The article, authored by independent researcher Nan Chen of Chengdu, China, maps ten components of a previously proposed dual-architecture AGI system onto the biological timeline of brain maturation. The central claim is deliberately provocative: machine intelligence may need to retrace the milestones of biological intelligence, acquiring physical tokens before social ones, building redundant capacity before compressing it, and pausing periodically to consolidate memory. Crucially, the author frames the thesis as falsifiable. If an AGI system could reach human-level physical and social reasoning while ignoring the proposed ordering, the core claim would collapse.

The developmental story begins at birth, with what the paper calls phase zero. The only fully myelinated structures in a newborn brain are the brainstem, spinal reflex arcs, and portions of thalamic and basal ganglia circuits. These are, in effect, factory pre-installed: respiration, grasp reflexes, and startle responses operate immediately without learning, at millisecond speeds. In the proposed AGI design, this maps onto a frozen reflex layer, a pretrained vision-language model with fixed weights that delivers predictions in a single forward pass. Meanwhile the prefrontal cortex, the seat of human intelligence, is almost entirely unmyelinated at birth. The most sophisticated part of the brain is built after contact with the world, not before, a fact that corresponds directly to the design principle that the AGI continuous learning layer starts with near-random weights.

From birth to age three, the brain runs two massive programs. First, synaptic overproduction: by ages two to three, cortical neurons average around 15,000 synapses each, roughly double the adult density, creating an enormous hypothesis space that experience will later prune. Second, myelination sweeps posterior to anterior, from visual cortex through somatosensory and motor regions to language areas and finally the prefrontal cortex. This sequence mirrors behavior: see first, then touch, then move, then speak, then plan. Two mechanisms launch during this window that the paper considers essential for AGI. Dopamine neurons fire not for reward itself but for prediction error, the gap between expected and actual outcomes, which grounds the architecture’s curiosity module. And hippocampal place-cell sequences replayed during sleep at roughly twenty times speed consolidate fragile memories into stable cortical storage, the biological rationale for a replay buffer that prevents catastrophic forgetting in artificial networks.

Between ages three and four, the system locks. Large-scale synaptic pruning begins, eliminating unused connections under a use-it-or-lose-it rule until roughly half of childhood synapses are gone by adolescence. The paper identifies this as the biological counterpart of token migration: patterns verified across enough interactions, exceeding an accuracy threshold and a minimum interaction count, are promoted and frozen into the reflex layer, while unverified candidates are discarded. Infantile amnesia ends at the same time, plausibly because declining hippocampal neurogenesis stops overwriting memory traces. In the AGI framework, experiences become retrievable knowledge only once abstracted into named tokens and written into a structural constraint graph; before that, they exist only implicitly as weight adjustments. The author illustrates the mechanism concretely: a robot that learned the physical token of force from a thousand box-pushing interactions can, when observing human arguments, use a learned transformation matrix to initialize a social token of persistence, reusing the causal structure of physical transfer as a representational prior rather than a metaphor.

The longest phase stretches from age three to about twenty-five, as the prefrontal cortex, responsible for planning, inhibition, and theory of mind, slowly myelinates last of all brain regions. The paper argues this ordering is a dependency, not a delay: social cognition is built on physical cognition. One cannot ground the social concept of a promise without first internalizing, through physical experience, the notion that actions constrain future outcomes, an intuition rooted in gravity, friction, and collision. In the architecture, this translates into a hard sequencing rule: social token acquisition presupposes that physical tokens are already internalized in the reflex layer. Self-modeling compounds the dependency, since mirror self-recognition in humans emerges at 18 to 24 months only after basic sensorimotor coordination, and a self-token in the framework must be composed from lower-level physical and social tokens.

A common objection is that the brain’s schedule is dictated by carbon, not computation: energy metabolism, molecular signaling, and cranial volume. The paper counters that carbon and silicon face deeply isomorphic optimization problems. Myelination boosts conduction speed from about one meter per second to roughly one hundred and cuts energy use, the same trade-off as migrating frequently used tokens from a slow, gradient-updated layer into a fast, single-pass layer. Synaptic overproduction followed by pruning echoes the lottery ticket hypothesis in machine learning, where large over-parameterized networks contain sparse subnetworks that generalize better after pruning than compact models trained directly. And the necessity of sensorimotor calibration before abstract reasoning parallels the collapse that occurs when a high-level classifier is trained on unconverged low-level features. The brain develops in sequence, the argument runs, not because it is flesh, but because it is a hierarchical learning system bootstrapping abstractions from raw data.

The author backed individual mechanisms with experiments on the MuJoCo physics engine, all run on a consumer MacBook CPU with no GPU. In a three-task continual learning benchmark on the Ant-v5 quadruped, elastic weight consolidation retained 93.2 percent of baseline performance after two task switches, synaptic intelligence 102.3 percent, and memory-aware synapses 82.7 percent, all above the 80 percent target, while unprotected training was erratic and, in a separate run, retained only 71.7 percent. The paper candidly notes that a 27 percent baseline variance limits the precision of those comparisons. A curiosity experiment showed that intrinsic prediction-error reward drove exploration roughly seven times deeper into novel territory than extrinsic reward alone. In a multimodal task on HalfCheetah-v5, temporally aligned vision, force, and audio signals predicted one another across all tested directions, while shifting the signals by just twenty steps drove performance below chance, direct evidence that shared representations are built on temporal synchrony.

Perhaps the most striking experimental finding concerns negative transfer. When a pretrained reflex layer was fine-tuned on genuinely novel physics, high-gravity environments, warm-starting with pretrained weights actively hurt: the frozen model scored better than any fine-tuned variant. A systematic gradient across five gravity levels identified a crossover threshold between roughly 1.25 and 1.5 times Earth gravity, beyond which training from scratch beats incremental adaptation. Weight-space analysis showed why: the pretrained solution sat too far from the new optimum. The revised routing strategy mirrors biology, where reflex arcs do not change with experience: default to frozen reflex predictions, train an independent learning layer from random initialization in the background, and switch when its confidence surpasses the reflex layer’s. The paper also argues that AGI may need explicit sleep-like cycles, offline consolidation phases without new input, to avoid overwriting old knowledge, and that the 25-year human timeline conflates clock time with experience time. If humans need on the order of 100 million effective physical interactions and simulators reach tens of thousands per hour, the equivalent developmental span could compress to about a year without skipping a single stage.

The author is careful about limits. The correspondence may be analogy rather than isomorphism, the experiments cover only six mechanisms at modest complexity, adult neuroplasticity suggests developmental stages are more flexible than the framework assumes, and the central falsification condition, an AGI reaching human-level reasoning without the ordering, cannot yet be tested because no such system exists. Still, the engineering paradigm on offer is a genuine inversion of the dominant scaling philosophy. Childhood, in this view, is not a bug but a feature: build an excessively redundant system, let experience prune it, freeze what works, and only then attempt abstraction. Nature took 3.5 billion years to find the sequence. Silicon need not take 25 years, but the paper bets it must still take them in the right order, physical before social, redundant before compressed, interactive before consolidated.

Subject of Research: Mapping human neurodevelopmental stages onto a dual-architecture artificial general intelligence framework

Article Title: Neurodevelopment as blueprint: biological validation of the dual-architecture AGI framework

Article References: Chen, N. (2026). Neurodevelopment as blueprint: biological validation of the dual-architecture AGI framework. Frontiers in Robotics and AI, 13, Article 1902952. https://doi.org/10.3389/frobt.2026.1902952

Image Credits: AI Generated

DOI: 10.3389/frobt.2026.1902952

Keywords: artificial general intelligence, neurodevelopment, dual architecture, catastrophic forgetting, token migration, continual learning, curiosity-driven exploration, multimodal alignment, synaptic pruning, myelination, negative transfer, MuJoCo

The trilogy began in 2016 with a collection on self-organization in the nervous system, continued in 2021 with an examination of understanding capacity in humans and machines, and now closes with a volume on the neurobiological foundations of cognition. Across these installments, the editors have pressed a set of deceptively simple questions. What is intelligence? How does human intelligence differ from that of other species? How could intelligence emerge in a physical universe at all? And what factors explain the startling breakthroughs of the last decade in machine learning, a field still built on ideas about digital neurons and perceptrons that date to the mid-twentieth century?

The editors’ answers begin with a distinction that most of the AI industry has blurred. All species can learn, they argue, but the capacity for understanding is uniquely human. They borrow a definition attributed to the developmental psychologist Jean Piaget: intelligence is what you use when you do not know what to do, when neither instinct nor prior training has prepared you for the situation. Novel situations, in this view, demand a dual feat, constructing an adequate response while simultaneously overcoming the inertia of previous learning, the legacy mindset that interferes with the construction. Learning drills in responses from the outside; understanding decides internally, and it is precisely this internal decision-making that current AI systems lack.

To ground this claim, the editorial assembles a century of psychological theory into a single technical picture. From Gestalt psychology it takes the insight that sensory units exist as segregated wholes before they acquire names or uses, and that this original segregation is what allows the world to appear imbued with meaning. From Piaget it takes the characterization of understanding as a system of operational coordinations that transform objects of thought the way actions transform material objects, treating any observed series as one sector of an unlimited range of possibilities. From William James it takes the image of attention hovering near a desired object until accumulated neural tension breaks through a barrier and the nervous wave pours into a tract that has long been awaiting its advent. And from Donald Hebb it takes the cornerstone idea of neuronal assemblies, groups of neurons acting as the basic functional units of the brain, whose interactions produce percepts that are genuinely new structures rather than mere sums of their parts.

The hard problem, the editors argue, is combinatorial. Meaning production is coextensive with choice, and choices explode combinatorially. An array of just ten objects affords 1,024 possible selections; accounting for permutations, in which A following B differs from B following A, raises the count to 3,628,800. A chess position with 32 pieces yields an astronomically large space of possibilities. Yet the human brain navigates these steeplechase marathons in real time while running on roughly the power of a dim light bulb. The proposed resolution is the Virtual Associative Structures proposal, or VaST, first outlined by Yufik in 1998, which holds that understanding compresses vast combinatorial spaces into minute volumes imbued with meaning.

VaST rests on a thermodynamic intuition. The brain, like any living system, seeks to secure steady energy inflows from energy-carrying stimuli while minimizing caloric costs and dissipative losses. Three mechanisms carry out this minimization. Neuronal packets self-organize in associative networks when tightly associated subnets fold into functional units, separated from their surroundings by energy barriers much as surface tension holds a raindrop or a snowflake together. Relational neurons link these packets across indefinite stretches of the stimulus stream. And complexes of coordinated packets form models, hierarchical structures in which upper-level neurons facilitate coordination among lower-level packets. When packets and models match incoming streams, the combinatorial space collapses, reducing intractable multitudes of possible combinations to a handful of meaningful candidates. Crucially, the editors note, spurious alternatives are not rejected; they simply remain beyond the boundaries of awareness.

The editorial grounds these abstractions in experimental findings. Studies of motor control show attentive rotation of population vectors in the motor cortex, while imaging work on the frontopolar cortex points to neurons specialized for apprehending relations among structured elements. Research on packet-based communication in the cortex, in which neurons emit structured bursts of activity, lends further support to the idea that packets serve as core functional components of cognition. The editors also cite classic perceptual learning experiments in monkeys: when animals actively attended to a touched finger, the cortical territory representing that finger expanded to more than twice its usual size, but no such expansion occurred in monkeys touched identically while attending elsewhere. Experience, in other words, reshapes the energy landscape only when attention is engaged.

From these pieces the editors build a synthesis that connects several major theoretical frameworks. VaST aligns closely with Karl Friston’s Active Inference Theory, in which brains minimize variational free energy, the discrepancy between predicted and received inputs. It also complements Yann LeCun’s Energy-Based Models, which minimize an energy function measuring compatibility between inputs and outputs while inferring latent variables, such as the correct segmentation of a sentence into words, that belong to neither. The editors propose that the latent variables in energy-based models correspond to the latent relational variables in VaST, and that the two minimization criteria are mutually reinforcing: improved compatibility entails prediction error reduction and vice versa. Both, they argue, are subsumed by a single deeper principle, maximizing energy inflows while minimizing costs and dissipative losses.

To illustrate the difference between perception, learning, and understanding, the editors turn to chess. A chimpanzee reaching for fruit must coordinate perceptual packets but does not understand the task; it would never envision that shorter sticks demand higher boxes. A chess player solving a checkmate problem, by contrast, manipulates a virtual structure, a superposition of all possible king-and-rook-against-king games that guides the execution of any particular game. In such a model, thinking of moving one piece is inseparable from awareness of how that move constrains the others. Once formed, the model becomes a unit in a larger repertoire, and each new model eliminates degrees of freedom in its constituents while acquiring its own identity. The editors argue that submitting machine outputs to Turing-test-style evaluations amounts to playing peekaboo with oneself, since the underlying process is either knowable from a chess book or readable in the code; the real question is what biological mechanisms produce understanding and how they can be approximated in tractable procedures.

The editorial closes with conversation starters for the next volume, ranging from the Libet phenomenon, which the editors reinterpret as the energy cost of bringing packet contents into awareness rather than as evidence against free will, to a speculative alignment between Piaget’s co-instantaneous coordination and Roger Penrose’s quantum-entangled cytoskeletal states, and to the observation that the Latin cogitare, to think, literally means to shake together. On their account, propagating waves of cortical activity shake neuronal packets together the way vibration sorts mixed nuts, driving the system toward Ockham-optimal packings. Whatever the fate of these individual hypotheses, the editorial’s central challenge to the AI community is unambiguous: deep learning compresses automatisms, but only self-organizing structures that create meaning, operate on minuscule energy budgets, and can explain themselves will cross the threshold from systems that perform to systems that understand.

Subject of Research: Neurobiological mechanisms of human understanding and their implications for achieving artificial general intelligence

Article Title: Editorial: Neurobiological foundations of cognition and progress toward artificial general intelligence

Article References: Yufik, Y., Kozma, R., & Kroger, J. (2026). Editorial: Neurobiological foundations of cognition and progress toward artificial general intelligence. Frontiers in Systems Neuroscience, 20, Article 1967487. https://doi.org/10.3389/fnsys.2026.1967487

Image Credits: AI Generated

DOI: 10.3389/fnsys.2026.1967487

Keywords: artificial general intelligence, machine understanding, neuronal packets, self-organization, VaST, active inference, energy-based models, Hebbian assemblies, cognitive neuroscience, combinatorial search, Piaget, Frontiers in Systems Neuroscience

The intellectual roots of the field trace back to Alan Turing, who in 1950 suggested that intelligence is not merely abstract computation but something demonstrated through dynamic interaction between a body and its environment. The review frames that insight as a theoretical foundation: a physical carrier is a necessary prerequisite for intelligence to operate in the physical world. In practice, the authors define embodied manipulation as a closed-loop process that takes embodied cognition as its engine and a physical robot as its carrier. Unlike traditional robotics, where perception and execution are decoupled modules stitched together, an embodied agent must simultaneously interpret human instructions, fuse low-level sensory data from multimodal sensors, and generate strategies adapted to its own mechanical constraints, correcting itself in real time through feedback.

Mathematically, the authors formalize this as a partially observable Markov decision process, a framework that acknowledges a sobering truth: a real robot almost never knows the true state of the world. Instead, it must maintain a probability distribution over possible states, updated with every observation and action. Where classical robot learning relied on tidy, low-dimensional state vectors, embodied manipulation confronts megapixel images, open-vocabulary language commands, tactile readings, and force signals all at once, forming a vast composite state space. Policies are therefore parameterized as deep neural networks trained to map this torrent of perception directly onto motor commands, maximizing expected cumulative reward over time.

What would an ideal embodied manipulator look like? The review identifies six hallmarks. It must achieve consistent multimodal perception, aligning vision, language, and haptics in a unified semantic space. It needs comprehensive multimodal understanding, the way large language models absorb web-scale knowledge. It must generalize across tasks and adapt zero-shot to unseen objects and scenes. It requires spatial intelligence, reasoning in three dimensions about geometry and the temporal consequences of its own actions. It needs basic physical commonsense, an intuitive grasp that objects fall, slide, and deform. And ultimately it must evolve, self-planning and self-correcting as environments shift.

Two competing technical philosophies currently dominate the field. The data-driven route treats manipulation as imitation at scale: collect enormous datasets of expert demonstrations, then train end-to-end vision-language-action models, or VLAs, that map what the robot sees and hears directly into what it does. Landmark systems such as RT-1, RT-2, PaLM-E, and the open-source OpenVLA exemplify this approach, and datasets like X-Embodiment, which aggregates a million-scale corpus spanning 22 robot morphologies and 527 skills, provide the fuel. Diffusion models, borrowed from image generation, have proven surprisingly effective here: the Diffusion Policy framework refines random noise into smooth, continuous action sequences, and successors like RDT-1B extend the idea to bimanual, cross-dataset learning.

The model-driven camp counters that data alone cannot carry robots through the open world. Researchers in this tradition build world models, internal simulations that let a robot imagine the consequences of its actions before committing to them. DayDreamer demonstrated that real robots can learn manipulation skills entirely through such imagined rollouts, updating their world model from live experience. Others inject the vast knowledge of multimodal large language models into the control loop: SayCan uses a language model to score which atomic skills actually serve a spoken instruction, while systems like ReKep generate spatial constraint functions from keypoints to solve manipulation trajectories without any task-specific training. Reinforcement learning adds a third pillar, fine-tuning imitation-trained policies through real-world trial and error, with frameworks like ConRFT balancing sample efficiency against safe execution.

Behind both paradigms lies a rapidly maturing infrastructure. Low-cost teleoperation rigs such as ALOHA and the handheld UMI gripper have slashed the price of collecting dexterous demonstration data, while Stanford’s HumanPlus tracks whole-body human motion to teach humanoids by shadowing. On the simulation side, GPU-accelerated engines like Isaac Gym and MuJoCo allow thousands of training environments to run in parallel, and newer frameworks such as Genesis and ManiSkill3 push toward photorealistic, four-dimensional worlds where policies can be trained before ever touching hardware. Generative techniques are now synthesizing data outright: RoboGen proposes its own skills and builds simulation scenes to practice them, while DemoGen mathematically replans existing trajectories to multiply datasets without a single new demonstration.

Yet the review is candid about how far robots remain from human dexterity. Vision-based policies depend on dense camera arrays that turn real workplaces into idealized laboratories, and a single modality can be catastrophically fooled; a robot wiping a table cannot feel whether it is pressing down or merely sliding the cloth. Contact-rich tasks like turning keys or opening valves expose the weakness of position-servo control that ignores force dynamics, and force sensors remain too expensive for mass deployment. Generalization is fragile, with minor lighting changes able to collapse performance, and full autonomy remains elusive. There is also a computational squeeze: the large models that perform best are precisely the ones too heavy for the edge processors a robot can physically carry, motivating compact architectures like SmolVLA, which achieves tenfold faster inference, and 1-bit compressed models like BitVLA.

The authors close with a plea that models and data should be treated as symbiotic rather than rival paradigms: data supplies the empirical knowledge that covers the world’s unpredictability, while models supply the interpretable framework that makes that knowledge usable. They also flag safety and ethics as unsolved frontiers, from concealed sim-to-real risks and adversarial attacks on embodied decision-makers to questions of job displacement and liability when autonomous machines cause harm. If the field’s trajectory holds, the milestone that matters may not be another benchmark score but the first robot that folds laundry in a stranger’s home as confidently as in its training lab, a quiet proof that intelligence, as Turing suspected, was always something a body does.

Subject of Research: Embodied intelligence for robotic manipulation, covering data-driven and model-driven approaches, their supporting infrastructure, and open challenges

Article Title: Embodied intelligence for robot manipulation: development and challenges

Article References: Song, H., Wang, L., Qiao, X., Chen, Y., Sun, D., & Sun, Z. (2025). Embodied intelligence for robot manipulation: development and challenges. Vicinagearth, 2(1), Article 8. https://doi.org/10.1007/s44336-025-00020-1

Image Credits: AI Generated

DOI: 10.1007/s44336-025-00020-1

Keywords: embodied intelligence, robot manipulation, vision-language-action models, world models, reinforcement learning, imitation learning, diffusion policy, multimodal perception, sim-to-real transfer, artificial general intelligence, humanoid robots, robot datasets

The review, conducted according to PRISMA guidelines and operationalised through an explicit PICOS framework, searched five databases, PubMed, Scopus, Web of Science, Dialnet and PsycINFO, for publications from 2013 to 2023. Two reviewers independently screened records at the title and abstract stages and again at full text, resolving disagreements through discussion until consensus was reached. Because AGI does not yet exist as a clinically deployable technology, the eligible literature is overwhelmingly theoretical, ethical or conceptual rather than experimental. The authors therefore adopted an explicit hierarchy-of-evidence framework that distinguishes empirical studies, secondary reviews, ethical and normative analyses, and theoretical or speculative work, weighting the synthesis accordingly. Of 72 records initially identified, 45 studies met the inclusion criteria and were synthesised thematically using NVivo software, with methodological quality appraised through author-developed criteria aligned with the SANRA scale for narrative reviews and the AACODS checklist for non-empirical sources.

The first major theme concerns diagnosis and treatment. AGI-adjacent systems, particularly large language models and multimodal architectures already deployed under the umbrella of narrow AI, have demonstrated genuine potential to support mental health assessment by integrating heterogeneous data streams: clinical notes, speech patterns, facial expression analysis, behavioural telemetry and self-reported symptoms. Conversational interfaces accessible around the clock and at negligible marginal cost could reduce stigma by offering an anonymous, judgement-free first point of contact for people reluctant to seek human help. For underserved regions where the ratio of mental health professionals to population is dangerously low, the prospect of near-universal access to competent psychological support is the review’s most compelling promised benefit.

Yet the same technical capabilities carry the mirror-image risks. The literature consistently warns, at the level of ethical and normative analysis, that automated emotional crisis management is fraught with danger: a system that handles a suicidal disclosure with a plausible but subtly wrong response could produce diagnostic error, false reassurance or lethal delay in escalation to human clinicians. The review stresses that such risks, documented for narrow AI in embryonic form, would be substantially amplified in any future AGI system, and that presentations involving psychosis and other severe mental illnesses pose particular challenges, because the conversational fluency of advanced models can mask a fundamental lack of clinical understanding and may even reinforce delusional content.

The second theme addresses governance. The review identifies emerging regulatory instruments, mandatory algorithmic audits, transnational data protection policies and clearer allocation of medico-legal accountability, as necessary preconditions for any safe integration of AGI into mental health care. Algorithmic audits would function as a technical safeguard, probing systems for bias, brittleness and unsafe behaviour before and during deployment, while data protection frameworks would govern the extraordinarily sensitive information that mental health applications inevitably collect. Accountability remains perhaps the thorniest question: when an autonomous-sounding system contributes to a harmful clinical outcome, the chain of responsibility spanning developers, deployers, clinicians and institutions must be legally defined in advance rather than litigated after tragedy.

The third and most philosophically provocative theme concerns human identity. Drawing on established work-identity and AI ethics literatures, the authors advance theoretical constructs such as ‘existential delaborisation’, the psychological dislocation that could follow if AGI displaces human beings from work that confers meaning, purpose and social standing. They also describe the algorithmic reconfiguration of biographical narratives: the possibility that AGI systems, by curating recommendations, memories and self-reflections, could subtly rewrite the stories people tell about themselves. These phenomena are presented explicitly as theoretical frameworks rather than empirically documented outcomes, but their implications intersect with well-established knowledge that unemployment, loss of occupational identity and social dislocation are robust predictors of depression, anxiety and substance misuse. Mass technological unemployment, should AGI fulfil the most aggressive projections, would therefore constitute not merely an economic problem but a population-scale mental health exposure.

The review also highlights the risk that biases embedded in training data could be perpetuated and amplified within therapeutic interventions. If the corpora on which AGI systems are trained under-represent certain cultures, languages, genders or diagnostic populations, the resulting systems may misinterpret idioms of distress, apply inappropriate diagnostic framings or deliver interventions calibrated to populations unlike the user. Because mental health care depends on trust, cultural sensitivity and nuanced understanding of individual context, such biases are not simply statistical inaccuracies but potential sources of harm and of widening inequity between well-served and marginalised communities.

What distinguishes this analysis from much of the public AGI debate is its disciplined separation of evidence from speculation. The empirical claims about diagnostic support and stigma reduction rest on studies of currently deployed narrow AI, while the warnings about existential and identity-level harms are clearly flagged as normative and theoretical extrapolation. This evidentiary transparency matters practically as well as intellectually: policymakers and clinicians reading the review can see precisely which conclusions are grounded in observed outcomes and which are reasoned projections, allowing proportionate rather than panicked or complacent responses. The authors argue that this evidence-differentiated approach should itself become a standard feature of AI-in-health governance, ensuring that regulatory urgency is matched to evidentiary strength.

The paper’s conclusion is deliberately double-edged. AGI integration in mental health offers the genuine possibility of democratised psychological care, extending support to billions of people who currently have none, while simultaneously posing clinical-safety and existential challenges that demand urgent ethical governance. Success, the authors contend, depends on regulatory frameworks that prioritise transparency, equity, accountability and privacy, and, crucially, on the preservation of authentic human connection as the irreplaceable core of mental health care. Machines may eventually match or exceed human clinicians in informational processing, but the therapeutic alliance, the felt experience of being understood by another conscious being, remains, on the current evidence, a uniquely human asset. Whether AGI arrives by 2030 or decades later, the review’s central message is that the window for building the ethical, legal and clinical infrastructure to receive it safely is open now, and it should not be allowed to close.

Subject of Research: The impact of artificial general intelligence on mental health

Article Title: The impact of artificial general intelligence on mental health (a systematic narrative review)

Article References: Moya, C. C., & Rodríguez, E. Q. (2026). The impact of artificial general intelligence on mental health (a systematic narrative review). Discover Mental Health. https://doi.org/10.1007/s44192-026-00589-z

Image Credits: AI Generated

DOI: 10.1007/s44192-026-00589-z

Keywords: artificial general intelligence, mental health, systematic narrative review, AI ethics, algorithmic bias, digital rights, AI-driven psychotherapy, human identity, evidence hierarchy, medico-legal accountability, algorithmic audits, existential delaborisation