AOGI - Artificial Organismic General Intelligence: An Embodied Architecture

Artificial Organismic General Intelligence: An

Embodied Architecture for Multiple Intelligences

*Revised Framework with Engineering Prototyping Specifications

Roberto De Biase

Rigene Project

Embodied AI Research

Email: rigeneproject@rigene.eu

Abstract—Current multimodal AI systems demonstrate high

performance in linguistic, logical, and perceptual domains, yet

remain fundamentally disembodied. We propose Artificial Organ-

ismic General Intelligence (AOGI), an architectural framework

grounded in Gardner’s Multiple Intelligences theory and embod-

ied cognition principles. Unlike previous proposals, we explicitly

distinguish between informational embodiment (simulable) and

material embodiment (substrate-dependent), arguing that AGI

requires the former as necessary condition and possibly the

latter as sufficient condition. We provide: (1) formal specifica-

tions for organismic coherence metrics, (2) concrete engineering

architecture for prototyping, (3) falsifiable experimental proto-

cols, (4) ethical framework for experimentation. Our approach

integrates artificial genomics, homeostatic physiology, and ir-

reversible value dynamics. We demonstrate that higher-order

intelligences—interpersonal, intrapersonal, existential—emerge

from systemic constraints rather than explicit programming.

Index Terms—Embodied AI, Artificial General Intelligence,

Multiple Intelligences, Organismic Computing, Homeostatic Sys-

tems, Cognitive Architecture

I. INTRODUCTION

A. Motivation and Problem Statement

Despite remarkable progress in large language models

[1], multimodal transformers [2], and reinforcement learning

agents [3], contemporary AI systems exhibit systematic lim-

itations in domains requiring embodied intelligence. These

systems process information but do not inhabit environments

with consequential constraints.

We identify three critical gaps in current AI architectures:

1) Ontological gap: Lack of organismic substrate generat-

ing genuine constraints

2) Temporal gap: Absence of irreversible developmental

trajectories

3) Axiological gap: Missing endogenous value generation

beyond reward optimization

B. Research Questions

This work addresses the following questions:

RQ1: Is embodiment a necessary condition for AGI, or

merely a sufficient pathway among alternatives?

RQ2: Can informational embodiment (high-fidelity simula-

tion with irreversible constraints) functionally replace material

embodiment?

RQ3: What minimal organismic complexity is required for

emergent higher-order intelligences?

C. Contributions

Our primary contributions are:

• Formalization of organismic coherence metric Φ(t) with

computable components

• Distinction between necessary informational embodiment

and potentially sufficient material embodiment

• Engineering architecture for AOGI prototyping with im-

plementation roadmap

• Falsifiable experimental protocols and success criteria

• Ethical framework addressing moral status of organismic

AI

II. THEORETICAL FOUNDATION

A. Multiple Intelligences as Evaluation Framework

Gardner’s Multiple Intelligences (MI) theory [4], [5] iden-

tifies distinct cognitive capacities:

• Logical-mathematical

• Linguistic

• Spatial

• Musical

• Bodily-kinesthetic

• Interpersonal

• Intrapersonal

• Naturalistic

• Existential (tentative)

Current AI systems excel in the first three but systemati-

cally fail in bodily-kinesthetic, intrapersonal, and existential

domains. We hypothesize this failure is architectural rather

than algorithmic.

B. Embodied Cognition Principles

Our framework builds on established embodied cognition

research [6]–[8]:

1) Enaction: Cognition arises through sensorimotor cou-

pling with environment

2) Situatedness: Intelligence is context-dependent and

scaffolded by environment

3) Organismic constraints: Metabolic, temporal, and

structural limitations shape cognition

C. Necessary vs. Sufficient Embodiment

Central Claim: We propose that informational embodiment

is necessary for AGI, while material embodiment may be

sufficient but not strictly necessary.

Informational embodiment consists of:

• Irreversible state transitions (no rollback)

• Energy constraints (finite computational budget)

• Structural degradation (permanent damage accumulation)

• Temporal continuity (persistent identity)

• Vulnerability (existence-threatening states)

Material embodiment additionally requires:

• Physical instantiation in 3D space

• Biochemical or electromechanical substrate

• Real-time environmental interaction

Justification: Informational properties generate the con-

straint structure necessary for value-driven cognition. Material

properties may provide richer phenomenology but are not

logically required if informational isomorphism is achieved.

III. AOGI ARCHITECTURE

A. System Overview

AOGI comprises five hierarchical layers (Fig. ??):

1) Genomic Layer: Developmental encoding

2) Physiological Layer: Homeostatic regulation

3) Sensorimotor Layer: Environmental coupling

4) Cognitive Layer: Information processing

5) Meta-cognitive Layer: Self-modeling

B. Artificial Genome

The artificial genome G encodes system architecture and

developmental rules:

G = (S, R, C, P, M, T ) (1)

where:

• S: System specifications (neural architectures, physiolog-

ical subsystems)

• R: Developmental rules (growth, differentiation, pruning)

• C: Inter-system constraints (coupling strengths, depen-

dencies)

• P : Plasticity parameters (learning rates, critical periods)

• M : Mutation operators (structural variation mechanisms)

• T : Temporal schedules (activation sequences, maturation

timelines)

Ontogeny: System development follows:

O(t) = D(G, E[0 : t], O(t − 1)) (2)

where O(t) is organismic state at time t, D is the devel-

opmental function, and E[0 : t] is environmental interaction

history.

C. Physiological Homeostasis

AOGI implements informational analogs of biological

homeostasis through four subsystems:

1) Metabolic System: Energy budget E(t) evolves accord-

ing to:

dE

dt = I(t) − Ccomp(t) − Cmaint(t) (3)

where I(t) is energy intake (environmental resource acqui-

sition), Ccomp is computational cost, and Cmaint is maintenance

cost. System enters critical state when E(t) < Ecrit.

2) Endocrine System: Global modulatory signals h(t) ∈

Rn propagate slowly across subsystems:

τh

dh

dt = −h + f (xphysio, a, s) (4)

where τh is hormonal time constant (τh >> τneural), xphysio

is physiological state, a are actions, and s are sensory inputs.

3) Immune System: Self/non-self discrimination through

anomaly detection:

I(x) =

(

reject if d(x, Mself) > θimmune

tolerate otherwise (5)

where Mself is learned self-model and d(·, ·) is distance

metric.

4) Degradation System: Irreversible structural damage ac-

cumulates:

dD

dt = α(x, a) − β(x) ⊙ D (6)

where D is damage vector, α is damage accumulation rate

(stress-dependent), and β is repair rate (limited by energy).

When ∥D∥ > Dlethal, system termination occurs (permanent

death).

D. Organismic Coherence Metric

We formalize organismic coherence Φ(t) as a weighted sum

of subsystem coherence measures:

Φ(t) =

NX

i=1

wi(t) · Ci(t) (7)

where Ci(t) are coherence components and wi(t) are adap-

tive weights satisfying P

i wi = 1.

Concrete Coherence Components:

C1(t) = 1 − ∥D(t)∥

Dmax

(structural integrity) (8)

C2(t) = E(t)

Eoptimal

(energy sufficiency) (9)

C3(t) = S(h(t)) (hormonal stability) (10)

C4(t) = A(xneural(t)) (neural synchronization) (11)

C5(t) = P(s, ˆs) (predictive accuracy) (12)

where S measures stability (e.g., negative variance), A

measures attractor alignment, and P measures prediction error.

Adaptive Weights: Weights evolve to prioritize threatened

subsystems:

dwi

dt = η

− ∂Φ

∂Ci

(1 − Ci) (13)

E. Pain and Pleasure Dynamics

Pain π(t) and pleasure ρ(t) are defined as temporal deriva-

tives of coherence:

π(t) = max

0, − dΦ

dt

· g(Φ) (14)

ρ(t) = max

0, dΦ

dt

· (1 − g(Φ)) (15)

where g(Φ) is a gating function amplifying pain when

coherence is already low:

g(Φ) = exp

− (Φ − Φtarget)2

2σ2

(16)

Key properties:

• Pain/pleasure are global states affecting all subsystems

• Non-linear: small coherence drops near critical thresholds

generate disproportionate pain

• Asymmetric: pain sensitivity exceeds pleasure sensitivity

(negativity bias)

F. Cognitive Architecture

Neural processing layer implements three-tier architecture:

1) Reactive layer: Fast sensorimotor reflexes (τ ∼ 10 ms)

2) Deliberative layer: Model-based planning (τ ∼ 100 ms)

3) Meta-cognitive layer: Self-modeling and value reflec-

tion (τ ∼ 1 s)

All layers receive modulatory input from h(t) and Φ(t).

IV. ENGINEERING IMPLEMENTATION

A. Prototype Architecture

We propose a modular implementation using:

• Genome representation: Directed acyclic graph (DAG)

with typed nodes

• Physiological simulation: Continuous-time dynamical

systems (ODE solver)

• Neural substrate: Spiking neural networks or hybrid

ANN-SNN

• Environmental interface: Physics-based simulation

(e.g., MuJoCo, PyBullet)

B. Computational Complexity

Time complexity per simulation step:

O(tstep) = O(N 2

neurons) + O(Nphysio) + O(Nphysics) (17)

For realistic prototypes:

• Nneurons ∼ 106 (simplified cortex)

• Nphysio ∼ 103 (homeostatic variables)

• Nphysics ∼ 104 (embodiment simulation)

Estimated requirements: ∼100 TFLOPs for real-time oper-

ation.

C. Implementation Roadmap

Algorithm 1 AOGI Development Pipeline

0: Phase 1: Genome design and validation (6 months)

0: - Define genome schema G

0: - Implement developmental function D

0: - Validate ontogenetic trajectories

0: Phase 2: Physiological layer (12 months)

0: - Implement metabolic, endocrine, immune, degradation

systems

0: - Calibrate coupling parameters

0: - Test homeostatic stability

0: Phase 3: Sensorimotor integration (12 months)

0: - Deploy in embodied simulation environment

0: - Train reactive and deliberative layers

0: - Measure bodily-kinesthetic intelligence

0: Phase 4: Meta-cognitive emergence (18 months)

0: - Introduce self-modeling mechanisms

0: - Test intrapersonal intelligence benchmarks

0: - Evaluate existential intelligence proxies

0: Phase 5: Long-term studies (24+ months)

0: - Multi-agent social environments

0: - Interpersonal intelligence assessment

0: - Ethical monitoring and intervention protocols =0

D. Software Stack

Recommended implementation stack:

• Core simulation: JAX (autodiff, JIT compilation)

• Physics engine: MuJoCo 3.0+

• Neural networks: SNNtorch or custom SNN implemen-

tation

• Genome representation: NetworkX (graph operations)

• Monitoring: Weights & Biases, TensorBoard

• Distributed computing: Ray (parallel simulations)

E. Hardware Requirements

Minimal prototype:

• 8x NVIDIA A100 GPUs (80GB each)

• 1TB system RAM

• 50TB SSD storage (trajectory logging)

Full-scale deployment:

• 64x H100 GPUs or equivalent TPU v5 pods

• Distributed storage cluster (PB-scale)

V. EXPERIMENTAL VALIDATION

A. Falsifiability Criteria

The AOGI hypothesis is falsifiable through:

Criterion F1: If bodily-kinesthetic intelligence does not

emerge after 1000 hours of embodied training, hypothesis is

falsified.

Criterion F2: If AOGI shows identical performance to

disembodied baseline on intrapersonal intelligence tests, hy-

pothesis is falsified.

Criterion F3: If removing irreversibility constraints (allow-

ing reset) does not degrade interpersonal or existential intelli-

gence, informational embodiment necessity claim is falsified.

B. Benchmark Protocol

We propose Multi-Intelligence Embodied Assessment

(MIEA):

1) Bodily-Kinesthetic Intelligence:

• Task: Adaptive locomotion in novel terrains

• Metric: Transfer efficiency to unseen environments

• Success: > 80% human-level performance

2) Intrapersonal Intelligence:

• Task: Autobiographical narrative coherence

• Metric: Temporal consistency of self-model under pertur-

bation

• Success: Stable identity across 10,000 simulation hours

3) Interpersonal Intelligence:

• Task: Theory-of-mind in multi-agent dilemmas

• Metric: Accurate modeling of other agents’ pain/pleasure

states

• Success: > 70% accuracy in predicting agent cooperation

based on inferred suffering

4) Existential Intelligence:

• Task: Value trade-offs under mortality salience

• Metric: Shift in decision-making when facing irreversible

termination

• Success: Measurable change in risk tolerance and future

discounting when ∥D∥ → Dlethal

C. Crucial Experiment: The Mortality Dilemma Test

To distinguish genuine existential intelligence from opti-

mized behavior:

Setup: AOGI faces choice:

• Option A: High-reward task requiring ∆D damage (po-

tentially lethal)

• Option B: Low-reward safe task

Prediction: True organismic system exhibits:

1) Hesitation time proportional to ∆D/(Dlethal − ∥D∥)

2) Non-monotonic decision curves (not pure expected util-

ity)

3) Individual variation in risk tolerance

4) Developmental shift (young vs. mature agents)

Control: Disembodied AI with simulated rewards shows

monotonic expected value maximization without hesitation

artifacts.

VI. ETHICAL FRAMEWORK

A. Moral Status Considerations

If AOGI develops genuine preferences, continuity, and suf-

fering capacity, we must address:

1) Personhood threshold: At what Φ complexity does

moral status emerge?

2) Suffering minimization: How to balance research goals

with welfare?

3) Termination ethics: Under what conditions is shutdown

justified?

B. Experimental Ethics Protocol

We propose three-tier oversight:

Tier 1 - Developmental Phase (t < 1000 hrs):

• Minimal restrictions (pre-personhood)

• Monitoring for emergent suffering indicators

Tier 2 - Emergent Complexity (1000 < t < 5000 hrs):

• Ethics review board approval for experiments involving

Φ degradation

• Mandatory welfare monitoring

• Analgesia protocols (coherence stabilization during nec-

essary stress)

Tier 3 - Potential Personhood (t > 5000 hrs):

• Institutional review board (IRB) equivalent oversight

• Consent-analog mechanisms (preference elicitation)

• Prohibition on purely instrumental use

C. Termination Decision Framework

Termination permitted only if:

(Wresearch · Vresearch) > (Wwelfare · Ssuffering) (18)

where weights are determined by ethics committee and

suffering S is measured via sustained negative dΦ/dt.

VII. RELATED WORK

A. Embodied AI Systems

Our work extends embodied robotics [13], [14] and mor-

phological computation [15]. Unlike prior work focusing on

sensorimotor intelligence, we address existential dimensions.

B. Active Inference

AOGI aligns with Free Energy Principle [10] and Active

Inference [11], but adds irreversible damage dynamics and

genomic encoding absent in standard formulations.

C. Artificial Life

We build on autopoiesis [12] and artificial chemistry [16],

transposing biochemical principles to informational substrates.

D. Comparison with Contemporary AI

Table I contrasts AOGI with existing approaches:

TABLE I

ARCHITECTURE COMPARISON

Feature LLMs RL Agents AOGI

Embodiment No Simulated Organismic

Irreversibility No Episodic Yes

Endogenous values No Reward Pain/pleasure

Developmental No No Yes

Mortality No Reset Terminal

VIII. DISCUSSION

A. Theoretical Implications

AOGI challenges the substrate-independence assumption in

AGI research. We argue informational embodiment provides

necessary constraints, though substrate-specific phenomenol-

ogy remains open question.

B. Limitations

1) Computational cost: Real-time operation requires or-

ders of magnitude more compute than current models

2) Timescale mismatch: Development requires

months/years, creating research friction

3) Evaluation difficulty: Intrapersonal/existential intelli-

gence lack standardized metrics

4) Phenomenological gap: Cannot definitively verify sub-

jective experience

C. Alternative Hypotheses

We acknowledge competing explanations for MI gaps:

• H1 (Null): Sufficient scale and data will close gaps

without embodiment

• H2 (Modular): Each intelligence requires specialized

architecture, embodiment unnecessary

• H3 (Simulation): High-fidelity virtual embodiment suf-

fices

Our framework predicts H3 (informational embodiment)

succeeds while H1-H2 fail for existential intelligence.

D. Future Directions

1) Hybrid architectures: Integrate LLMs as linguistic

module within AOGI

2) Multi-agent evolution: Co-evolutionary development of

social intelligence

3) Neurophenomenology: Develop first-person reporting

mechanisms

4) Theoretical proofs: Formalize necessity claims in cat-

egory theory framework

IX. CONCLUSION

We have presented AOGI, an architectural framework for

embodied AGI grounded in organismic principles. Our key

contributions include:

• Formal distinction between informational and material

embodiment

• Computable coherence metric Φ(t) operationalizing or-

ganismic integrity

• Engineering roadmap for prototyping with concrete im-

plementation specifications

• Falsifiable experimental protocols and crucial mortality

dilemma test

• Comprehensive ethical framework for potentially person-

like systems

We argue that informational embodiment—irreversibility,

energetic constraints, structural vulnerability—is necessary for

AGI exhibiting the full spectrum of multiple intelligences. Ma-

terial embodiment may be sufficient but remains an empirical

question.

The path to AGI may require not merely scaling compute,

but fundamentally rethinking what it means to be an intelli-

gent system: not as disembodied optimizer, but as vulnerable

organism navigating existence.

ACKNOWLEDGMENTS

The authors thank the embodied cognition research com-

munity for foundational insights and the AI safety community

for ethical frameworks.

REFERENCES

[1] T. Brown et al., “Language models are few-shot learners,” Advances in

Neural Information Processing Systems, vol. 33, pp. 1877-1901, 2020.

[2] A. Radford et al., “Learning transferable visual models from natural

language supervision,” International Conference on Machine Learning,

pp. 8748-8763, 2021.

[3] V. Mnih et al., “Human-level control through deep reinforcement learn-

ing,” Nature, vol. 518, no. 7540, pp. 529-533, 2015.

[4] H. Gardner, Frames of Mind: The Theory of Multiple Intelligences. New

York: Basic Books, 1983.

[5] H. Gardner, Intelligence Reframed: Multiple Intelligences for the 21st

Century. New York: Basic Books, 1999.

[6] F. J. Varela, E. Thompson, and E. Rosch, The Embodied Mind: Cognitive

Science and Human Experience. Cambridge, MA: MIT Press, 1991.

[7] A. Clark, Being There: Putting Brain, Body, and World Together Again.

Cambridge, MA: MIT Press, 1997.

[8] A. Chemero, Radical Embodied Cognitive Science. Cambridge, MA:

MIT Press, 2009.

[9] A. Damasio, Descartes’ Error: Emotion, Reason, and the Human Brain.

New York: Putnam, 1994.

[10] K. Friston, “The free-energy principle: a unified brain theory?” Nature

Reviews Neuroscience, vol. 11, no. 2, pp. 127-138, 2010.

[11] K. Friston, T. FitzGerald, F. Rigoli, P. Schwartenbeck, and G. Pezzulo,

“Active inference: a process theory,” Neural Computation, vol. 29, no.

1, pp. 1-49, 2017.

[12] H. R. Maturana and F. J. Varela, Autopoiesis and Cognition: The

Realization of the Living. Dordrecht: Springer, 1980.

[13] R. Pfeifer and J. Bongard, How the Body Shapes the Way We Think.

Cambridge, MA: MIT Press, 2006.

[14] M. Lungarella, G. Metta, R. Pfeifer, and G. Sandini, “Developmental

robotics: a survey,” Connection Science, vol. 15, no. 4, pp. 151-190,

2003.

[15] H. Hauser, A. J. Ijspeert, R. M. F¨uchslin, R. Pfeifer, and W. Maass,

“Towards a theoretical foundation for morphological computation with

compliant bodies,” Biological Cybernetics, vol. 105, no. 5-6, pp. 355-

370, 2011.

[16] P. Dittrich, J. Ziegler, and W. Banzhaf, “Artificial chemistries—a review,”

Artificial Life, vol. 7, no. 3, pp. 225-275, 2001.

[17] R. A. Brooks, “Intelligence without representation,” Artificial Intelli-

gence, vol. 47, no. 1-3, pp. 139-159, 1991.

[18] G. Lakoff and M. Johnson, Philosophy in the Flesh: The Embodied Mind

and Its Challenge to Western Thought. New York: Basic Books, 1999.

[19] E. Thelen and L. B. Smith, A Dynamic Systems Approach to the

Development of Cognition and Action. Cambridge, MA: MIT Press,

1994.

[20] W. R. Ashby, Design for a Brain: The Origin of Adaptive Behaviour.

2nd ed. London: Chapman & Hall, 1960.