OpenKnowledge - OpenLaw
Vision Document for the OpenKnowledge's OpenLaw Project
By Seun Akinwunmi ·
Intelligence Infrastructure for an AI-Powered World
AI is becoming one of the most important general-purpose technologies in the world.
The cost of accessing intelligence is falling rapidly. Models are becoming more capable, inference is becoming cheaper, and capabilities that once required teams of highly specialized people can increasingly be accessed through software.
I have been thinking about the best way to leverage AI to accelerate development in developing countries, and I have come to realize that the general intelligence of frontier models still requires local context to become genuinely useful.
A model can be incredibly intelligent and still struggle to solve the problems of a particular society if it does not understand how that society works.
There is also a real risk that developing countries could become locked out of some of the benefits of AI because of geopolitical, economic, infrastructure, or commercial constraints. The models we depend on may be developed elsewhere, the companies controlling them may change their policies, access may become more expensive, or geopolitical circumstances may make certain systems unavailable.
So I have been thinking about what infrastructure we can build ourselves.
Build the Knowledge Infrastructure
My goal is to use AI to dramatically lower the cost of accessing useful knowledge, education, and institutional capability in developing countries, while reducing the risk of being locked out of the benefits of AI.
I believe this requires building something that general AI companies have little incentive to build themselves:
Local knowledge infrastructure.
A general AI company has an enormous incentive to make its models broadly intelligent and useful to billions of people.
But there is a limit to how deeply a general-purpose model can be expected to understand every society in the world.
It has much less incentive to deeply structure every government procedure, obscure regulation, ministry circular, state-specific rule, local institution, educational pathway, public service, or local process in every developing country.
That work is highly contextual and often expensive to maintain.
Yet this is precisely the information people and AI agents need in order to operate effectively within a society.
This is what I mean by knowledge infrastructure.
It means finding, digitizing, structuring, connecting, verifying, contextualizing, and maintaining the knowledge that describes how a society works — and building systems that allow both people and intelligent machines to use it.
OpenKnowledge is the broader idea I am exploring.
The initial implementation is OpenLaw, focused on Nigerian law.
I am deliberately starting with law because it provides a relatively well-defined domain where authoritative knowledge, jurisdiction, interpretation, provenance, and context matter enormously. It is also a domain where the consequences of getting information wrong are significant, making it a useful environment for exploring reliable knowledge retrieval and reasoning.
OpenLaw is not intended to be simply another legal chatbot.
The initial goal is to build a system that can make Nigerian legal knowledge easier for people to understand and easier for intelligent systems to use.
OpenLaw can do the expensive work once — finding, structuring, connecting, verifying, and explaining this knowledge — and make the resulting capability cheap and reusable for everyone.
Instead of thousands of people independently spending two hours researching the same problem, the work can be done once and accessed in seconds.
We need to build the knowledge, distribution, localization, and product infrastructure that becomes increasingly valuable as intelligence gets cheaper.
Why Build This Layer?
There is another reason this matters.
Nigeria is not currently at the frontier of AI model development.
The world's most capable commercial models are developed primarily in the United States and China, and the infrastructure required to train frontier models — enormous datasets, specialized hardware, capital, research organizations, and massive amounts of compute — is concentrated in a relatively small number of companies and countries.
We should absolutely participate in AI research and model development where we can.
But we shouldn't assume that our path into an AI-powered future requires us to first build the next frontier foundation model.
There is another layer that is much more accessible to us.
Knowledge infrastructure.
Building a comprehensive, machine-readable, contextual understanding of Nigerian law, institutions, procedures, education, businesses, public services, and other domains is difficult — but it is vastly more accessible than building frontier-scale compute infrastructure.
And it is something we can actually own.
This is particularly important because knowledge infrastructure can sit above the model layer.
That infrastructure can be used by increasingly capable models, whether those models are commercial, open-source, locally deployed, or eventually developed here.
The model can change.
The knowledge layer can remain.
We should be building that layer now, while the cost of doing so is relatively small compared with building the physical and computational infrastructure required for frontier AI.
From Models to Agents
This becomes increasingly important as AI moves from answering questions to performing tasks.
Imagine that, several years from now, we want to build an Agentic OS for Nigerian government workers.
An AI system that can understand government processes, retrieve the relevant rules and documents, interact with internal systems, coordinate tasks, and help civil servants do their jobs.
Would we want that entire system to depend on a commercial model API?
Maybe.
Commercial models may be the best option for many applications.
But we should also have the option of taking increasingly capable open models and building the Nigerian-specific infrastructure around them ourselves.
The model provides the intelligence.
The knowledge infrastructure provides the context.
The tools and workflows provide the ability to act.
That separation could allow us to build systems that remain useful even as the underlying models change.
The same principle applies beyond government.
An agent helping a Nigerian citizen might need to know:
Is this legal?
Which institution is responsible?
What documents are required?
What process should I follow?
What are my rights?
Can this process be completed on my behalf?
A sufficiently capable agent should eventually be able to use a knowledge infrastructure to answer these questions and, where appropriate, act on the user's behalf.
This is why I don't want to think about OpenLaw simply as a legal chatbot.
It could eventually become an Agentic API for Nigerian law — a service that allows other AI systems to retrieve authoritative legal information, understand applicable rules and context, and reason about the legality of actions.
And if the knowledge layer expands beyond law into government institutions and public processes, it could eventually enable agents to help people navigate and perform parts of the bureaucracy that currently require significant time, money, connections, or specialized knowledge.
An agent could potentially move from:
"What are my rights?"
to:
"What should I do?"
to:
"Can you help me do it?"
That is where knowledge infrastructure begins to become capability infrastructure.
Knowledge Before Capability
There is an important asymmetry here.
Physical infrastructure and frontier compute require enormous amounts of capital, hardware, energy, and specialized expertise.
Knowledge infrastructure has a different set of economics.
It still requires substantial work — collecting information, digitizing it, structuring it, verifying it, connecting it, keeping it current, and building systems that can reliably use it.
But once that work is done, it can potentially be reused millions of times.
A government procedure documented and structured once can help thousands of citizens.
A legal rule correctly connected to its jurisdiction and exceptions can be retrieved by thousands of people and potentially thousands of agents.
An educational pathway can be made accessible to anyone who needs it.
This creates a compounding effect.
Do the expensive knowledge work once. Make the resulting capability cheap to access.
Knowledge infrastructure is not a replacement for physical infrastructure, compute, or AI research.
It is a layer we can begin building now while we continue to develop the others.
We should be preparing this infrastructure for a world where AI agents become increasingly capable.
This Is an Experiment
I don't know exactly what OpenLaw will become.
That's important.
This is an exploratory project for me.
I'm building it because I want to investigate what becomes possible when increasingly capable AI is combined with a deep, structured understanding of a particular society.
I don't have the final architecture figured out.
I don't have the final product figured out.
I don't even know whether the larger OpenKnowledge idea will evolve into what I currently imagine.
The goal right now is to build, experiment, and learn in public, and hopefully over time make the idea more concrete.
Maybe the project evolves into a much larger OpenKnowledge platform.
Maybe it becomes infrastructure that other developers and organizations build on.
Maybe it becomes an Agentic API for Nigerian law.
Maybe it eventually becomes part of the infrastructure for AI agents operating across Nigerian government and society.
Maybe it becomes something I haven't imagined yet.
And maybe some of these ideas turn out not to work.
That's okay.
The project does not need to have its final form today.
I want to discover what is useful by actually building it.
And I hope that, as the idea becomes more concrete, other people will find the problem interesting enough to join me.
The Bigger Vision
The ambition extends beyond law and beyond Nigeria.
We aim to democratize the benefits of AI for education, knowledge, and human capability among citizens of developing countries around the world, starting with Nigeria.
There are billions of people living in societies where useful knowledge exists but is difficult to access.
People don't know what their rights are.
They don't know what opportunities are available to them.
They don't know how institutions work.
They don't know what is possible.
They may have the ability to do something but lack the knowledge required to discover that possibility.
That represents enormous wasted human potential.
AI gives us an opportunity to change the economics of access to knowledge.
But to take advantage of that opportunity, we need more than intelligent models.
We need to build the knowledge infrastructure that allows those models to understand and operate within our societies.
Knowledge infrastructure is one of the layers of the AI stack that developing countries can begin building today.
We should not wait until AI becomes ubiquitous and then discover that the infrastructure around it was built entirely elsewhere.
We can start building the layer that we can realistically build today, while continuing to participate in model research, compute, and the other parts of the AI ecosystem.
Nigeria can become not only a consumer of AI, but a place where AI is given the knowledge and context required to become genuinely useful to Nigerian people and institutions.
And perhaps the same architecture can eventually be adapted to other developing countries.
The long-term goal is not to build the world's smartest model.
It is to make increasingly powerful intelligence useful, accessible, and locally relevant.
To reduce the cost of knowledge.
To reduce the cost of education.
To reduce the cost of navigating institutions.
And ultimately, to reduce the cost of capability itself.
What becomes possible when access to useful knowledge is no longer determined primarily by where you live, what you earn, or who you know?
Resources
Github Repo: https://github.com/204070/openlaw-api Youtube: OpenKnowledge Vision Stream - YouTube
Tags: tech, governance