Artificial Intelligence and Genealogy

This review explores quantitative genealogical and forensic analyses of genetic data, illustrated with case examples that introduce the core principles of probabilistic inheritance and analytical techniques for determining ancestral lineages and identity matches. The discussion emphasizes the use of short DNA polymorphism variant sequences, including single-nucleotide polymorphisms (SNPs), short tandem repeats (STRs), and microhaplotypes (MHs), as biomarkers. Forensic DNA index system databases were discussed The illustrative case examples include tracing the ancestral origins of domesticated dogs from wolf populations by applying principles of inheritance and probabilistic analyses to identify their likely dual ancestry ancestors. Another example tracks common ancestors before the branching of the DRD4-7R dopamine D 4 receptor gene variant tree, characterized by the 7-repeat variant VNTR (variable number tandem repeats), serving as a biomarker to analyze the natural selection process that led to beneficial outcomes despite the adverse effects of disorders like ADHD (attention-deficit hyperactivity disorder) and novelty-seeking behavior. Additionally, another case example discusses tools that address the limitations of DNA samples in non-invasive prenatal paternity testing (NIPPT) by utilizing maternal blood samples to obtain fetal DNA. The alternative artificial intelligence (AI) approach in genealogical analysis was discussed, focusing on two distinct AI methods for addressing complex problems. The traditional algorithmic AI approach relies on programming analytical methodologies into expert systems with clearly defined statistical inference techniques to solve these problems analytically. However, it struggles to address unexpected scenarios and cannot manage gaps in missing genetic data. The alternate branch of the AI approach addresses these issues by employing machine learning to derive solutions without specifying the exact methodologies or algorithms needed to tackle these problems. Case studies illustrate the principles and history of neural computation alongside machine learning, demystifying the AI magic that excels at summarizing large datasets while lacking explainable reasoning in its generated results. The advantages and practical limitations of machine learning provide informed criteria for evaluating suitable AI model performance. The limitations include the requirement for a large dataset for training to identify suitable solutions and learn from examples to summarize data through input-output correlation, which is a time-consuming process. Depending on the AI models using supervised or unsupervised learning paradigms, they may or may not need human guidance during training with known solutions. These tools can help fill gaps in genealogical databases, and while these outcomes are possible, they may not always have a rational explanation. Their solutions may not lead to the optimal one, as other alternatives could be more effective. Until future AI models improve their reasoning abilities, it is wise to exercise caution when concluding. Nonetheless, these systems effectively automate the organization of genealogical databases from written and spoken records, and they can produce automated responses to users’ inquiries by utilizing natural language processing to address questions from everyday users.