Bio-Intelligence as Engineering

Substrates, interfaces, experiments — a transdisciplinary landscape survey with an experimental programme

Preliminary note on the position of this work

Three axes are already occupied in this strand:

This work reopens neither the consciousness question nor the organisation question. It treats both as boundary conditions and asks the engineering question: parameters, bottlenecks, cost, legal status, access — and from these an experimental programme that can be started with existing workshop infrastructure.

Evidence status. Every factual claim carries a marker: [P] peer-reviewed or primary source verified by DOI, [S] secondary source or vendor statement, [O] open, a claim only, not independently replicated. Vendor statements on energy efficiency and performance are [S] or [O] throughout, because the vendors are also stakeholders.

Exclusions. Military, weapons and defence applications are excluded across all strands. In this field that is not a formal clause: several of the referenced works on cyborg insects and biohybrid sensing are funded through naval research grants. The exclusion therefore bites substantively and is applied to specific approaches in section 7.

Availability. This work runs on the platform. Tools, worked experimental documentation, build documentation and teaching modules live at syntagma-science.com, where they are maintained, versioned and developed further. The text itself — landscape survey, bottleneck analysis, ρ/δ analysis, guardrails, open questions and all evidence — stays free under CC BY 4.0, fully citable, and is deposited as a reference edition pointing to the platform.

Since 2026-10-06 the order is reversed: not an archive holding with an offer attached, but a platform with the finding deposited freely alongside. Nothing about what is open changes — the finding stands complete in the free part and stays there.

1 A three-level distinction

The umbrella term “bionics” covers three technically unconnected approaches. Conflating them is the most common source of misplanning in this field.

Fig. 1 — Three levels of transfer from nature. Separate competences, no inheritance between levels.

The levels do not inherit from each other. B1 demands materials and manufacturing competence, B2 tissue culture plus control engineering, B3 electrophysiology plus coding theory. Mastery of B1 yields nothing for B3 except manufacturing capacity for peripherals — which, per section 9, is the real point of connection.

2 Substrate taxonomy

Six substrate classes with the parameters that decide feasibility. Cost figures are orders of magnitude for a single build, not quotations.

Fig. 2 — Substrate map: maturity against access barrier, colour encodes ethical precautionary load (author’s classification, not a measurement).

The column “animal-free achievable” is not standard in the literature and is introduced here; its justification is approach N1 in section 8.

3 State of the art

3.1 Neuronal wetware

Platforms. Two commercial routes exist.

Demonstrated computation. The state of the art is more modest than market communication suggests, but not nil.

Maturation and vascularisation are the two physiological bottlenecks. Without vascular supply organoids form a necrotic core, which limits size and maturation [P]. Vascularisation approaches:

Timeline: a paper dated 19 Aug 2026 shows that under suitable conditions organoids track a multi-year developmental clock and express features of later brain-development stages [P] (Faravelli, Antón-Bolaños et al., Nature, doi:10.1038/s41586-026-10877-x). This shifts the maturation-limit question without removing it.

3.2 Non-neuronal living substrates

Fungal memristors. Shiitake mycelium (Lentinula edodes), dehydrated and wired, shows pinched hysteresis, i.e. memristive behaviour. Reported: 95 % memristive fidelity at 5 Vpp/10 Hz and 90 ± 1 % up to 5.85 kHz, with samples remaining functional after dehydration [P] (LaRocco et al., PLOS One, doi:10.1371/journal.pone.0328965). Radiation resistance is mentioned there only in the discussion as an argument for aerospace use; no primary study on it has been identified [O].

Fungal logic. Boolean and electrical gates in fungal colonies have been documented since 2020/2021 [P] (Adamatzky, Beasley et al.), largely from one research group and not broadly reproduced independently; reading fungal spiking as “language” is contested in the literature and is not adopted here [P]. Switching times are in the seconds range; these systems are not intended to replace digital logic but to fuse sensing and computation in the substrate itself.

Slime mould. Physarum polycephalum produces approximate solutions to network optimisation problems; sensing and computing devices based on Physarum come predominantly from one laboratory (Unconventional Computing Laboratory, UWE Bristol; EU project PhyChip); no reliable device count is documented [O]. The link to kognition-ohne-zentrum (≥0393) is substantive, not merely thematic: the substrate computes without a central instance.

Assessment. This class is inferior to neuronal wetware in maturity and performance, and clearly superior in accessibility and ethical load. It is the only branch that can be run without approval, without a biosafety level, and without animal material.

3.3 Biohybrid robotics

Actuation. Skeletal muscle delivers precise directed contraction, cardiac muscle delivers autonomous rhythm. Both are built into soft-robotic skeletons [P]. Development lines:

Documented limit: fatigue. At four contractions per second over 30 minutes performance drops measurably; at one contraction per second it does not [S]. The final version of the tendon paper reports fatigue tests over more than 7000 cycles; the four-versus-one comparison is not found there [O]. The mechanism is open [O].

Cyborg insects. Terrestrial cyborg insects have been documented since 1997; steering is by stimulation at the antennal bases, readout by probes implanted in the antennal lobe. Reported values: >85 % steering success, 92.8 % static odour recognition, 84.8 % during locomotion [P] (Huang et al., Adv. Mater. Technol. 2025, doi:10.1002/admt.202501158; funding source of this paper not established). The Sniffbot (2026) mounts a locust antenna plus miniaturised electrophysiology on a vehicle and localises odours in still air faster than metal-oxide sensors [P] (bioRxiv 10.64898/2026.02.17.706375).

For context: a substantial part of this line is funded through naval research grants and explicitly aimed at explosives detection. Section 7 covers the resulting exclusion decision.

3.4 Nanobiology

Practical access is comparatively good here: oligonucleotides can be ordered commercially, design tools (caDNAno, oxDNA) are open, but validation requires AFM or gel electrophoresis.

3.5 The semiconductor side as a yardstick

Without this comparison the wetware efficiency claim cannot be checked.

Six to eight orders of magnitude, depending on the reference value, separate the biological synapse (fJ) from the organic artificial neuron (nJ); CMOS spiking neurons, by contrast, already operate in the femtojoule range. The testable gap therefore runs between device classes, not between “biological” and “technical”. The comparison with a data centre additionally conflates substrate level and system level.

The three values measure different events: the organic figure is per neuron spike of a single device, the biological figure per synaptic event (not per action potential; for a whole action potential the cortical energy budget gives about 7×10⁸ ATP per neuron [P]), the CMOS figure per spike of an optimised isolated circuit at very low supply voltage. Chip level is higher (picojoules per synaptic operation, in order of magnitude). A breakdown of life support against computation energy for organoids has not been published; the system power figures in circulation are vendor claims [S].

4 Bottlenecks

Bottleneck EP4 is the structurally hardest. As long as no robust learning rule exists, every wetware application is either a reservoir with a digital readout classifier — in which case the intelligence sits in the classifier — or a demonstration of stimulus-response differences.

5 Audit of common claims

The last row exemplifies a pattern: the same percentage means characteristic-curve fidelity, classification quality or control success rate depending on context. Citations must carry the referent.

6 ρ/δ analysis (instance of FORSCH-0083)

Who carries the risk (ρ), who holds disposal (δ)?

A3 is where the asymmetry becomes technical rather than rhetorical. The occupational-safety grounds for BSL-2 with human cell material are real: contamination risk for the worker, cross-contamination, disposal of human biological material. Equally real is that an infrastructure monopoly follows from it which is not derived from the level of risk but has grown historically. The countermeasure is not rule circumvention but building shared laboratory infrastructure and working the approval-free substrate classes consistently. A5 is the weakest-covered point in the literature and the starting point of approach N1.

7 Guardrails for this strand

Four rules that follow from existing commitments and are applied here to specific approaches.

L1 — No military, no weapons, no defence. In this field the affected lines are primarily cyborg insects and biohybrid sensing: explosives detection with locust antennae, remote-controlled insects for reconnaissance, nano-tattoo steering. The exclusion does not cover the underlying sensory physiology or odour coding as such. It covers setups whose purpose is detection of ordnance, reconnaissance or targeting. The civilian parallel application — gas leak localisation, environmental monitoring — is permissible, but must be kept separate in purpose, publication form and funding source, otherwise the exclusion has no effect.

L2 — No animal material. The exclusion bites before the laboratory animal, namely in consumables. Affected: Matrigel and related matrices (murine sarcoma extract), foetal calf serum, animal-derived serum substitutes, animal antibodies in characterisation, insect preparations in biohybrid sensing, animal-derived skeletal and cardiac muscle cells in actuation. Approach N1 follows from this, and from that in turn that muscle-driven biohybridics is not pursued practically in this strand, only referentially. The same applies to remote platforms whose culture chain contains animal components; their use requires a written statement from the operator (see V6).

L3 — Precaution regarding the substrate. While the consciousness question is open, a precautionary regime applies that does not wait for an answer. Operationally: no stimulation regimes built on aversion and avoidance where alternatives exist; no combination of nociceptor differentiation with closed-loop feedback; a documented maturation ceiling per series; a termination protocol fixed in advance and published with the work; no adoption of sentience vocabulary without a measurand.

L4 — Provenance and consent. Cell lines only with documented consent that covers the computational purpose. Where consent does not cover it, this must be recorded in the work and not disposed of by reference to anonymisation.

8 Novel approaches

Five proposals that the surveyed literature does not occupy, or occupies only marginally.

N1 — An animal-free wetware stack

Observation. Organoid research is routinely presented as a replacement for animal experiments. The culture chain nevertheless still contains animal components. The standard matrix Matrigel is murine sarcoma extract; more than 14,000 publications rest on it, while the most common xeno-free whole-matrix alternative is cited in the low double digits [P] (Wolff et al., Adv. Sci. 2025, doi:10.1002/advs.202508734; count date and database not stated there).

Available building blocks.

Gap. For cerebral organoids operated as a computing substrate, no published end-to-end animal-free chain with an electrophysiological endpoint exists. What exists are individual comparisons for vascular, intestinal and tumour organoids. A preliminary search (09/2026) confirms the finding; the systematic search with a documented strategy is outstanding (verification item 31).

Proposal. A systematic comparison with an electrophysiological endpoint: identical cell line, four matrix conditions, endpoints firing rate, burst rate, network synchrony, maturation trajectory, batch spread. This approach is immediately publishable as a review and collaboration-ready as bench work, because it is useful to any organoid laboratory independently of the computing purpose. It is at the same time the condition under which wetware work in this strand is compliant with its own rules at all.

N2 — A substrate ethics gradient instead of a binary decision

The debate runs binary: organoids either can be conscious or cannot. What would be operationally useful is a gradient over measurable proxies that scales requirements rather than switching them. Candidate axes: cell count; fraction of differentiated neurons; presence of nociceptive cell types; network synchrony; maturation duration; presence of closed sensorimotor feedback; multi-regionality (assembloid). The gradient claims no measurement of consciousness. It ranks setups by precautionary need and makes requirements justifiable before a threshold can be established. Compatible with the PIS-12 scale logic from FORSCH-0333.

N3 — Spectro-temporal coding as a wetware benchmark

The wetware task with sensory grounding demonstrated so far is a tactile classification across eight electrodes. Audio is the obvious unoccupied neighbouring case and methodologically more favourable: the input signal is temporally structured by nature, feature extraction (filter bank, onset, modulation spectrum) is established, and hard digital baselines exist. Proposed benchmark in three stages — onset discrimination, timbre classification at constant pitch, rhythm pattern recognition — each against a reservoir baseline of equal readout complexity. The core of the evaluation is showing that discriminability arises in the substrate and not in the classifier: an identical classifier on shuffled data must fail.

N4 — A ρ/δ access index for biocomputing platforms

A reproducible index over five axes: monetary threshold, waiting time, jurisdiction, required biosafety level, data return rights. Applicable to any platform, publishable as a table with a methods section, and suited to lift asymmetry A3 from section 6 out of the anecdotal. The index measures not quality but the distribution of disposal.

N5 — Terminological correction B1/B2/B3

The three-level distinction in section 1 is itself a result. German-language technical and funding literature uses “bionics” for all three levels, which systematically misdirects calls, curricula and project plans: B1 competences are budgeted for B3 undertakings. A short position paper with a decision table can be run as a sub-paper to this work.

9 Connections

Cross-references within the FORSCHUNG strand. Register check of 2026-09-22; numbers provisional where assigned.

Application fields without strand reference.

The peripheral-manufacturing row is the most robust: the competence at hand lies in the peripherals, not the substrate. That is not a limitation but the point of greatest leverage, because peripherals in this field are consistently expensive, proprietary and poorly documented. This work makes no links to commercial applications.

10 Experimental programme

Eight experiments ordered by rising access barrier. Cost figures are orders of magnitude for a first build, excluding labour.

The descriptions state question, principle, measurand, controls and cost order of magnitude. They deliberately omit process parameters: drying temperatures and durations, layer thicknesses, deposition and patterning parameters, mixing ratios, suppliers and part numbers are not given here. This withholds nothing needed to assess the work — the finding and the argument stand complete in the free part, and refuting them needs no bill of materials. It separates finding from replication.

Tier 0 — approval-free, animal-free, workshop

V1 — Mycelium memristor bank. Question: How do memristive characteristics scatter across substrate, species, degree of drying and electrode material? Setup: Grain or hemp substrate, edible-mushroom culture, stainless-steel and PEDOT:PSS-coated ITO electrodes, defined drying stages, I–V curves over frequency and amplitude. Measurand: Area and symmetry of the pinched hysteresis loop, retention over cycles, sample-to-sample spread. Contribution: The published work shows feasibility on few samples. A scatter study with a standardised drying protocol and an open dataset is missing. Requirements: Function generator, oscilloscope or DAQ card, humidity chamber. EUR 500–1500. Law: no restriction. Ethics: unproblematic.

V2 — Physarum reservoir. Question: Does the network topology serve as a reservoir for a temporal classification task? Setup: Agar plate, oat flakes as attractors, electrode grid, readout of the resistance matrix over time, digital readout classifier. Measurand: Separability against a random baseline and against an echo-state network with equal readout size. Contribution: Direct evidence for the thesis of ≥0393. German-language literature effectively absent. Requirements: EUR 200–600. Law/ethics: unproblematic.

V3 — Open MEA hardware. Question: Can a transparent MEA be produced to sufficient quality with workshop means? Setup: ITO-coated glass, patterning, PEDOT:PSS deposition on contact pads, carrier PCB, connection to an Intan headstage, acquisition through the Open Ephys chain; validation initially on fungal and Physarum samples, not cell culture. Measurand: Per-electrode impedance, noise floor, crosstalk. Contribution: The open build instructions (transparent MEA; OpenMEA) are published but rarely rebuilt independently. A documented rebuild with a deviation report is a contribution in its own right and the precondition for everything further. Requirements: EUR 1500–4000 depending on headstage. Law/ethics: unproblematic.

V4 — Systematic comparison of animal-free matrices (desk tier). Question: Which xeno-free matrices are documented for neural 3D culture with an electrophysiological endpoint, and which are merely asserted? Setup: Full literature search, extraction grid (matrix, cell type, endpoint, comparator arm, batch count, funding, stakeholder interest), grading levels. Result: A publishable review plus a draft bench protocol for N1. Requirements: labour only. Law/ethics: unproblematic.

V5 — Actuation without animal cells. Question: What actuation performance is reachable with hygromorphic and mycelium-based composites relative to muscle actuators? Setup: Moisture-driven fibre composites and mycelium composites, 3D-printed anisotropic carriers, measurement of stroke, force, cycle count, time constant. Measurand: Work per cycle, power density, fatigue, comparison against published muscle values. Contribution: The comparison is not drawn in the biohybrid literature because the communities are separate. A negative result is equally publishable. Requirements: EUR 300–1000, mostly existing manufacturing.

Tier 1 — remote access, external infrastructure

V6 — Spectro-temporal coding on an organoid platform (N3). Question: Do organoids produce distinguishable response patterns to spectro-temporally coded audio stimuli, and does the separability reside in the substrate? Setup: Local feature extraction (filter bank, onset detection, modulation spectrum), mapping onto eight stimulation channels, recording through the platform API, offline evaluation. Three stages: onset, timbre at constant pitch, rhythm patterns. Controls: identical classifier on temporally permuted stimuli; identical classifier on blank-channel recordings; reservoir baseline of equal readout complexity. Measurand: Classification accuracy against all three controls, discriminability per stage. Contribution: An unoccupied modality, drawing on existing audio competence, directly compatible with the published Braille protocol. Access: apply for a free slot through the open call, otherwise subscription. The platform’s ethics status is to be documented in advance, not presupposed. Requirements: labour plus any platform fee. L2 reservation: The platform’s culture chain (matrix, medium, supplements) must be queried in writing before applying. If it contains animal components, V6 cannot be run under guardrail L2; the N3 benchmark then runs on the animal-free reservoirs from V1 and V2.

Tier 2 — collaboration-dependent

V7 — DNA coding with physical endpoint control. Question: How does a self-designed coding scheme with error correction behave under real synthesis and real readout? Setup: Local encoder/decoder, short oligos ordered commercially, readout via a sequencing service or a nanopore collaboration. Measurand: Raw per-base error rate, recovery rate after correction, cost per kilobyte. Requirements: EUR 300–2000 depending on length and readout.

V8 — Wetware with own culture. Only after V1–V6 and only under three conditions: a documented laboratory collaboration at the appropriate biosafety level, a matrix chain per N1, a precautionary protocol per L3. Without all three the step is permissible neither legally nor under the strand’s own rules.

Recommended order: start V4 and V1 in parallel. V4 is pure desk work with an immediate publication target; V1 produces the first own measurement series and generates the hardware need that V3 then covers. V6 is applied for only once V3 shows the evaluation chain works and the platform operator has confirmed that the culture chain is animal-free; otherwise the N3 benchmark runs on the reservoirs from V1 and V2.

11 Open questions

1. Does a learning rule exist for dissociated and organoid networks that does not reduce to a reservoir plus digital readout classifier? [O]
2. What is the actual energy demand of a wetware system including life support, per task solved? Reliable figures are missing. [O]
3. Does the multi-year developmental clock (Nature, 08/2026) shift the precautionary threshold, and if so, measured against what? [O]
4. Are xeno-free matrices electrophysiologically equivalent for cerebral organoids? Documented for vascular organoids, open here. [O]
5. At what electrode density does the readout side move from sampling to state estimation? [O]
6. Can radiation resistance of dehydrated fungal memristors be confirmed under real high-altitude exposure? [O]
7. Which Austrian and Union provisions apply specifically to human iPSC culture without genetic modification, and where does the analogy to tissue-safety law end? — preliminary search 09/2026 (not legal advice): for culture without genetic modification in a non-clinical laboratory, worker protection (ASchG § 40(5), Ordinance on Biological Agents) and data protection (GDPR Art. 9) are in the foreground; by their wording the Tissue Safety Act and the Genetic Engineering Act do not apply, the latter however does with any genetic modification. Risk group, derivation phase and consent remain open in the individual case; to be confirmed by a qualified expert before any wetware work. [O]
8. Is the culture chain of the available remote organoid platforms (matrix, medium, supplements) animal-free? Public operator information does not name matrix, serum or supplements [S]. Without a written statement from the operator, V6 cannot be run under guardrail L2. [O]

12 Sources

Selection of load-bearing references. Full verification: see the verification list in werk/.

Not cited: vendor blogs as evidence for performance claims, news aggregators, social media posts. Where vendor figures appear in the text they are marked [S] and identified as statements by a stakeholder.

13 Glossary