Jędrzej Wydra

The publication is the first Polish commentary on the EU Artificial Intelligence Act (AI Act), which governs the development, deployment, and use of AI systems in the European Union. It provides a detailed, article-by-article analysis of the regulation, covering risk classification, prohibitions on high-risk AI, design and testing requirements, provider and distributor obligations, certification procedures, market supervision, and the protection of fundamental rights. The commentary also explores the relationship between the AI Act, GDPR, and DSA, referencing official EU documents and the draft Polish AI law.

The work, created through collaboration between lawyers, data scientists, and AI engineers, offers nearly 100 practical examples of applying legal provisions in real-world scenarios. These examples address risk classification, cybersecurity, ethical concerns, system labeling, certification, and incident response. This interdisciplinary approach bridges legal, ethical, and technical perspectives, offering practical guidance for lawyers, public officials, IT specialists, and business leaders implementing AI solutions in compliance with EU law.

Estimating time of death from insect evidence traditionally relies on the “law of total effective temperature,” with developmental parameters derived from linear regression methods such as Ikemoto and Takai’s approach. However, these methods provide only point estimates, neglecting population variability and uncertainty.

We developed an alternative statistical method that simultaneously estimates interval values for developmental parameters and identifies component populations within datasets. The approach fits finite mixtures of Weibull distributions to development time data using the Expectation–Maximization (EM) algorithm, allowing the capture of individual-level variation.

Using developmental datasets of Creophilus maxillosus and Necrodes littoralis (Staphylinidae), our model achieved 95% coverage near the nominal level, while classical methods captured only 59% and 75% of cases. This improvement demonstrates that mixture-based interval estimation better reflects biological heterogeneity and enhances the reliability of postmortem interval calculations. Beyond forensic entomology, the method offers a generalizable framework for thermal ecology and applied biological modeling.

This study examines the reliability of fingerprint experts in assessing the individualization value of minutiae in latent fingerprint traces. Despite the central role of fingerprint evidence, examiner variability and lack of verification protocols raise concerns about reliability.

Thirty Polish experts identified and rated seven minutiae in two fingerprint traces of differing quality on a five-point scale. Inter-rater agreement was measured with Krippendorff’s alpha, and correlations were used to explore the link between selection frequency and perceived value.

Results showed low agreement in both selection and evaluation (α = 0.39 for high-quality and 0.42 for low-quality traces) and a strong negative correlation between selection frequency and perceived uniqueness (r = -0.84 and -0.89). These findings highlight significant variability in expert judgments and suggest that widely selected features are viewed as less discriminative.

The study emphasizes the need for systematic verification. Multi-expert reviews could improve reliability and reduce the risk of forensic identification errors.

We investigated independence testing for functional data, which may be either univariate or multivariate. Broadly speaking, our approach involved first reducing the dimensionality of the functional data using basis expansion and then applying the distance of mean embedding—a flexible measure of independence. We enhanced this method for pairwise independence by incorporating marginal aggregation, as well as asymmetric and symmetric aggregation measures, to improve test performance and adapt it to mutual independence testing.

Our methods are compared with tests based on distance covariance and the Hilbert–Schmidt independence criterion. To evaluate their effectiveness, we presented simulation studies and two real data examples using air pollution and chemometric data sets. The new testing procedures demonstrated favorable finite-sample properties, effectively controlling the type I error rate and exhibiting competitive power, making them viable alternatives to covariance-based tests.

Accurate reconstruction of ambient temperature at a crime scene is vital for estimating postmortem interval (PMI). The standard approach corrects weather station data using on-site measurements and linear regression, but it often requires several days of data collection.

This study introduces two functional data analysis (FDA) models that reconstruct ambient temperature using only temperature readings while shortening the required measurement period. The concurrent regression model supports mid-term reconstruction, while a Fourier-based model enables accurate short-term estimation.

Both models were tested on data from six quasi-indoor environments, including heated and unheated buildings and underground conditions. The concurrent model achieved near-perfect accuracy after six days, and the Fourier model matched its performance after only four to five hours.

These FDA-based methods, compatible with existing forensic procedures, show that reliable PMI temperature correction can be achieved with minimal measurement time.

Aggregations of juvenile insects are key models for studying social interactions shaped by competition, cooperation, and parental effects. This study investigates these dynamics in Necrodes littoralis, a carrion beetle that forms large larval colonies on vertebrate carcasses. Adults produce a feeding matrix from exudates that generates heat and may influence larval fitness.

We hypothesized that exploitative competition dominates larval behavior, while cooperative interactions arise from collective exodigestion. Using experimental colonies with manipulated parental effects (present/absent) and varying larval densities (0.02-1.9 larvae/g), we measured fitness components, development time, and thermogenesis.

Results showed a strong negative group-size effect on larval fitness in colonies with parental effects and in high-density groups without them. However, development time shortened and heat production increased with density, indicating partial benefits of aggregation. Positive fitness effects emerged only at low densities without parental influence.

These findings suggest that larval societies of N. littoralis are primarily shaped by exploitation competition, moderated by both density and parental effects.

Estimating postmortem interval (PMI) from insect evidence relies on thermal summation constants (k), which describe developmental duration in accumulated degree-days or hours (ADD/ADH). Standard k values represent mean population traits and may misestimate cases involving unusually small or large specimens.

To address this limitation, we developed a statistical formula that calibrates k for specific insect evidence without requiring explicit “k versus size” models. The approach integrates the general thermal summation constant (with standard error), published species size ranges, and measurements of individual insects collected at the death scene.

The method was validated using datasets of Creophilus maxillosus and Necrodes littoralis (Coleoptera). For atypically sized insects, the formula reduced mean errors by approximately 25 ADD in C. maxillosus and 40 ADD in N. littoralis.

This calibration approach enables more accurate PMI estimation, particularly for outlier specimens, and can be applied in forensic casework where species-specific size–development data are limited. It provides a practical, reproducible way to account for biological variation without requiring additional experimental datasets.

Previous methods of evaluating evidence from handwriting examinations were usually associated with a redefinition of how these examinations are to be made. Here, we propose the likelihood ratio method for handwriting evidence evaluation, which is fully compatible with the current handwriting examination protocols.

The method focuses on the similarity between handwriting samples, quantified using the Jaccard index from the results of a typical forensic handwriting comparison. The numerator of the likelihood ratio is the probability of a given class of similarity, assuming that a given person wrote the questioned sample. The denominator is the probability of the same similarity class, assuming that a randomly selected person wrote the questioned sample. The similarity distribution to quantify the numerator is derived from comparisons across reference handwritings. To calculate the denominator, we proposed developing similarity distributions relevant to particular forensic scenarios. In the proof-of-concept study, we developed the distribution for the simulation scenario.

Empty puparia, the hardened exoskeletons left after fly emergence, are often found at death scenes with long postmortem intervals (PMI). While the post-eclosion interval (PEI)—the time between fly emergence and collection—cannot be directly measured, it affects PMI estimation.

This study simulated PMI estimation for empty puparia of Protophormia terraenovae and Stearibia nigriceps under varying PEIs (0-90 days) and temperature conditions.

PEI did not affect PMI for specimens collected in winter or early spring (December-April). From late spring to autumn (May-November), PMI increased with PEI, most strongly in summer and only slightly in autumn.

The shortest PMI always corresponded to a PEI of 0, confirming that the minimum PMI is obtained with this assumption. For spring samples, oviposition likely occurred the previous year, while in summer, this was indicated only for longer PEIs.

These results refine PMI estimation from empty puparia and identify the conditions under which PEI can be ignored or not.