HumanGenesis: Evolution, Regulation, and Synthesis

• HumanGenesis is a multifaceted concept encompassing human origins, genomic regulation, and synthetic modeling that unifies evolutionary, demographic, cognitive, and computational research.
• It examines evolutionary processes like introgression and migration alongside human-specific regulatory controls and digital human reconstructions.
• Research in HumanGenesis employs diverse methodologies—from genomic sequencing and statistical modeling to generative AI—to elucidate human diversity and emergence.

In the cited literature, HumanGenesis is used to frame a set of questions about how humans originate, diversify, are regulated at the genomic level, and are synthetically modeled. The term spans hypotheses of hominin speciation through introgression, demographic and population-genetic reconstructions of global ancestry, genome-scale analyses of human-specific regulatory architecture, models of the transition from genetic to memetic or cognitive regimes, and generative systems that synthesize human genotypes, digital humans, or expressive human motion (Nygren, 2018, Elhaik et al., 2012, Glinsky, 2019, Carter, 2010, Kenneweg et al., 2024, Jiang et al., 2022, Li et al., 13 Aug 2025). This breadth implies that HumanGenesis is not a single theory, but a family of origin narratives and constructive models operating at evolutionary, genomic, cognitive, and computational scales.

1. HumanGenesis as a plural research concept

One major usage treats HumanGenesis as a problem of deep human origins: which lineages, introgression events, and demographic processes produced Homo and its close relatives. A second usage treats it as a problem of reconstructing population structure and migration from genome-wide markers, including archaic introgression and regional demographic history. A third shifts the emphasis from lineage branching to regulatory control, asking how fixed human-specific regulatory changes and endogenous retroviral elements shape development, survival, and pathology. A fourth treats HumanGenesis as constructive modeling: generating synthetic human genotypes, digital human radiance fields, or photorealistic videos of human motion (Nygren, 2018, Elhaik et al., 2012, Glinsky, 2019, Glinsky, 2023, Kenneweg et al., 2024, Jiang et al., 2022, Li et al., 13 Aug 2025).

The coexistence of these usages is consequential. In some works, HumanGenesis is primarily historical and phylogenetic; in others, it is mechanistic and regulatory; in still others, it is explicitly generative, meaning the artificial production of genomes, memory traces, or human-centered visual data. This suggests that the term functions less as a stable disciplinary label than as a recurring attempt to unify origin, inheritance, and generation within a single conceptual frame.

2. Origin hypotheses, speciation, and deep-time narratives

A particularly explicit evolutionary usage appears in the proposal that hominin evolution was caused by introgression from Gorilla (Nygren, 2018). That paper argues that gorilla introgression into the ancestor of Pan and Homo around the Pan–Homo split, together with lineage sorting of a hybrid genome, explains the observation that about 30% of the gorilla genome shows lineage sorting between humans and chimpanzees and accounts for the shared chromosome 5 NUMT “ps5” found in Gorilla, Pan, and Homo. In that reconstruction, roughly 15% of gorilla-derived genomic segments become part of the chimpanzee lineage and roughly 15% become part of the human lineage, while the same introgression event is further extended to the divergence of Australopithecus and Paranthropus, including the proposed reclassification of Au. deyiremeda as Paranthropus deyiremeda (Nygren, 2018). The paper itself states that this interpretation is intentionally bold and not a mainstream consensus, and it explicitly notes that standard explanations invoke incomplete lineage sorting and that formal D-statistics, $f_4$-ratio tests, or phylogenetic network models are not presented.

A distinct deep-time narrative appears in the demographic model of reproductively advantageous and disadvantageous regions (Enflo et al., 2017). In that framework, neutral genetic lineages in a reproductively disadvantageous region decay exponentially as $(1-\varepsilon)^n$, while a reproductively advantageous source region can come to dominate the gene pool of a chain of regions through long-term asymmetric migration. The model is used to explain low genetic variation in modern humans and Western Neanderthals, the overwhelmingly African origin of modern mtDNA and Y-chromosomes, the young age of much of the European gene pool, and the observation that East Asians carry more Neanderthal-derived nuclear DNA than Europeans (Enflo et al., 2017). Its importance within HumanGenesis lies in replacing singular bottleneck narratives with long-duration demographic asymmetry.

An even broader cosmological variant places HumanGenesis within cometary panspermia (Wickramasinghe, 2012). In that account, life originated cosmically as a unique event and Earth was a recipient rather than the original cradle; subsequent evolution was driven not only by mutation and selection but also by episodic acquisition of cosmically derived genes, often through viral or horizontal transfer mechanisms (Wickramasinghe, 2012). The paper explicitly presents this as a cosmic theory of life rather than a conventional terrestrial evolutionary model. A plausible implication is that, within the HumanGenesis literature, “origin” can range from hybrid speciation within African apes to gene-flow models across continents to explicitly extra-terrestrial accounts of the origin and diversification of life.

3. Population-genetic reconstruction of migrations, admixture, and regional structure

A more empirical usage of HumanGenesis centers on global ancestry reconstruction. The GenoChip was introduced as a dedicated, non-medical genotyping platform for genetic anthropology, with over 130,000 autosomal and X-chromosomal SNPs, an unprecedented number of Y-chromosomal and mtDNA SNPs, and AIM ascertainment from over 450 populations (Elhaik et al., 2012). Its design explicitly targets population structure, migration history, and archaic hominin introgression while excluding medically relevant markers. In comparative analyses of Africans, Europeans, and Asians, the GenoChip autosomal SNPs had mean $F_{ST}=0.10$ and the X-chromosomal SNPs had mean $F_{ST}=0.13$, exceeding the corresponding commercial arrays and thereby increasing resolution for continental and sub-continental structure (Elhaik et al., 2012). This work situates HumanGenesis within the practical infrastructure of population genomics.

An exploratory total-evidence analysis combined five SNP datasets into a joint matrix of 4025 individuals and 3146 autosomal SNPs and applied model-based clustering across $K=2$ to $K=16$ (Li, 2011). Despite the sparse overlap, the analysis recovered coherent continental structure, a distinct South Asian component, fine-scale East and Southeast Asian differentiation, northern and southern Native American structure at $K=16$, and a notable association between the ancestry profiles of Andamanese, Malaysian and Philippine Negritos, Lesser Sunda populations, and the Khoisan–Pygmy component resolved within Africa (Li, 2011). The paper treats these findings as exploratory and explicitly states that cross-study data confrontation is hypothesis-generating rather than definitive. Within HumanGenesis, its contribution is methodological: it shows that even thinly overlapping autosomal datasets can reveal branching and admixture patterns relevant to early out-of-Africa history.

The first region-wide genomic analysis of Oceania further deepens this population-genetic interpretation (Quinto-Cortés et al., 2024). It generated and analyzed genome-wide data from 981 individuals in 92 populations across 58 islands and 30 countries, disentangling Papuan and more recent Austronesian ancestries and characterizing archaic introgression separately in each component (Quinto-Cortés et al., 2024). The study reports that Papuan Highlanders retain nearly 100% Papuan ancestry, that Lapita-associated ancient genomes cluster within the Austronesian ancestry-specific clade connected to present-day Polynesians, and that Denisovan ancestry is largely confined to Papuan segments, typically at 2–5% within Papuan ancestry across many Oceanian populations (Quinto-Cortés et al., 2024). A reverse migration from western Polynesia back to the southern Philippines is inferred from IBD, outgroup $f_3$, and the directionality index $\psi$. In this setting, HumanGenesis becomes a layered regional history in which out-of-Africa dispersal, archaic introgression, Austronesian expansion, founder effects, and later back-migrations are jointly modeled.

4. Human-specific regulatory architecture and retroviral control

Another major strand of HumanGenesis moves away from species trees and toward the regulatory basis of human-specific phenotypes. One study analyzed 35,074 human-specific regulatory single-nucleotide changes fixed in modern humans and located in differentially accessible chromatin regions during human versus chimpanzee neurogenesis in cerebral organoids (Glinsky, 2019). GREAT assigned 34,010 of these SNCs to 8,405 genes, and gene set enrichment analyses linked those genes to more than 1,000 anatomically distinct adult human brain regions, more than 200 common disorders, and more than 1,000 rare-disease records (Glinsky, 2019). The same work identified 645 genes associated with autosomal dominant inheritance, 849 with autosomal recessive inheritance, and 2,273 genes associated with premature death, embryonic lethality, and pre-, peri-, neo-, and post-natal lethality phenotypes in mouse models (Glinsky, 2019). It also notes that 99.8% of the 35,074 hsSNCs are shared with archaic humans and only 64 are unique to modern humans, implying that most of this regulatory scaffold predates the split between archaic and modern Homo (Glinsky, 2019).

A closely related but more expansive regulatory account is given by the analysis of human embryo retroviral LTR elements (Glinsky, 2023). That work defines a global retroviral genomic regulatory dominion composed of 8,839 highly conserved fixed LTR elements derived from LTR7, MLT2A1, and MLT2A2, linked to 5,444 downstream target genes and organized into 26 genome-wide multimodular genomic regulatory networks and 69,573 statistically significant genomic regulatory modules at Binominal FDR q-value threshold $0.001$ (Glinsky, 2023). Individual GRNs occupy from 5.5% to 15.09% of the human genome and contain 529–1486 retroviral LTRs and 199–805 downstream genes (Glinsky, 2023). The enriched biological domains include regulation of nervous system development, synapse part, postsynaptic membrane, neuron projection, regulation of membrane potential, cell-cell adhesion, and offspring-survival phenotypes (Glinsky, 2023). The paper further connects these LTR-linked networks to human-specific regulatory sequences, rapidly evolving TADs, and species-biased expression in human–chimpanzee hybrid iPSC and cortical spheroids, with especially strong enrichment in Dentate Granule Cells and brain modules linked to learning and memory (Glinsky, 2023).

Taken together, these studies recast HumanGenesis as a problem of regulatory architecture. Human distinctiveness is attributed not primarily to novel protein-coding genes, but to fixed human-specific changes in cis-regulatory regions and to exapted endogenous retroviral elements that orchestrate embryogenesis, survival, brain development, and disease susceptibility (Glinsky, 2019, Glinsky, 2023). This suggests a shift from asking when humans branched from other apes to asking how conserved genes were rewired into human-specific developmental and physiological programs.

5. From genetic evolution to memetic and cognitive generation

HumanGenesis has also been framed as a transition in the mode of information transmission. A quantitative account distinguishes a first phase of hominid development governed by neo-Darwinian genetic evolution from a second phase governed by memetic evolution (Carter, 2010). In that model, the biological phase is characterized by rapid increase of brain size within Homo, whereas the later phase, beginning with Homo sapiens, is dominated by technological evolution and hyperbolic population growth following the Foerster law

$(1-\varepsilon)^n$0

with $(1-\varepsilon)^n$1 years (Carter, 2010). The same paper argues that $(1-\varepsilon)^n$2, where $(1-\varepsilon)^n$3 is the human generation timescale and $(1-\varepsilon)^n$4 is the order of $(1-\varepsilon)^n$5 bits of genomic information, and interprets the origin of Homo as a possible hard step while characterizing the emergence of Homo sapiens as “rather an automatic process” once large-brain evolution had entered a maximal regime (Carter, 2010). Within HumanGenesis, this is a transition from genome-limited to meme-limited dynamics.

At the cognitive level, a related generative formalism appears in the Generative Episodic–Semantic Integration System, which models memory as the interaction between a Cortical-VAE and a Hippocampal-VAE embedded in a retrieval-augmented generation architecture (D'Alessandro et al., 17 Oct 2025). The model reproduces semantic generalization, recognition, serial recall effects, gist-based distortions, and constructive episodic simulation under explicit capacity constraints (D'Alessandro et al., 17 Oct 2025). Semantic memory is represented by a conditional $(1-\varepsilon)^n$6-VAE over latent content $(1-\varepsilon)^n$7 and embeddings $(1-\varepsilon)^n$8, episodic memory by a compressed latent key $(1-\varepsilon)^n$9, and retrieval by query–key matching over a key–value store (D'Alessandro et al., 17 Oct 2025). Although this work addresses human cognition rather than evolutionary origins, it extends the HumanGenesis motif into the computational genesis of human-like memory: recall, imagination, and distortion are treated as outputs of bounded generative systems rather than passive storage.

6. Synthetic HumanGenesis: genomes, digital humans, and future lineages

In computational genomics, HumanGenesis has become literally generative. A diffusion-based system for complete synthetic human genotypes introduces a model trained on gene-wise PCA embeddings of Project MinE genotypes and 1000 Genomes haplotypes, using a continuous DDPM on latent genotype space (Kenneweg et al., 2024). Synthetic-only training with the best generator yields classifier recovery rates of about 94.26% for ALS with an MLP classifier and 93.02% for 1000 Genomes ancestry classification with an MLP classifier, while augmenting 5% real data with synthetic genotypes raises ALS accuracy from 70.96% to 84.83% and 1KG ancestry accuracy from 29.01% to 83.98% (Kenneweg et al., 2024). The paper explicitly states that complete synthetic genotypes can, by standard protocols, be expanded into full-length, DNA-level genomes (Kenneweg et al., 2024). Here HumanGenesis denotes the controlled generation of synthetic human genomic profiles and populations.

A visual-computational analogue appears in HumanGen, a 3D human generation system that combines a pretrained 2D StyleGAN-human generator, an anchor image, tri-plane features, an SDF geometry module, and a two-stage blending scheme for appearance (Jiang et al., 2022). On 3D human generation benchmarks, it reports FID 20.97 and depth difference 0.0201, outperforming EG3D, StyleSDF, GNARF, and a 2D-G + PIFu baseline (Jiang et al., 2022). The anchor image links 2D latent editing to 3D human radiance fields, allowing off-the-shelf 2D latent editing methods to be lifted into 3D (Jiang et al., 2022). In this usage, HumanGenesis refers to the creation of complete, editable digital humans.

The term is used even more directly in the four-agent framework “HumanGenesis: Agent-Based Geometric and Generative Modeling for Synthetic Human Dynamics” (Li et al., 13 Aug 2025). That system integrates a Reconstructor based on 3D Gaussian Splatting and deformation decomposition, a Critique Agent that uses multi-round MLLM-based reflection to identify poor regions, a Pose Guider with time-aware parametric encoders, and a Video Harmonizer that couples hybrid rendering with diffusion and feeds back through a Back-to-4D loop (Li et al., 13 Aug 2025). On HumanVid, it reports PSNR 21.934, SSIM 0.802, LPIPS 0.223, FID-VID 76.20, and user rating 4.103, outperforming AnimateAnyone, AniGS, MIMO, and Champ; on NeuMan it reports SSIM 0.891, PSNR 26.992, and LPIPS 0.081, outperforming Vid2Avatar, NeuMan, and HUGS (Li et al., 13 Aug 2025). HumanGenesis here names a framework for synthetic human dynamics rather than biological origin, but the semantic continuity is clear: the system generates photorealistic, intention-driven human motion with explicit geometric and generative priors.

A prospective, population-genetic form of synthetic HumanGenesis appears in simulations of interstellar crews (Marin et al., 2021). The HERITAGE code models a closed multi-generational human population with 46 chromosomes, 23 pairs, 2,110 loci, 10 allelic forms per locus, meiosis with crossing-over and unilateral conversion, and mutation from cosmic ray bombardments under the neutral hypothesis (Marin et al., 2021). The authors conclude that centuries-long deep-space travel produces small but unavoidable genetic differentiation under Earth-like background doses and stronger divergence under larger doses (Marin et al., 2021). This suggests a future-oriented HumanGenesis in which new lineages emerge through isolation, drift, and radiation rather than through terrestrial population structure alone.

7. Controversies, epistemic status, and methodological boundaries

The HumanGenesis literature is unusually heterogeneous in evidential status. Some contributions are instrumentation or methods papers grounded in standard population genetics, such as GenoChip, Oceania-wide ancestry-specific analyses, synthetic genotype diffusion models, or 3D/diffusion systems for digital humans (Elhaik et al., 2012, Quinto-Cortés et al., 2024, Kenneweg et al., 2024, Jiang et al., 2022, Li et al., 13 Aug 2025). Others are theoretical syntheses or speculative narratives. The gorilla introgression hypothesis explicitly contrasts its interpretation with the standard coalescent-based incomplete-lineage-sorting account and notes the absence of formal likelihood or Bayesian tests in the paper itself (Nygren, 2018). The cosmic theory of life presents horizontal gene transfer and retroviral integration as vindications of cometary panspermia, but its extra-terrestrial extension remains a distinct cosmological thesis rather than a mainstream account of human evolution (Wickramasinghe, 2012).

Even within genomics, caution is required. The hsSNC study is explicit that its evidence is bioinformatic and correlative and that direct experimental validation of individual regulatory SNCs remains largely to be done (Glinsky, 2019). The retroviral LTR study depends on GREAT-based target assignment, ontology enrichment, and DAG validation; its statistical scale is large, but its mechanistic claims remain tied to regulatory inference rather than universal locus-by-locus causality (Glinsky, 2023). The cross-study SNP synthesis based on 3,146 overlapping SNPs was framed as exploratory from the outset (Li, 2011). The diffusion-based synthetic genotype work demonstrates non-copying empirically through NN adversarial accuracy and distance checks, but states that no theoretical privacy guarantee is provided in the absence of differential privacy (Kenneweg et al., 2024). The synthetic human dynamics framework notes continuing difficulty with highly dynamic scenes, fine-grained facial expressions, and intricate hand motion (Li et al., 13 Aug 2025).

These limits are not merely technical. They indicate that HumanGenesis is a boundary concept linking population genetics, developmental regulation, cognition, and generative modeling. In some contexts it denotes the origin of Homo; in others, the origin of modern human regulatory architecture; in others, the artificial generation of genomes, memories, or videos of humans. The unifying feature is constructive ambition: each usage seeks to explain how human form, function, or representation comes into being, whether through introgression, migration, regulation, memetics, or synthesis.