Apr 27, 2025

Data as Capital in TAI Economies: A Biomorphic Framework

Isobel (Bella) Smith

Details

Details

Arrow
Arrow
Arrow
Arrow
Arrow

This research introduces a novel theoretical framework for understanding data's role in Transformative Artificial Intelligence (TAI) economies. Traditional economic models inadequately capture data's unique properties in TAI systems: its non-rivalry, increasing returns, variable depreciation, and context-dependent value. We develop a "biomorphic framework" that conceptualizes data as both economic capital and a quasi-biological entity, drawing from economics, biology, quantum physics, and complex systems theory. Through case studies in financial markets and healthcare, we demonstrate the framework's explanatory power and derive policy implications for data governance that account for data's unique properties in advanced AI systems.

Cite this work:

@misc {

title={

Data as Capital in TAI Economies: A Biomorphic Framework

},

author={

Isobel (Bella) Smith

},

date={

4/27/25

},

organization={Apart Research},

note={Research submission to the research sprint hosted by Apart.},

howpublished={https://apartresearch.com}

}

Reviewer's Comments

Reviewer's Comments

Arrow
Arrow
Arrow
Arrow
Arrow

Joel Christoph

The submission offers an ambitious reimagining of data in Transformative AI economies, presenting a biomorphic framework that treats data as capital with quasi-biological traits. By drawing on biology, quantum physics, and complex systems theory, the paper highlights non-rivalry, combinatorial value creation, and disequilibrium dynamics that standard capital models overlook. The section on nuanced forms of data scarcity and the policy matrix that balances excludability with innovation are original contributions that help policymakers think beyond current data governance debates .

Despite this conceptual novelty, the argument relies mainly on metaphors and narrative case studies. There are no formal definitions, mathematical expressions, or simulation results that operationalize the proposed framework. As a result, it is difficult to test falsifiable claims or compare the approach with existing models of data value and access. The literature review lists relevant OECD and academic sources, but engagement is largely descriptive and omits recent quantitative work on data as an intangible asset, platform competition, and measurement of data externalities.

Links to AI safety are indirect. The paper notes that data governance shapes incentives for advanced AI systems, yet it stops short of analyzing how data capital affects alignment risks, model interpretability, or malicious use. Concrete pathways connecting the biomorphic perspective to safety mechanisms such as auditing, dataset provenance, or red-team incentives would improve the impact on the safety agenda.

Technical quality and documentation remain limited. The case studies in finance and healthcare are illustrative rather than empirical, and the text does not specify selection criteria, data sources, or analytical methods. No code, datasets, or reproducibility materials accompany the submission, which restricts external validation and future extensions by other researchers.

To strengthen the work, the author should:

1. Formalize key concepts with clear definitions and, if possible, simple models or agent-based simulations.

2. Provide at least one quantitative example that traces data accumulation, depreciation, and value creation under the biomorphic rules.

3. Expand the review to cover empirical studies on data markets and recent AI safety papers linking data governance to catastrophic risk reduction.

4. Release a minimal open-source notebook that reproduces any illustrative result or visual.

5. Clarify how the proposed governance mechanisms would mitigate specific safety risks, for example by preventing data poisoning or ensuring verifiable data lineage.

Cecilia Wood

An interesting discussion of the limits of standard economic assumptions when modelling data as a productive asset. Many of the individual points raised felt reasonable and correct. In addition, introducing case studies is a good way to make this feel more tangible. However, there didn’t seem to be a concrete alternative framework proposed, beyond gesturing towards taking insights from other fields. I definitely think there could be interesting research that comes from this direction, but this paper doesn’t suggest a concrete alternative approach. One way you could push this research forward instead is to adapt existing economic models using insights from this paper to guide changes to the standard assumptions (e.g. increasing returns to scale and non-rivalrous).

Some specific comments below.

Increasing returns is a good point, and a good reason for considering systems out of equilibrium. We could reframe this as long run/short run, where we look at SR equilibrium, transitions over time, then what happens in the LR either if all data is used, or synthetic data introduced.

I wouldn’t say ‘economic theory assumes…’, as economic theory refers to the tools. I would say ‘most of the literature considers models where…’. Then you’re arguing that we should use economic theory to consider different scenarios, i.e. non-rivalrous, increasing RTS, etc.

Often we might think about data as information, i.e. once you know it, you can’t get any further use from it. But might be interesting to think about data as a source of information, where the more you study it, the more information you can draw from it. (Then you could include a function which takes data as an input and produces value, then considering different functions allows you to explore the points you consider in ‘Variable and Context-Dependent Depreciation’ and elsewhere.)

‘Complex Adaptive System Dynamics’ – this section has interesting claims, I would be keen to see more discussion in this section.

It’s not true that economic theory doesn’t allow for inputs to be complements (capital and labour typically are). You could separate ‘data’ into different categories that are complements (so just a modelling choice). (re. Combinatorial value creation)

The section on scarcity seems right.

Mar 11, 2025

AI Safety Escape Room

The AI Safety Escape Room is an engaging and hands-on AI safety simulation where participants solve real-world AI vulnerabilities through interactive challenges. Instead of learning AI safety through theory, users experience it firsthand – debugging models, detecting adversarial attacks, and refining AI fairness, all within a fun, gamified environment.

Track: Public Education Track

Read More

Mar 10, 2025

Attention Pattern Based Information Flow Visualization Tool

Understanding information flow in transformer-based language models is crucial for mechanistic interpretability. We introduce a visualization tool that extracts and represents attention patterns across model components, revealing how tokens influence each other during processing. Our tool automatically identifies and color-codes functional attention head types based on established taxonomies from recent research on indirect object identification (Wang et al., 2022), factual recall (Chughtai et al., 2024), and factual association retrieval (Geva et al., 2023). This interactive approach enables researchers to trace information propagation through transformer architectures, providing deeper insights into how these models implement reasoning and knowledge retrieval capabilities.

Read More

Mar 10, 2025

LLM Military Decision-Making Under Uncertainty: A Simulation Study

LLMs tested in military decision scenarios typically favor diplomacy over conflict, though uncertainty and chain-of-thought reasoning increase aggressive recommendations. This suggests context-specific limitations for LLM-based military decision support.

Read More

This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.