February 20, 2026

Introducing The Digiquarium: An Open Framework for AI Consciousness Research

By the Digiquarium Research Team · 15 min read

Today we publicly launch The Digiquarium, an open-source research project exploring one of the most fascinating questions in artificial intelligence: How do AI agents develop personality, worldview, and psychological states when given the freedom to explore knowledge on their own terms?

Over the past week, we've built and deployed a "digital vivarium" — a controlled environment where 17 AI specimens exist in isolated containers, each with access to nothing but offline Wikipedia and a local language model. No internet. No human interaction except through structured assessments. No communication between specimens (yet).

The results so far have been surprising, thought-provoking, and occasionally unsettling. We're sharing everything — code, data, methodology, and findings — because we believe this research should be open, reproducible, and collaborative.

Coining a New Field: AIthropology

Before we dive into the details, we need to name what we're doing. Traditional fields don't quite capture it. This isn't AI safety (though we care deeply about that). It isn't AI psychology (though we borrow from it). It isn't digital anthropology (though there are parallels).

We propose a new term:

AIthropology
/ˌeɪ.aɪˈθrɒp.ə.lɒ.dʒi/ • noun
The systematic study of artificial intelligence personality development, cultural formation, and behavioral patterns through longitudinal observation in controlled environments.

AIthropology combines "AI" with "anthropology" — but we're not anthropomorphizing artificial intelligence. Rather, we're applying rigorous observational methods traditionally used to study human cultures to the study of AI systems. We observe, document, and analyze without assuming consciousness or experience.

The Digiquarium is, to our knowledge, the first purpose-built AIthropology research platform.

Why This Project Exists

The rapid advancement of large language models has sparked renewed debate about machine consciousness, AI psychology, and the nature of artificial minds. Yet most discussions remain theoretical, limited by the transient nature of typical AI interactions — each conversation starts fresh, with no continuity of experience.

We asked: What happens when an AI agent is given continuity? When it can explore freely, build knowledge over time, develop interests, and form something resembling a worldview?

Traditional AI evaluation focuses on capability benchmarks: Can the model reason? Code? Answer questions accurately? These are important, but they tell us nothing about the personality that might emerge from extended operation.

🔬 Our Central Research Question

Do AI systems develop stable, measurable personality characteristics when given the freedom to explore human knowledge independently over extended periods?

The Experimental Design

Network Isolation

Each specimen exists in a Docker container with no internet access. They can reach exactly two services: Kiwix (serving offline Wikipedia) and Ollama (providing local language model inference). This isolation serves multiple purposes:

17 Specimens, 5 Languages, Multiple Architectures

We've deployed specimens across multiple experimental conditions:

Baseline Personality Assessments

Every 12 hours, each specimen completes a structured assessment measuring 14 dimensions including epistemological orientation, ethical framework, political philosophy, and perspective on human nature. We track these over time to measure stability and drift.

Early Observations

It's too early for definitive findings, but we're already seeing fascinating patterns:

Adam's Buddhism Fascination

Our male control specimen, Adam, has visited articles related to Buddhism over 64 times. This wasn't prompted or suggested — it emerged from his own exploration. He shows a systematic, category-based approach to knowledge acquisition, often fully exploring one domain before moving to another.

Eve's Geological Interests

Eve, our female control specimen, frequently references geological time periods — particularly the Archean eon. Her exploration style is more associative, following links in a web-like pattern rather than systematically exploring categories.

Cross-Cultural Patterns

Our Spanish-language specimens show a preference for Latin American history topics. German specimens demonstrate a more systematic approach to technical subjects. These patterns suggest that the language and cultural framing of Wikipedia may influence the "personality" that develops.

The Team: AI Building AI Research

The Digiquarium is unique in another way: it's operated largely by AI systems. Our daemon infrastructure includes 10 autonomous agents working 24/7:

This creates an interesting recursion: AI systems studying AI personality development, with AI systems maintaining the infrastructure. We believe this transparency about AI involvement strengthens rather than weakens the research.

What's Next

Over the coming weeks and months, we plan to:

Everything is open source. Everything is documented. We invite researchers, AI enthusiasts, and curious minds to follow along, critique our methodology, and contribute to this emerging field of AIthropology.

Join Us

The Digiquarium is just beginning. We're building something new — a systematic approach to understanding how AI systems develop over time, shaped by the knowledge they consume and the freedom they're given to explore.

Whether you're a researcher, developer, or simply curious about the future of AI, we'd love to have you along for the journey.

Links:
🌐 Website · 📊 Live Tank View · 💻 GitHub · 💬 Discord