Programme

This paper will examine the current and potential impact of various applications of artificial intelligence (AI) to the field of academic publishing in psychiatry. At present, AI algorithms assist in data analysis, allowing researchers to process large datasets quickly and uncover complex patterns that would be challenging to detect manually. In psychiatry, this capability can potentially help integrate data from genetics, neuroimaging, and clinical assessments. AI-driven natural language processing (NLP) tools might also facilitate systematic reviews and meta-analyses by automating the extraction and synthesis of information from vast bodies of published literature.

In publishing, AI can help to streamline the peer review process in certain ways. Automated systems can screen manuscripts for methodological rigor, ethical compliance, and potential conflicts of interest, thereby reducing the burden on human reviewers by prompting them to consider certain matters, and possibly accelerating the publication timeline. So far, however, AI shows no sign of replacing peer-review entirely.

AI-powered tools are already used to help with dissemination of research findings by generating summaries and identifying key insights, making information more accessible to a broader audience.

In the future, AI has the potential to enhance psychiatric publishing in various other ways. Predictive analytics might identify emerging trends and research gaps in the literature, guiding future studies and funding priorities. AI could also facilitate more robust collaborations by connecting researchers with complementary expertise and interests. Additionally, the integration of AI in digital platforms could democratise access to cutting-edge research, promote global knowledge sharing, and accelerate advancements in clinical care.

As AI continues to evolve, its applications in research and publishing hold the potential to drive significant progress in understanding and treating mental disorders. It is essential that these developments are accompanied by openness about the use of AI in publishing, with clear declarations by authors and publishers about the use of specific applications in their work.

Artificial intelligence (AI) is revolutionising the healthcare landscape and geriatric medicine. This study aimed to provide a comprehensive overview of AI applications in geriatric care, focusing on common age-related conditions. Furthermore, this study provides examples from real-world clinical practice to test the ability of AI to diagnose neurodegenerative disorders compared to human clinicians.

To evaluate AI's ability of AI to diagnose neurodegenerative disorders, Claude 3.5 Sonnet, a large language model, was tested using clinical cases from the Center for Cognitive Decline and Dementia. AI diagnoses were compared to those formulated by clinicians, revealing that AI performs better when clear signs and symptoms are present but struggles to differentiate between subtly different clinical presentations. Despite these advancements, AI faces challenges in diagnosing conditions with subtle symptoms and in navigating the complexities of geriatric cases. Further research is needed to develop and evaluate AI-based geriatric support systems, ensuring they effectively integrate with clinical practice and improve patient outcomes

Today, machine learning models are increasingly incorporated in expert decision-making in a number of fields, for instance to improve medical diagnosis or legal reasoning. The high accuracy and speed of such models make them very useful assistants to human decision-making; moreover, algorithms may help to overcome cognitive biases and inconsistency in the judgment of clinicians, judges, and other experts. At the same time, machine learning systems also raise multiple legal, ethical, and social concerns for experts, their decisions' stakeholders, and society at large. For this reason, interdisciplinary research teams are working, within the so-called human-centered paradigm, on regulatory frameworks and design guidelines to ensure an efficient but also equitable application of machine learning.

One crucial issue with applying machine learning models is to understand how the interaction with algorithms affects human reasoning and decision-making tasks when high uncertainty and/or high stakes are involved. Ever since the first automation support systems were introduced in industrial operations, researchers have pointed out that replacing or augmenting human capabilities is not straightforward and typically leads to unpredictable consequences and possibly mistakes. One particularly troubling problem in this connection is over-reliance: human specialists tend to excessively rely on suggestions from the systems, sometimes leading to surprising and unjustifiable errors in decisions. While human-computer interaction (HCI) engineers are working hard on practical strategies to counter over-reliance, the phenomenon itself lacks a precise definition and the cognitive mechanisms behind over-reliance are not yet fully explored.

In this paper, we address the issue of over-reliance from a cognitive perspective, aiming to better understand this phenomenon, its cognitive determinants, and the possible strategies to overcome it. We proceed as follows. First, we review the multidisciplinary discussion of over-reliance, which is currently spread among different disciplines. In particular, we contrast the notion of complacency, as discussed by human factors engineers, and that of automation bias studied by social psychologists, reviewing the most recent efforts to overcome over reliance by replacing AI support with cognitive forcing. We also discuss recent real-world examples of erroneous decisions due to the over-reliance on automation and analyze a few instances of primary evidence to illustrate the cognitive mechanisms behind the phenomenon. Second, we propose an analysis of over-reliance as a predictable and systematic deviation from sound Bayesian reasoning, here assumed as the relevant normative model of rational thinking and decision-making. More specifically, we suggest that over-reliance is a kind of “base rate fallacy”, a well-studied phenomenon in the cognitive science literature. Third, we discuss the implications of our model for the current discussion of AI-assisted reasoning and decision-making. In particular, we compare our proposal with competing ones in the literature and we show how over-reliance relates to other issues emerging from the debate on transparent and trustworthy AI. Interestingly, our analysis implies both that over-reliance may persist in the presence of common countermeasures based on the accuracy and transparency of the machine learning models, and that higher accuracy may actually increase over-reliance, rather than decreasing it. We conclude by discussing such problematic findings with reference to selected cases in HCI studies, suggesting how training, evaluation, and communication may improve the deployment of machine learning models in reasoning and deciding under uncertainty.

Forensic psychiatric patients are subjected to significant levels of stigma, largely due to the widespread misconception that associates mental illness and criminal history with social dangerousness. This stigma has profound effects on their treatment, rehabilitation, and reintegration into society. With the advent of Generative AI and the increasing presence of AI-generated content across the media landscape and beyond, this stigma is being shaped and redefined in complex and evolving ways. The introduction of AI-driven content creation and analysis is not only influencing public perceptions but also reshaping the narratives surrounding forensic psychiatric patients in intricate and multifaceted ways. These technological developments have the capacity to either reinforce existing stigmatising attitudes or to challenge and deconstruct them, depending on how they are applied. The complexity of this dynamic underscores the need for a careful and nuanced approach to the integration of AI in contexts that impact vulnerable populations, such as those within the forensic psychiatric domain.

This presentation explores the influence of Generative AI on representations of forensic mental health. It examines how these models could transform the visual portrayal of forensic psychiatric phenomena, addressing both the potential and the limitations of AI-generated content in accurately reflecting complex mental health scenarios. By examining recent advancements and practical examples, this presentation aims to provide a nuanced perspective on how AI-driven content creation could impact the stigma surrounding forensic psychiatry.

The literature on forensic psychiatric evaluations reveals ongoing debate regarding the differences between expert and non-expert judgments. While some studies suggest that professionals, such as psychiatrists, may be prone to similar biases and heuristics as non-experts, other research highlights significant differences in decision-making processes. Oskamp (1965) and Jackson (1986) found little difference in the accuracy of clinical judgments between experts and laypeople. However, Grøndahl et al. (2009) observed notable differences, with professionals showing more nuanced evaluations based on case-specific information, challenging the view that experts are no more reliable than non-experts. Göranson et al. (2022) further explored how different professional groups utilize specific information in diagnosing severe mental disorders, finding slight variations in the types of information prioritized and the tendency to diagnose severe conditions. Svensson et al. (2022) examined the decision-making processes of Swedish forensic psychiatric experts, revealing a complex and iterative approach influenced by the availability of information and time constraints, which may impact decision quality. This body of research underscores the need to further investigate the factors influencing expert judgments in forensic settings. The aim of this integrative review is to explore the decision-making processes in Forensic Psychiatric Investigations and to identify strategies for improving the quality of these processes. The review seeks to understand the types of information relied upon by experts in Forensic Psychiatric Investigations how this information and other available resources are utilized, and to delineate the decision-making processes occurring within these investigations.

The perceived fairness of an algorithm is a complex and multidimensional concept that relates to whether the algorithm treats all individuals or groups fairly when processing data and making decisions. It not only affects the trust of individuals in algorithms, but also the overall acceptance and reliance of society on the technology. With the increasing use of big data and machine learning techniques in predictive algorithms, crime prediction software has become an important part of modern policing. However, the presence of data bias may lead to unfairness in algorithmic results, which in turn affects policing decisions, threatens the perceived fairness of algorithmic decisions, and widens social inequality. Three main types of bias types exist in current crime prediction software: design bias, training bias, and interaction bias. These biases permeate and are amplified in algorithms through predictive models and manifest themselves in practice as feedback loops. Reducing the negative impact of data bias by enhancing algorithmic transparency and interpretability, improving the diversity and fairness of datasets, and optimizing feedback mechanisms can be beneficial in informing the assessment of future algorithmic fairness-aware practices.

The emergence of Cognitive Warfare (CW) as a new frontier in contemporary conflict has expanded far beyond the military domain, now permeating everyday life. Both State and Non-State actors increasingly exploit cognitive warfare tactics, using Neuroscience and Artificial Intelligence (AI) to influence perceptions, manipulate decisions, and control behavior on a mass scale. Neuroscience provides insight into the brain's decision-making and emotional processes, enabling sophisticated manipulation strategies. Meanwhile, AI leverages these insights through microtargeting, algorithmic manipulation, ad-hoc Disinformation campaigns, radicalising people, and posing a continuous and serious threat to public, national and global security with the aim to undermine democracies. This paper explores also how AI can be applied to mitigate the threats of CW and provide enhanced protection for citizens.

Forensic psychiatric evaluations are mandatory for the insanity assessment but face significant challenges related to their accuracy, accountability, and transparency. The lack of standardised methods makes it extremely difficult for experts to agree on insanity evaluations. A comprehensive understanding of the assessment processes and the biases influencing forensic psychologists and psychiatrists is essential for improving their reliability and standardising the methodology.

This systematic review aims to identify the factors that influence assessments of insanity. The investigation focuses on the pieces of information (e.g., personal information and psychiatric information about the defendant, information about the defendant's criminal record and the committed offence) which tend to guide the decision-making process of psychologists and psychiatrists appointed by the Court to assess the defendant's insanity.

This systematic review adheres to the PRISMA guidelines. The search has been conducted on the following electronic databases: PubMed, CINAHL, PsycINFO, Web of Science and Scopus. It includes studies searched through reference lists of relevant review papers and references of included/excluded studies with no limitations on the year of publication. Two reviewers independently screened titles and abstracts for eligibility, followed by a detailed review of full papers. Discrepancies have been resolved through discussion or consultation with a third reviewer.

This systematic review seeks to enhance comprehension regarding the limitations and strengths inherent in the decision-making processes of expert psychologists and psychiatrists in charge of evaluating the defendant's insanity, a fundamental initial step towards standardising forensic evaluations.

Despite the central role in criminal trials, there is limited research on the decision-making processes of experts in forensic psychiatry. The present study aims to explore the impact of sociodemographic, psychopathological, and criminological factors on forensic psychiatric determinations concerning criminal responsibility and social dangerousness in stalker perpetrators evaluations. We analyzed 108 forensic psychiatric reports provided by 15 Italian forensic psychiatrists. Defendants' psychiatric symptom severity was evaluated through the Brief Psychiatric Rating Scale (BPRS), the Clinical Global Impression (CGI) scale, and the Global Assessment of Functioning (GAF). The sample consisted of 80% men and 20% women. Nearly all subjects had a positive psychiatric history (95%), with 41% having experienced involuntary hospitalization in the past. On average, the stalking duration was 29 months, with a longer duration for subjects not taking pharmacotherapy and shorter for those treated with antidepressants (p<0.05). In 66% of cases with male stalkers, the victims were women and vice versa (p<0.001). No significant differences were found in age, education, marital status, alcohol or drug use, criminal record, or psychiatric history between responsible and not criminally responsible defendants. In particular, defendants judged not criminally responsible (Not-CRDs) presented with more severe psychiatric symptoms, were more likely male (p<0.05), 40% had a diagnosis of schizophrenia (p<0.001), and a higher rate than previous involuntary hospitalization (43%, p<0.05). In comparison, personality disorders were more associated with subjects considered to be criminally responsible (56%, p<0.001). All the female offenders were considered not criminally responsible, raising the possibility of gender bias in forensic judgments concerning defendants' mental responsibility. The 84% of male subjects were considered socially dangerous (p<0.05), and only 38% of the subjects considered dangerous had been involuntarily admitted in the past (p<0.05). Moreover, the social dangerousness evaluations highlighted how older experts (mean age: 61.2 years, 13.3 SD) expressed more non-social danger evaluations than lower mean age experts (mean age: 50.4 years, 10.0 SD) (p<0.05).

Risk assessment tools are increasingly utilized within the criminal justice system. These tools are standardized, scientifically validated models designed to estimate the likelihood of reoffending, particularly in relation to serious offenses. However, their use has sparked debate, especially regarding the accuracy and fairness of these predictions. Concerns have been raised that risk assessments may contribute to or exacerbate existing inequalities within the criminal justice system, particularly affecting members of ethnic minority groups. Furthermore, advancements in algorithmic risk predictions introduce new challenges, further intensifying the debate on fair risk assessment. This presentation will explore several risks associated with risk assessment, as well as the normative issues that arise in this context.

The ability to recognize the authenticity of emotions expressed by other people has a strong impact on social interactions, as faces are the most informative stimulus during an interaction. Difficulties in the recognition of authentic emotional states may lead to wrong interpretations of others’ intentions, which in turn may cause aggressive or violent behavior. According to recent data collected by our group, incarcerated men perform worse than controls on the Emotion Authenticity Recognition (EAR) test. Preliminary fMRI results show that the correct identification of a genuine (vs. simulated) emotion in a community sample is linked to increased activation in the right middle temporal gyrus and in the left lateral occipital cortex. The former area is involved in emotion processing and social cognition, through its connections to the default mode network and the theory of mind network (Xu et al., 2019). The latter area is involved in visual processing and cognitive processes such as attention and working memory (Nie et al., 2019; Salmon et al., 1996). This suggests that impaired performance in the EAR test might be linked to difficulties in social cognition and cognitive processes required for reasoning in emotional contexts. The literature shows the presence of alterations in both these abilities in the convicted population (Mariano et al., 2017; Shumlich et al., 2019), which are confirmed in the data collected by our group.

In addition to punishment for crimes, many contemporary criminal law systems incorporate a form of preventive detention. Preventive detention refers to the incarceration of dangerous offenders – often diagnosed with a mental illness – based solely on the need to prevent future harm. Its primary objective is the protection of the public from individuals deemed dangerous, rather than retribution. However, this form of detention, for instance in a forensic psychiatric facility, significantly infringes upon the fundamental right to freedom, as enshrined in the European Convention on Human Rights (ECHR) and many national constitutions. Consequently, the imposition of preventive detention must not only adhere to national criminal law but must also comply with national and international constitutional rights law. This requires a proportionality assessment, balancing the fundamental right to freedom of the individual against the collective interest in security and crime prevention.

This paper presents the results of a comparative legal study on proportionality assessments in legal decision-making about preventive detention orders. It compares case law on the German preventive detention measures of Sicherungsverwahrung (§66 German Criminal Code) and Unterbringung in einem psychiatrischen Krankenhaus (§63 German Criminal Code) with the Dutch preventive detention measure Terbeschikkingstelling met dwangverpleging (art. 37a Dutch Criminal Code). Additionally, the paper analyzes relevant case law from the European Court of Human Rights (ECtHR) concerning Article 5 of the ECHR. This comparative analysis reveals significant variations in how proportionality assessments are conducted across jurisdictions, highlighting fundamental differences in the relationship between criminal and constitutional law.

The outcome of Swedish forensic psychiatric investigations (FPI) can have a major impact on legal decisions concerning mentally disordered offenders. However, the decision-making process of these investigations is largely unstudied. Semi-structured interviews with Swedish experts (n=38) working with FPI:s indicate that the decision-making process is highly flexible, potentially allowing for unique adaptations in each case and high validity while at the same time possibly posing challenges with reliability. A follow-up case vignette study of just over a third of the Swedish experts working with FPI:s showed that they formulate and entertain multiple hypotheses throughout the decision-making process and were ready to revise these hypotheses when presented with new information. There was substantial variation among the experts regarding which hypotheses they deemed most relevant. While the experts grew more certain in their opinions on severe mental disorder, a key medicolegal construct in Swedish law, there was considerable variation in their opinion throughout the process and at the end of it. If used properly, the variation in both process and outcome could be used to safeguard and possibly increase the reliability and validity of the final decision of Swedish forensic psychiatric investigations.

The present study examined how judges incorporate scientific information in their decision-making. We utilized a mixed-methods approach, including Natural Language Processing techniques and qualitative analysis, to analyze data from semi-structured interviews with 34 U.S. criminal court judges. Sentiment analysis revealed differences in emotional tone between judges with varying degrees of belief in scientific determinism and based on gender, age, geographical region, and professional background. Structural Topic Modeling identified key considerations, including determinism, responsibility, treatment needs, and moral concerns. Qualitative analysis enriched these results by unraveling the philosophical and legal considerations that judges grapple with when considering scientific explanations for defendants’ behavior. This study’s findings underscore the nuanced interplay between scientific understandings of behavior, personal beliefs, and judicial decision-making. We discuss these insights in relation to decision-making and suggest ways to better integrate scientific knowledge into judicial practice.

The concept, scope, and academic framework of cognitive forensic science remain to be clarified, and the core theoretical system has yet to be established. The promotion of interdisciplinary integration in the development of this discipline and the provision of new pathways and theoretical support for legal education and theoretical research require further elaboration. By retrieving data from the Web of Science Core Collection, all English-language publications related to cognitive forensic science up to August 19, 2024, were collected. A refined search was conducted to identify papers with even narrower focus and higher relevance. A total of 2,741 papers on cognitive forensic science from 1975 to 2024 were obtained, along with 217 high relevance papers from 2004 to 2024. Using the Bibliometrix software, a visual analysis of authors, institutions, and keywords was performed, along with co-occurrence, clustering, and timeline analysis, to illustrate the distribution trends and research evolution patterns in the field. Overall, the volume of academic publications in the field has shown a gradual upward trend, with a substantive shift in research focus around the turn of the 21st century. The field has evolved from focusing on criminals to encompassing all participants in judicial processes and from concentrating on individuals with mental illness or psychological abnormalities to exploring general decision-making behaviors in the justice process. This shift reflects the maturation of the discipline. However, the research field still lacks a robust global collaborative network, and there is insufficient international cooperation among scholars.

The aim of forensic mental healthcare is to treat offenders with severe psychological problems, so they can safely reintegrate into society. To make reintegration sustainable, it is not only necessary to minimize the risk of new crimes as much as possible, but also important to contribute, where possible, to repairing the harm that the offenders have caused to victims and their families with their crimes. Restorative justice is gaining traction globally as it presents a promising approach to help restoration after crime. However, its application within forensic psychiatry remains underexplored. Core questions to be answered concern the motivations and the decision-making processes for engaging in victim-offender contact in forensic psychiatric settings. Within this presentation I will, first, explore the theoretical underpinnings of restorative justice—such as safety, responsibility, and voluntariness—within the context of forensic psychiatry. Next, based on the results, some challenges and opportunities of implementing restorative justice interventions in forensic psychiatry will be considered. Finally, I will highlight the potential for more holistic and reparative approaches to offender rehabilitation.

Victim-offender mediation is a form of restorative justice in which victim and offender aim to repair the harm that was caused by the offence. This contact between victims and offenders can also be beneficial within a forensic mental health context. In the Netherlands forensic social workers supervise the network of patients residing in forensic psychiatric hospitals and in that capacity they also interact with victims. This presentation focuses on working with victims during offender treatment from the perspective of these forensic systemic counselors. Based on interviews, group discussions and a survey, their work and their perceptions as well as procedures relevant for the involvement of victims during treatment of offenders will be discussed. It will become clear that forensic systemic counselors experience considerable tension between the interests of patients and victims, and that procedures and cooperation with stakeholders are important factors in their decision-making processes.

Proteomics, the science that studies proteins on a large scale, is transforming forensic investigation by enabling the precise analysis of biomarkers in biological samples. This emerging technology offers significant benefits, such as the identification of individuals and the detection of illicit substances. However, its application in the field of mental health raises important ethical questions relating to privacy and consent.

The ability to identify genetic predispositions to mental illnesses, such as schizophrenia, bipolar disorder and depression, through specific biomarkers, can help with diagnosis. However, the use of this information outside the medical context, especially by third parties such as employers, exposes individuals to risks of discrimination and stigmatisation. Disclosure of proteomic data without informed consent can result in violations of human rights, such as the right to privacy and non-discrimination.

For example, a genetic predisposition to mental illness can be interpreted as an indicator of future disability, leading companies to avoid hiring individuals with such characteristics in an attempt to minimise health insurance costs and absences from work. Such discriminatory scenarios have occurred in previous cases, such as the use of genetic tests by entities like the US Air Force and companies like DuPont to exclude workers with genes associated with sickle cell anaemia.

Given the invasive potential of proteomics, it is essential that its use is strictly regulated, with a focus on protecting privacy and prohibiting discrimination. The implementation of legal safeguards that guarantee explicit consent and transparency in the use of proteomic data is crucial to balancing scientific progress with respect for human rights. Only with these measures will it be possible to utilise the innovations of proteomics ethically and fairly.

This talk provides an analysis of recent uses of neuroscience in prison litigation, specifically in the context of prolonged solitary confinement. Using the U.S. as a case study, the talk reviews selected court cases from the past five years (2019-2024) where neuroscience has been introduced as complementary evidence of the psychological harms caused by isolation. The talk will notably focus on the types of normative claims that neuroscience can help substantiate from both constitutional and penological perspectives. Ultimately, the goal is to highlight the growing presence of neuroscience in prison litigation, highlighting a shift in its courtroom use—moving from supporting individual defense or mitigation strategies to addressing broader structural issues within the criminal justice system.

Contemporary criminal law upholds the position that humans, as individual beings, can be held accountable for their actions by default, unless a certain predefined condition is met that negates accountability. In different jurisdictions there are typically four such conditions formalized in an unaccountability doctrine: infancy, unconsciousness, cognitive impairment, and the so-called ‘insanity’. Conceptually each of these conditions, along with the baseline idea of humans being accountable are being explicitly or implicitly developed and justified based on the conjunction of a selection of ethical tenets, and a selection of tenets about the nature of the phenomenon of human action and by extension – about the human as an actor. Considering that the tenets about human action were traditionally developed based on predominantly unsystematic observations of humans acting in naturalistic settings, analytical inductive inferences of varying degrees of structuredness and transparency made from those observations, and conceptual products of introspection and reasoning, from the contemporary reliabilist physicalist perspective, there is a substantial possibility that the tenets may be partially or completely erroneous products of unreliable approaches to the analysis of unreliable data. In such a situation a tenet would become an unjustified belief.

In a similar vein, proponents of the physicalist worldview – natural scientists, in particular neuroscientists increasingly draw their attention to tenets that underlie the legal concepts of (un)accountability and derivatives, attempting to assess their validity. Similar efforts are directed towards assessment of assumptions at lower levels of abstraction. However, these efforts share two key problems that attract critique: 1) there is no established framework for systematic ascertainment of legally relevant tenets and assumptions; 2) The overall ‘neurolaw’ discussion appears fragmented, lacking an overarching unifying thematic framework, and instead proposes insights from different branches of neuroscience, unsystematically relating them to various tenets at higher levels of abstraction or assumptions at lower levels of abstraction. Taking a reliabilist physicalist stance, in my presentation I tackle these problems by presenting a justifiable scientific framework for validation and further development of tenets about the nature of the phenomenon of human action & about human as an actor that underlie legal concepts of (un)accountability. From within the framework I argue for the ‘biopsychology of the decision-making process’ as the key relevant physicalist praxeological framework for assessing the validity of tenets, and as a framework for their further development.

In criminal proceedings, offenders are sentenced based on doctrines of culpability and punishment that theorize why they are guilty and why they should be punished. Throughout human history, these doctrines have largely been grounded in legal-policy constructions around retribution, safety, deterrence, and closure, mostly derived from folk psychology, natural philosophy, sociocultural expectations, public-order narratives, and common sense. On these premises, justice systems have long been designed to account for crimes and their underlying intent, with experience and probabilistic assumptions shaping theoretical discourses on the nature of crimes and offenders’ punishability. As scientific discoveries, inventions, and methodologies progressively developed to refine such doctrines and displace long-held assumptions, criminal courtrooms have increasingly witnessed counsels and judges relying on scientific evidence to submit, dispute, or validate claims. For instance, over the last century, criminal courtrooms have selectively admitted neuroscientific models, exams, and insights claiming to revolutionize our understanding of who is culpable and deserving of punishment. Most recently, advancements in epigenetics have promised even more profound challenges to long-standing criminal law doctrines. This article examines the reasons reversibility and inheritability of epigenetic markers might warrant revising culpability and punishment and concludes that epigenetic findings are not yet robust enough to justify such revisions. ----- The paper based on this abstract was recently published into the Criminal Justice Ethics journal, and can be downloaded open access from https://doi.org/10.1080/0731129X.2024.2376444 . It has never been presented at conferences, and I believe this would represent a meaningful opportunity to do so!

At present, there is a progressive transformation of the doctor-patient relationship: from a certain hierarchy and paternalism in the doctor-patient relationship to the gradual distancing and automation of these relationships. The role of technology and digitalization of healthcare is a determining factor in this transformation and brought new challenges to medical civil liability. Although the advantages of new medical technological instruments and the use of artificial intelligence, the fact is that there are some possible dangers to the patient, which can suffer damages that have to be compensated. The aim of this presentation is to address these challenges posed to medical civil liability, resulting from the use of new technologies and artificial intelligence to civil liability.

Access to and collection of user data to form databases for training Artificial Intelligence (AI) as a medical device is a pressing need in the context of biomedical innovation. AI models depend on representative data to ensure the quality, safety and efficacy of the medical devices they are part of. However, access to such data must be harmonised with the fundamental right to privacy enshrined in European Union (EU) law, especially the European Convention on Human Rights and the General Data Protection Regulation (GDPR). This article analyses the tension between the need to diversify databases for AI training and the legal requirements of the personal data protection regime, exploring the viability of different legal grounds for their collection. The study starts from the premise that although consent is widely used as a legal basis for data processing, its practical application may be limited, especially in situations where retrospective collection of health data is required. As an alternative, the article advocates using the legitimate interest of the data controller as the most appropriate legal basis, despite it is not a silver bullet. It argues that this approach, combined with the legal basis of the need to ensure the quality and safety of medical devices, is compatible with the requirements of the GDPR and allows for greater flexibility in the re-use of data, provided that appropriate technical and organisational measures are implemented to protect the rights of data subjects. The analysis concludes that the use of legitimate interest in the safety of AI-based medical devices, combined with measures such as pseudonymisation, can effectively balance the need for innovation in healthcare with the protection of individual rights, contributing to a safer and more innovative environment for the implementation of AI in medical devices.

Adoption of Artificial Intelligence (AI) in Psychiatric Advance Directives (PADs) raises a number of legal and ethical concerns, especially regarding the patient’s capacity to make decisions for their treatment. This paper aims at finding out whether AI integration in PADs calls for changes in current legislation or in ethical standards. AI's capacity to forecast mental health emergencies and aid in decision-making presents novel obstacles: while in the process of positively enhancing patient care by making provision for early interventions, it poses risks of infringing with the patient’s autonomy if it is not well regulated.

This research is centred on the ways of applying ethical AI with the aim of promoting the patient’s self-determination in cases of the patient’s limited decision-making capacity. A useful consideration is therefore to avoid the technology from becoming an authoritative source of judgement to supplant human judgement or make unilateral judgements that contradict the patient’s evolving preferences.

In legal systems’ view, this study proposes that they ought to adapt by adopting guidelines that place emphasis on such concepts as informed consent, transparency, and human supervision. In addition, there are several features that an ethical perspective towards AI in PADs should consider, including revising directives regularly, making AI explain its reasoning, and involving collaborative decision-making between the patient and physicians. Suggestions for those on patient autonomy are raised to ensure that AI does not erode it but instead enrich it while stressing the need for integrated collaboration of the lawmakers, ethicists, and mental health professionals to provide solid protection.

The digital revolution has introduced technological advancements that are transforming contemporary society (Santos, 2024), while also bringing unprecedented challenges to criminal procedure, profoundly altering the way digital evidence is collected, processed, and used (Valente, 2018). The use of artificial intelligence (AI) in obtaining digital evidence, while offering efficiency, raises questions about its admissibility in light of fundamental rights and procedural legality (Monteiro, 2023). Thus, a brief analysis of the role of AI in obtaining digital evidence within the Portuguese and European legal context is sought, with a focus on the most recent case law and the challenges related to the protection of defence rights, privacy, and the integrity of evidence.

Considering that AI, through advanced techniques such as machine learning and predictive analysis (Monteiro, 2023), has shown potential to optimize the collection and analysis of large volumes of digital data, concerns arise regarding its compatibility with fundamental rights, such as privacy and the right to a fair trial (Fidalgo, 2022).

In Portugal, the Cybercrime Law and the Code of Criminal Procedure regulate digital evidence, but they do not specifically address AI, highlighting the need for legislative adaptation (Conde Correia, 2014). This can be inferred, for instance, from the Supreme Court's ruling of October 11, 2023, case no. 184/12.5TELSB-R.L1-A.S1, which addressed the issue of seizing electronic communication data in the context of a criminal investigation.

Therefore, it is safe to say that AI offers advantages by allowing for the quick and efficient analysis of large volumes of data, but it raises concerns about the transparency of algorithms, often considered "black boxes," which jeopardizes the right to defence (Fidalgo, 2022). In this regard, the Court of Justice of the European Union (CJEU) ruled in the Ministerio Fiscal and Digital Rights Ireland cases, emphasizing the need to ensure that data collection respects fundamental rights.

In summary, digital evidence faces various challenges due to its volatility and immateriality, which may compromise its integrity and admissibility (Monteiro, 2023), thus highlighting the need for legislative reform (Conde Correia, 2014).

Works cited

Doctrine

• Correia, João Conde, Prova digital: as leis que temos e a lei que devíamos ter, in Revista do Ministério Público 139, 2014.

• Fidalgo, Sónia, A utilização de inteligência artificial na obtenção de prova digital em processo penal, Instituto Jurídico, Faculdade de Direito da Universidade de Coimbra, 2022.

• Monteiro, António Pedro Pinto, Garantias Processuais e Prova Digital, III Jornadas a Sul do Direito Civil e Processual Civil”, Universidade de Évora, 2023.

• Santos, Margarida, A inteligência artificial, a obtenção da prova digital e a (des)proteção dos direitos fundamentais – breves considerações, 9° Congresso Internacional de Direito da Lusofonia: Governação, Justiça e Desenvolvimento Sustentável, 2024, in press.

• Valente, Manuel Monteiro Guedes, O Reforço dos Princípios Constitucionais na Obtenção de Prova no Mundo Digital, in RDPJ, ano 2 n.º 3, 2018.

Jurisprudence

• Supreme Court, October 11, 2023, case no. 184/12.5TELSB-R.L1-A.S1.

• Court of Justice of the European Union, April 8, 2014, joined cases C‑293/12 and C‑594/12.

• Court of Justice of the European Union, case C-207/16.

Based on the studies and proposals of the High-Level Expert Group on Artificial Intelligence (AI), the EU Treaties, the EU Charter of Fundamental Rights and International Law, the European Union presented the Ethical Principles of AI: the Principle of Respect for Human Autonomy, the Principle of Prevention of Harm, the Principle of Fairness and the Principle of Explainability. How can we guarantee the application of these principles in cases of Autonomous Vehicles (AV) action? What guidelines should VA software developers follow when creating a VA programme? Should they allow the VA to learn autonomously in order to decide how to solve ethical dilemmas, such as decide whether to run over one person or another if it is impossible to avoid a collision? Is it permissible for a machine to replace man when it comes to deciding who should live or die?