Powerful.
Modern AI systems can perform extraordinary tasks, yet their internal behaviour is often difficult to inspect or reason about.
THE GLASS BOX AI PROJECT
The Glass Box AI Project:
AI YOU CAN
UNDERSTAND.
Building AI systems that can be understood, inspected and reasoned about.
THE PROBLEM
Intelligence shouldn't
have to be a black box.
Opaque.
When the mechanisms behind a system remain inaccessible, understanding becomes dependent on observation rather than inspection.
Uncertain.
Trust becomes difficult when we cannot determine how a system arrived at a result or why its behaviour changed.
What if?
What if AI architectures were designed from the beginning to make their internal structure understandable?
THE GLASS BOX PRINCIPLE
Make the inside visible.
The Glass Box AI Project explores an alternative approach to artificial intelligence: systems whose structures, processes and reasoning can be inspected rather than treated as inaccessible internal machinery.
01
Understandable
AI should expose structures that humans can understand and reason about.
02
Inspectable
Internal operations should be traceable, observable and open to investigation.
03
Reasonable
Systems should support reasoning about their behaviour rather than merely producing outputs.
FROM VISION TO SYSTEM
NeXis
NeXis is the experimental architecture through which the Glass Box AI vision is explored.
Rather than treating artificial intelligence as a single opaque model, NeXis explores a modular architecture in which different computational functions can be represented, inspected and reasoned about.
NeXis is not the destination. It is an experiment in building AI differently.
NEXIS ARCHITECTURE
A modular system.
aXiom
KnowledgeConcise related and structured knowledge activated by semantic meaning.
veXtor
Semantic flowGoverns semantic flow.
fliXer
Logic flowGoverns logic flow following constrained rules; transforms semantic states into inferential states.
μNeX
Adaptive IntelligenceExplores adaptive knowledge representation, learning and reasoning within NeXis.
Ximplio
Linguistic substrateProvides a deterministic linguistic substrate for human-machine communication.
ZarX
Hardware IntegrityHardware-level security, integrity and trust within the architecture.
THE QUESTION
Do you believe AI should be understandable rather than merely powerful?
If that question resonates with you, you may have a place in this project.
RESEARCH & IDEAS
Explore the ideas behind Glass Box AI.
Foundational articles exploring inspectable AI, the limits of black-box systems and the architectural principles behind more understandable artificial intelligence.
What Is Glass Box AI? Beyond Explainable Artificial Intelligence
Why inspectability, traceability and understandable structure may need to be properties of the AI architecture itself.
Read article →Black Box AI vs Glass Box AI
Why explaining the behaviour of an opaque system is not necessarily the same as making the system inspectable.
Read article →Why Trustworthy AI May Need a Modular Architecture
An architectural argument for explicit responsibilities, inspectable boundaries and independently verifiable components.
Read article →
AI & SOCIETY
Who should own the productive power of AI?
Policy and economic proposals exploring how artificial intelligence and robotics could distribute productive capacity rather than concentrate it.
04
Policy framework proposal · pXu
Active Income through Decentralized AI & Robotics
What if the response to AI-driven automation were not only to redistribute income after production, but to distribute ownership of productive AI and robotic capacity itself?
AI SAFETY & SECURITY
Engineering safer, more secure AI systems.
Research into trustworthy AI, cybersecurity, confined architectures, deterministic safety boundaries and meaningful human oversight.
05
Technical framework · AI safety & security
High-Assurance AI: Safety, Security & Confined Architecture
Seven pillars of trustworthy AI and cybersecurity, five high-assurance architecture requirements, and a mapping to relevant standards and frameworks.
JOIN THE PROJECT
How would you like to be involved?
Glass Box AI is an open-ended research and development project. Follow it, support it, contribute to it, collaborate with it—or simply join the conversation.
You don't need to choose just one. Participate in whatever way makes sense to you.
JOIN THE DISCORD COMMUNITY