Does a wetware computer contain living material?

In the sense used here, yes; biological activity contributes to information processing.

Is a simulated neural network wetware?

No. A simulation running on ordinary electronics does not contain a living computational component.

Does task adaptation prove consciousness?

No. Changes in measurable behavior do not settle questions about subjective experience.

AI Glossary / AI and machine learning / Advanced

Wetware computer

Definition

A wetware computer uses living biological material as part of its information-processing system. In AI-related research, the term often refers to experimental systems that connect cultured neurons or other biological components to electronic interfaces for receiving inputs and producing measurable outputs.

What it means

Conventional software can simulate neurons while running entirely on electronic hardware. A wetware system instead includes living material whose physical activity contributes to the computation. Hardware, software and biology may all be necessary parts of the complete setup.

The term is broad and does not identify one architecture or a standard commercial product. A demonstration of adaptive activity in a laboratory is different from a reliable general-purpose computer. Descriptions should specify the biological preparation, electronic interface, task and role of ordinary software rather than implying that a dish of cells independently performs an entire application.

How it works

In a neuron-based experiment, an electrode interface can stimulate a culture and record electrical activity. Software maps task information into stimulation patterns and maps recorded activity into an output, such as a simulated movement or classification signal.

A feedback loop may change stimulation according to the outcome. Researchers examine whether activity changes in a task-relevant way, comparing results with controls and alternative explanations. Other experimental designs use biological dynamics as a reservoir, with a separate readout trained to interpret responses.

Living components need suitable environmental support and vary across preparations and over time. Interface design, signal processing and evaluation therefore matter alongside the cells themselves. An observed change in performance does not by itself establish consciousness, human-like understanding or a scalable computing advantage.

Three practical examples

Basic neuroscience: A laboratory studies how a cultured network responds to structured feedback in a controlled simulated task.
Computing research: Investigators explore whether biological dynamics can provide useful transformations for a separately trained readout, comparing them with electronic baselines.
Experimental platform development: Engineers design recording, stimulation and monitoring systems that help researchers reproduce measurements across biological preparations. This supporting infrastructure is part of the overall computing system.

Think of it this way

Think of a laboratory instrument that incorporates a living component rather than only manufactured parts. The component responds dynamically, while electronics deliver signals and interpret measurements. The analogy should not be stretched into calling the preparation a miniature person or assuming that biological material automatically supplies a complete brain.

Worked example

Imagine an invented experiment with two possible inputs, A and B. Each input is encoded as a different stimulation pattern. Researchers record activity from a fixed set of electrodes and train a simple readout to distinguish the input classes.

On forty held-out trials, the readout identifies thirty correctly, giving 75% accuracy. That result alone does not prove learning occurred in the biological network: the readout may be exploiting a stable response difference or an artifact in the stimulation equipment.

The study therefore needs controls, repeated preparations and a clear account of which parameters changed. The example explains experimental attribution; it is not a report of a published wetware result or a recipe for handling biological material.

Illustrative case study

Fictional case: BioCircuit Research compares a living-network interface with a simulated reservoir on a small signal-classification task. Its first presentation highlights only the energy used by the recording electronics.

Reviewers ask the team to include environmental support, interface equipment and ordinary computation in the comparison. The revised report also shows variation between preparations and the effort required to keep measurements stable. This invented case illustrates fair system-level comparison without claiming that either approach is superior.

Common misconceptions and mistakes

Wetware is not the same as a conventional artificial neural network, and neuromorphic electronic hardware need not contain living cells. A biological element also does not eliminate the need for software, sensors or conventional processors.

Avoid treating provocative wording about sentience in a paper title as an established consensus about subjective experience. Task adaptation and consciousness are different claims. Do not infer a production-ready technology from a limited laboratory demonstration, or compare biological and electronic energy use while excluding different parts of their supporting systems.

Business use

For most businesses, wetware computing is a research topic requiring specialized scientific and engineering expertise. Potential relevance may lie in experimental platforms, interfaces and basic studies of biological information processing rather than immediate replacement of ordinary servers.

Evaluate claims against the exact task, controls, reproducibility, lifetime, total support requirements and comparison baseline. Any project involving living material also needs appropriate institutional governance and expertise. The useful question is what the demonstrated system can reliably do, not whether the word biological makes it intrinsically more intelligent or efficient.

Check your understanding

A biological computing demo uses a trained software readout. Can all observed classification ability automatically be attributed to learning in the cells?

Answer: No. The contributions of the readout, interface, biological dynamics and possible artifacts must be investigated separately.

Related terms

Reservoir computing
Biological computing
cultured neuronal network
brain organoid
neural interface
neuromorphic computing
biohybrid system
reproducibility

Learn next

Read one experimental paper and draw the full system: biological component, stimulation, recording, software and environment. Mark which parts learn and which are fixed. Identify the controls and held-out evaluations. Compare the paper's actual evidence with any broader claims made in a headline or product description.

Sources

Kagan and colleagues, In vitro neurons learn and exhibit sentience when embodied in a simulated game-world: https://pubmed.ncbi.nlm.nih.gov/36228614/ (primary abstract and Figure 1 caption consulted for electrode interfacing and feedback configurations). Cai and colleagues, Brain organoid reservoir computing for artificial intelligence: https://www.nature.com/articles/s41928-023-01069-w (publisher abstract consulted; full text not accessed). Direct Kagan DOI and alternative PDF access failed. These sources describe research demonstrations; the first title reflects the authors’ interpretation, not established proof of consciousness.

Previous: Western Electric rules Next: Wh-pronoun