de-identification, aggregation, and other privacy-enhancing technologies (NIST, 2023).
Another concern is the secondary use of data. Information initially collected for patient treatment may subsequently be used for research, AI model training, pharmaceutical development, or commercial applications. Such secondary use raises questions about whether individuals understood how their information would be used and whether their consent adequately covers future applications. Effective governance therefore requires clear rules concerning the purpose of data collection, permitted uses, retention periods, access rights, and data-sharing arrangements.
7.1.2 Data Security Threats
Privacy protection and cybersecurity are closely connected but are not identical. Privacy concerns primarily address appropriate collection, use, disclosure, and control of personal information, whereas security focuses on protecting information and systems against unauthorized access, alteration, destruction, or disruption.
AI systems used in life sciences may be exposed to conventional cybersecurity threats such as unauthorized access, phishing, ransomware, insider threats, insecure databases, and compromised cloud infrastructure. They may also face AI-specific attacks. For example, attackers may attempt to manipulate training
116
datasets, extract sensitive information from models, infer characteristics about individuals, or exploit vulnerabilities in AI-enabled applications.
The consequences of a security breach can be particularly serious in healthcare. Unauthorized access to medical records can result in financial harm, discrimination, reputational damage, or psychological distress. Manipulation of clinical or laboratory data could also affect medical decisions and research conclusions. In the United States, the HIPAA Security Rule requires covered entities and business associates to implement administrative, physical, and technical safeguards to protect the confidentiality, integrity, and availability of electronic protected health information (HHS, 2026).
7.1.3 Privacy-Preserving Approaches
Several technical and organizational measures can reduce privacy and security risks in AI-enabled life-science systems. Data minimization involves collecting and processing only the information necessary for a specific purpose. De-identification and pseudonymization can reduce direct identification risks, although they should not automatically be considered complete protection against re-identification.
Encryption should be applied to sensitive data both during transmission and while stored. Strong identity and access-management mechanisms should ensure that researchers, clinicians, developers, and administrators can access only the information required for their responsibilities. Multi-factor authentication, role-based access control, audit logs, vulnerability assessments, and continuous monitoring can further strengthen security.
Privacy-enhancing technologies can also support AI research. Techniques such as federated learning allow models to be trained across distributed datasets without necessarily transferring all raw data to a central repository. Differential privacy introduces controlled statistical noise to reduce the possibility of identifying individuals from analytical outputs. Secure multi- party computation and other privacy-preserving approaches can similarly enable collaboration while reducing direct exposure of sensitive datasets. However, these methods may involve trade-offs between privacy, computational cost, data utility, and model accuracy. NIST specifically recognizes that privacy-enhancing techniques can sometimes affect accuracy
117
and therefore require careful consideration within broader AI risk management.
7.1.4 Governance, Consent, and Accountability
Technical safeguards alone cannot guarantee responsible data use. Organizations need comprehensive data-governance policies covering data ownership, consent, access, retention, sharing, breach response, and accountability. The OECD recommends governance mechanisms that combine privacy-protective data use with transparent communication, appropriate authorization processes, de-identification, security practices, and continuous reassessment as technologies and risks evolve (OECD, 2015).
The NIST AI Risk Management Framework provides a broader risk-management approach that encourages organizations to consider trustworthy AI throughout the system lifecycle, including design, development, deployment, use, testing, and evaluation. Its trustworthiness characteristics include security, resilience, accountability, transparency, privacy enhancement, and fairness (NIST, 2023).
In life sciences, informed consent is particularly important. Individuals should, where applicable, be informed about what data are being collected, why they are being collected, who may access them, and whether they may be used for future research or commercial purposes. Consent procedures should also recognize that AI systems may create new uses for data that were difficult to anticipate when the information was originally collected. Consequently, governance frameworks should provide mechanisms for transparency, withdrawal where legally and technically feasible, and responsible secondary use.
Data privacy and security are fundamental requirements for trustworthy AI in life sciences. The enormous value of health, genomic, and biomedical datasets must be balanced against the rights and interests of individuals whose information makes AI research possible. Effective protection requires a combination of privacy-by-design principles, data minimization, encryption, access controls, de-identification, privacy-enhancing technologies, cybersecurity monitoring, informed consent, institutional governance, and continuous risk assessment. The WHO stresses that AI in health should be
118
developed around ethical and human-rights principles, while frameworks such as the NIST AI RMF provide practical approaches for managing emerging risks. Ultimately, protecting data privacy is not merely a technical or legal obligation; it is essential for maintaining public trust and ensuring that AI-driven advances in life sciences produce meaningful benefits without compromising individual rights.
7.2 ETHICAL ISSUES IN AI-BASED HEALTHCARE AND RESEARCH
Artificial intelligence (AI) is increasingly being integrated into healthcare and life-science research, including medical diagnosis, clinical decision-making, drug discovery, medical imaging, genomics, patient monitoring, and public-health surveillance. AI can process large and complex datasets at a speed that is difficult for humans to achieve and can potentially improve diagnostic accuracy, treatment planning, and research productivity. However, because healthcare involves human life, dignity, privacy, and fundamental rights, the use of AI creates significant ethical challenges. The World Health Organization (WHO) emphasizes that ethics and human rights should remain central to the design, development, deployment, and use of AI in health (WHO, 2021).
Figure 7.2: Ethical Issues in AI-Based Healthcare and Research
119
7.2.1 Privacy and Protection of Health Data
Healthcare AI depends heavily on large datasets containing sensitive information such as medical histories, diagnostic reports, genetic profiles, images, laboratory results, and behavioral information. The collection, storage, sharing, and secondary use of such data can create serious privacy concerns. Patients may not always understand how their information will be processed by AI systems or whether their data will be shared with researchers, technology companies, hospitals, or other institutions. In research, large datasets may also be combined from different sources, increasing the possibility of re-identification even when obvious identifiers have been removed. Therefore, strong data-governance mechanisms, appropriate consent procedures, access controls, encryption, and responsible data-sharing practices are essential. Privacy and security are recognized as major limitations and ethical concerns associated with healthcare AI (Topol, 2019).
7.2.2 Algorithmic Bias and Health Inequalities
One of the most important ethical concerns is algorithmic bias. AI systems learn patterns from historical data, and if the training data reflect existing social, economic, gender, racial, geographical, or healthcare inequalities, the resulting algorithm may reproduce or even amplify those inequalities. Bias can arise from under-representation of particular populations, poor-quality data, inappropriate selection of variables, or the use of biased outcomes as prediction targets.
A well-known example was reported by Obermeyer et al. (2019), who demonstrated racial bias in a healthcare algorithm used to identify patients requiring additional care. The algorithm used healthcare expenditure as a proxy for health needs, but because healthcare spending differed systematically between racial groups, the algorithm underestimated the healthcare needs of Black patients. This example demonstrates that removing race as an explicit variable does not necessarily eliminate discrimination. Ethical AI therefore requires systematic assessment of datasets, model performance, and outcomes across different demographic groups.
7.2.3 Transparency and Explainability
Many advanced AI models, particularly deep-learning systems, operate as complex “black boxes.” They may produce highly accurate predictions
120
without providing an understandable explanation of how a particular conclusion was reached. This creates a significant problem in healthcare because clinicians and patients may need to understand why a particular diagnosis, risk score, or treatment recommendation was generated.
Lack of explainability can also make it difficult to identify errors, challenge inappropriate decisions, or determine responsibility when harm occurs. UNESCO emphasizes that transparency and explainability are important conditions for protecting human rights and ensuring effective accountability in AI systems (UNESCO, 2021). Consequently, healthcare AI should provide explanations appropriate to the clinical context while avoiding unnecessary technical complexity.
7.2.4 Human Oversight and Clinical Responsibility
AI should support healthcare professionals rather than automatically replace human judgment in situations involving significant clinical consequences. An AI system may identify patterns that a clinician overlooks, but it can also generate incorrect predictions because of incomplete data, unusual cases, or differences between the population used for training and the population receiving care. Excessive dependence on AI may lead to automation bias, in which healthcare professionals accept algorithmic recommendations without sufficient independent evaluation.
The ethical principle of human oversight therefore requires clinicians to retain meaningful responsibility for decisions affecting patients. AI recommendations should be considered alongside clinical evidence, professional judgment, and patient preferences. The WHO similarly stresses accountability and the importance of ensuring that AI remains responsive to the healthcare workers and communities affected by its use (WHO, 2021).
7.2.5 Informed Consent and Patient Autonomy
AI also raises difficult questions about informed consent. Traditional consent generally involves explaining the purpose, benefits, risks, and alternatives associated with a medical intervention or research study. With AI, however, patients may not fully understand how their data will be used or how an algorithm will influence clinical decisions. In research, individuals may
121
consent to the collection of biological or medical data without anticipating that the information could later be used to train AI models for different purposes.
Respect for autonomy requires that patients and research participants receive meaningful information about significant uses of AI and their data. Where appropriate, consent processes should explain data reuse, sharing, potential risks, and the role of AI in decision-making.
7.2.6 Safety, Reliability, and Accountability
AI systems used in healthcare must be sufficiently accurate, reliable, and clinically validated. A model that performs well in a laboratory or controlled research environment may perform poorly in another hospital, geographical region, or demographic population. Changes in clinical practice, patient characteristics, equipment, and data quality can also reduce model performance over time.
Ethical deployment therefore requires continuous monitoring rather than relying solely on pre-deployment testing. Responsibility must also be clearly assigned among developers, healthcare institutions, clinicians, and other stakeholders.
Ethical discussions of healthcare AI emphasize not only technical performance but also questions of traceability, responsibility, and the consequences of system failures (Morley et al., 2020).
7.2.7 Ethical Issues in AI-Based Life-Science Research
AI-driven research introduces additional concerns related to research integrity, authorship, data ownership, reproducibility, and responsible innovation. Researchers may use AI to generate hypotheses, analyze datasets, identify drug candidates, or produce scientific text. However, AI-generated outputs can contain inaccurate information, hidden biases, or fabricated references. Researchers therefore remain responsible for verifying AI-generated results and ensuring that scientific conclusions are supported by reliable evidence.
Another concern is unequal access to AI technologies. Well-funded institutions may possess greater computational resources, high-quality datasets, and specialized expertise than institutions in low- and middle-income settings. Without deliberate efforts to promote inclusion, AI could widen existing global inequalities in healthcare and scientific research. WHO 122
identifies equitable access, inclusiveness, and public benefit as central considerations for responsible AI in health (WHO, 2021).
Ethical AI in healthcare and life-science research requires more than technical accuracy. Privacy, fairness, transparency, informed consent, human oversight, safety, accountability, and equitable access must be incorporated throughout the AI lifecycle. AI should be designed to complement human expertise while protecting patient autonomy and dignity. As AI becomes increasingly embedded in healthcare and research, multidisciplinary governance involving clinicians, researchers, ethicists, patients, policymakers, and technology developers will be essential. Ultimately, the ethical objective should not simply be to create more powerful AI systems, but to ensure that these systems contribute to safer, fairer, more transparent, and human-centered healthcare and scientific progress.
7.3 REGULATORY POLICIES AND LEGAL FRAMEWORKS
The rapid integration of artificial intelligence (AI) into life sciences has created significant opportunities in medical diagnosis, drug discovery, biomedical research, genomics, personalized medicine, clinical decision-making, and public health. At the same time, the use of AI in these sensitive domains raises complex legal and regulatory questions concerning patient safety, privacy, informed consent, accountability, discrimination, intellectual property, and the reliability of automated decisions. Consequently, regulatory policies are increasingly shifting from technology-neutral principles toward risk-based frameworks that consider the potential impact of AI on individuals and society. The World Health Organization (WHO) emphasizes that AI applications in health should place ethics, human rights, accountability, and public benefit at the centre of their design and deployment (WHO, 2021). (Figure 7.3)
123
Figure 7.3: Regulatory Policies and Legal Frameworks
7.3.1 Global Regulatory Approaches
There is currently no single international legal framework governing AI across all areas of life sciences. Instead, regulation is developing through a combination of AI-specific legislation, medical-device regulation, data-protection laws, research-ethics requirements, and professional standards. This fragmented regulatory environment reflects differences in national legal systems and healthcare structures.
The European Union (EU) has adopted one of the most comprehensive approaches through the EU Artificial Intelligence Act (Regulation [EU] 2024/1689). The legislation follows a risk-based model in which AI systems are subject to different requirements depending on their potential level of harm. AI applications associated with healthcare and medical devices may fall within the high-risk category, requiring stronger controls relating to risk management, data governance, technical documentation, transparency, human oversight, accuracy, robustness, and cybersecurity. This approach is particularly relevant to life sciences because AI systems can directly influence diagnosis, treatment, and other high-impact decisions.
124
The United States generally regulates AI-enabled medical technologies through existing sector-specific regulatory structures rather than through a single comprehensive AI law. The U.S. Food and Drug Administration (FDA) evaluates AI-enabled medical devices under its medical-device regulatory framework, with attention to safety and effectiveness and the intended use and technological characteristics of the device. The FDA maintains an AI-enabled medical-device list to provide greater transparency regarding authorized products.
This framework is especially important for machine-learning systems used in medical imaging, diagnosis, monitoring, risk assessment, and treatment support.
The WHO provides an important international governance framework even though its recommendations are not directly equivalent to national legislation. Its guidance calls for AI in healthcare to respect human autonomy, promote human well-being and safety, ensure transparency and explainability, establish responsibility and accountability, promote inclusiveness and equity, and ensure that AI technologies are sustainable and responsive to public needs (WHO, 2021). Similarly, UNESCO's Recommendation on the Ethics of Artificial Intelligence, adopted in 2021, establishes a broader global framework emphasizing human rights, fairness, transparency, accountability, privacy, and environmental and social considerations (UNESCO, 2021).
7.3.2 Data Protection and Privacy
Data protection represents one of the most significant legal issues associated with AI in life sciences. AI systems frequently require large datasets containing electronic health records, genomic information, medical images, laboratory results, and other sensitive personal information. Improper collection, processing, sharing, or re-identification of such data can cause substantial harm to individuals.
In India, the Digital Personal Data Protection Act, 2023 (DPDP Act) establishes a legal framework governing the processing of digital personal data. It defines obligations for data fiduciaries and provides rights and duties concerning personal data. The framework is particularly relevant to AI-based healthcare systems because developers and healthcare institutions may process
125
large quantities of personal information when training or operating AI models. Effective governance therefore requires appropriate notice, lawful processing, security safeguards, access controls, and mechanisms for addressing individual rights.
7.3.3 Indian Regulatory and Ethical Framework
India is developing a combination of statutory, regulatory, and ethical mechanisms for governing AI in biomedical research and healthcare. The Indian Council of Medical Research (ICMR) published its Ethical Guidelines for Application of Artificial Intelligence in Biomedical Research and Healthcare in 2023. These guidelines provide a framework for ethical decision-making during the development, deployment, and adoption of medical AI. They address ethical principles, stakeholder responsibilities, ethics review, governance, and informed consent. Importantly, the guidelines apply to developers, researchers, clinicians, ethics committees, institutions, sponsors, and funding organizations, demonstrating that responsible AI governance requires shared accountability rather than responsibility being placed solely on technology developers (ICMR, 2023).
7.3.4 Accountability, Liability, and Human Oversight
A major unresolved legal question concerns responsibility when an AI system produces an incorrect or harmful recommendation. For example, if an AI diagnostic system fails to identify a serious disease, responsibility could potentially involve the software developer, healthcare institution, clinician, data provider, or manufacturer. Traditional liability frameworks may not always adequately address such situations because AI systems can be complex, continuously updated, and dependent on data from multiple sources.
Regulatory frameworks therefore increasingly emphasize human oversight. AI should generally support rather than completely replace professional judgment in high-stakes medical decisions. Healthcare professionals must be able to understand the intended function and limitations of an AI system and intervene when its recommendations appear unreliable. Developers and institutions should also maintain documentation, validation records, audit trails, and mechanisms for reporting adverse events.
126
7.3.5 Regulatory Challenges and Future Directions
AI regulation in life sciences remains an evolving process. One important challenge is the adaptive nature of machine-learning systems. Conventional medical-device regulation often evaluates a product at a particular point in time, whereas AI systems may change when algorithms are retrained or updated with new data. Regulators must therefore develop mechanisms for continuous monitoring, post-market surveillance, algorithmic auditing, and controlled modification.
Future regulatory policies are likely to emphasize risk-based governance, transparency, explainability, data quality, cybersecurity, continuous monitoring, human oversight, and international harmonization. Regulatory authorities will also need to balance safety with innovation so that excessive compliance requirements do not prevent beneficial AI applications from reaching patients and researchers.
Overall, effective regulation of AI in life sciences should not be viewed simply as a mechanism for restricting technological development. Rather, it provides the legal and institutional foundation necessary for trustworthy innovation. A strong regulatory ecosystem can protect patients and research participants while simultaneously encouraging responsible development of AI technologies. The long-term objective should therefore be a governance model in which innovation, safety, privacy, human rights, scientific integrity, and public benefit are treated as interconnected requirements.
7.4 SOCIAL IMPACT AND HUMAN-AI COLLABORATION
Artificial intelligence (AI) is increasingly influencing life sciences by transforming healthcare delivery, biomedical research, drug discovery, diagnostics, genomics, public health, and patient management. Its social impact, however, extends beyond technical performance. AI systems can influence how healthcare resources are distributed, how professionals make decisions, how patients interact with healthcare institutions, and how societies understand health and disease. Consequently, the integration of AI into life sciences should be approached as a human-AI collaboration rather than as a simple replacement of human expertise. The World Health Organization (WHO) emphasizes that AI in health should be developed and deployed with
127
ethics, human rights, accountability, and public benefit at its core (World Health Organization [WHO], 2021). (Figure 7.4)
Figure 7.4: Social Impact and Human-AI Collaboration
7.4.1 Social Impact of AI in Life Sciences
One of the most significant positive social impacts of AI is its potential to improve access to healthcare and scientific knowledge. AI-based diagnostic systems can analyze medical images, laboratory results, genomic information, and patient records rapidly, potentially supporting earlier detection of diseases and improving clinical decision-making. In regions where specialist healthcare professionals are limited, AI-assisted tools may provide additional decision support to frontline healthcare workers. Similarly, AI can accelerate drug discovery by identifying molecular relationships, predicting drug–target interactions, and assisting researchers in screening large numbers of potential compounds. These capabilities may reduce research time and support the development of more personalized approaches to treatment (WHO, 2021).
AI can also affect the social organization of healthcare work. Routine administrative activities, documentation, image analysis, and data processing can increasingly be supported by automated systems. This may allow healthcare professionals to devote more time to communication, complex 128
decision-making, and patient-centred care. The broader objective should therefore not be to remove human involvement but to redistribute human effort toward tasks requiring empathy, contextual understanding, ethical judgment, and interpersonal communication. Topol (2019) similarly argues that appropriately implemented AI can potentially strengthen rather than weaken the human dimensions of medicine.
However, these benefits are not automatically distributed equally. Access to advanced AI systems may be concentrated in wealthy healthcare institutions, countries, or populations with better digital infrastructure. This creates a potential digital health divide between technologically advanced and resource-constrained communities. If AI becomes an important component of diagnosis or treatment while certain populations remain poorly represented in training datasets, existing inequalities may be reproduced or intensified. UNESCO therefore emphasizes inclusion, fairness, human dignity, and the avoidance of discrimination as important principles for AI development and deployment (UNESCO, 2021).
Algorithmic bias represents another important social concern. AI systems learn from historical data, and those data may reflect existing inequalities in healthcare systems. A widely discussed example is the study by Obermeyer et al. (2019), which demonstrated racial bias in a healthcare algorithm used for managing patient populations. The algorithm relied on healthcare costs as a proxy for healthcare needs, a choice that reflected underlying inequalities in healthcare expenditure and consequently underestimated the needs of some Black patients.
This example illustrates that an algorithm does not need to explicitly use race or another protected characteristic to produce socially unequal outcomes.
7.4.2 Human-AI Collaboration
Human-AI collaboration provides an alternative to the idea that AI will simply replace scientists, physicians, researchers, and other professionals. AI systems are particularly effective at processing enormous datasets, identifying statistical patterns, generating predictions, and performing repetitive computational tasks. Humans, in contrast, contribute contextual knowledge, ethical reasoning, emotional intelligence, professional responsibility, and the
129
ability to interpret complex social circumstances. Combining these capabilities can produce better outcomes than relying exclusively on either humans or machines.
In clinical practice, for example, an AI system may identify a potentially abnormal radiological feature, estimate a patient's risk of developing a disease, or recommend possible treatment options. The clinician can then evaluate that information against the patient's symptoms, medical history, preferences, social circumstances, and other factors that may not be adequately represented in the algorithm. This creates a collaborative decision-making process in which AI provides analytical support while the human professional retains responsibility for interpretation and action. Char et al. (2018) emphasize that implementing machine learning in healthcare requires careful attention to ethical challenges rather than assuming that technical performance alone guarantees beneficial outcomes.
Effective human-AI collaboration requires meaningful human oversight. Professionals should understand the intended purpose, limitations, uncertainty, and potential failure modes of an AI system before incorporating its recommendations into practice. AI outputs should not automatically be treated as objective or infallible. In high-risk applications, mechanisms should exist for human review, intervention, correction, and, where necessary, rejection of an AI recommendation. The NIST AI Risk Management Framework similarly promotes approaches for identifying, assessing, and managing AI risks throughout the system lifecycle (Tabassi, 2023).
Another essential element is transparency and communication. Patients and research participants should, where appropriate, know when AI contributes substantially to decisions affecting them. Healthcare professionals should also receive adequate training to understand AI-generated recommendations and recognize situations in which an algorithm may be unreliable. Such practices are important for maintaining trust between patients, professionals, institutions, and technology developers.
Human-AI collaboration also requires multidisciplinary participation. AI developers cannot independently determine what constitutes an acceptable healthcare outcome. Physicians, nurses, biomedical researchers, ethicists,
130
patients, legal experts, policymakers, and affected communities should participate in the design, evaluation, and governance of AI systems. Such participation can help identify social risks that may remain invisible from a purely technical perspective.
7.4.3 Building a Human-Centred AI Future
The future of AI in life sciences should therefore focus on augmentation rather than substitution. AI should enhance human capabilities while preserving human autonomy, dignity, accountability, and professional judgment. Organizations should evaluate AI systems not only according to accuracy or efficiency but also according to fairness, accessibility, safety, explainability, privacy, and their effects on different social groups.
A human-centred approach also requires continuous monitoring after deployment. AI systems may perform differently when used with new populations, hospitals, laboratories, or datasets. Consequently, evaluation should continue throughout the system's operational life rather than ending when an algorithm achieves satisfactory performance during development. WHO (2021) and UNESCO (2021) both stress the importance of accountability, equity, human rights, and responsible governance in ensuring that AI produces social benefit.
Ultimately, the social impact of AI in life sciences will depend less on the technology alone and more on how societies choose to design, govern, and use it. Responsible human-AI collaboration can improve scientific productivity, healthcare quality, and access to services while preserving the uniquely human dimensions of care and judgment. The objective should not be to create systems in which humans become passive recipients of machine decisions, but systems in which AI provides powerful analytical capabilities and humans remain informed, responsible, and empowered decision-makers.
131