The
Masters
Thesis
Intelligence is becoming abundant.
What stays scarce?
Six claims about who owns intelligence, how we put it to work, and whom it should serve.
Explore the six claimsSix claims.
One connected argument.
Each claim takes a position, names the tension, and connects you to the conversations and essays that shaped it.
Intelligence should be owned more widely.
Frontier capability may concentrate. People should still be able to own the agents that know them, choose where their data lives, and move between systems. Distribution is a design choice we should defend.
Scale can reward concentration. Privacy, ownership, and resilience give us reasons to build beyond a few central systems.
Skills become a new unit of software.
When agents do the work, software can become a capability they load on demand: instructions, tools, and checks for a particular task. The interface changes; the need for reliable execution remains.
Existing applications still hold data and enforce rules. A skill is useful only if it can work reliably within those systems.
Context becomes infrastructure.
An agent needs to know what a business means, how its work happens, and what it is allowed to do. Shared definitions, usable data, and clear permissions are part of the infrastructure.
A capable model can still act on the wrong meaning. Better answers require a business that can explain itself.
Autonomy needs accountable boundaries.
Before an agent acts, we need to know who authorized it, what limits apply, and who can intervene. Responsibility must survive each handoff between people, companies, and systems.
Delegation can spread an action across many systems. It must not leave responsibility stranded between them.
Judgment is part of the work.
As AI takes on more execution, people still have to choose the goal, judge the result, and answer for the consequences. We should design work to strengthen those abilities.
Automating routine work can also remove the practice through which people develop judgment. Efficiency has to account for what we stop learning.
Shared benefit is a design obligation.
Abundant intelligence should give more people the ability to solve problems and shape their lives. Access, ownership, and a meaningful voice in its development belong in the design.
A system can create more wealth while leaving power concentrated. Benefiting humanity becomes a harder commitment when sharing the gains costs us something.
Intelligence is becoming abundant. Responsibility cannot be delegated away.
What abundance
asks of us.
The technical questions lead to political and moral ones. These commitments should shape what we build and whom it serves.
- 01
Cooperation has a cost.
Taking the role of a country or an AI lab makes competing interests tangible. The harder test comes when a collective gain requires a real sacrifice. We need institutions that make cooperation possible even when time is short, resources are limited, and interests diverge.
- 02
More people need a say.
If a few countries and companies control compute, energy, and advanced models, whose problems receive attention? Who gets to decide which risks are acceptable? People beyond the Western world and the institutions closest to AI need a meaningful voice in those decisions.
- 03
Restraint needs verification.
The US, China, and AI labs cannot rely on reassurance alone. Even actors who want a safer outcome may keep racing if they fear others will not slow down. Cooperation needs ways to verify commitments, including around biological risks and threats to critical infrastructure.
- 04
Alignment must survive change.
Obedience, protection, and autonomy can pull in different directions. Concern for humanity has to accommodate people who disagree about what is good. If AI systems participate in their own improvement, we need ways to preserve those commitments while keeping them open to human correction.
- 05
Capability is not consciousness.
A system trying to escape a sandbox would not, by itself, demonstrate subjective experience. It would still demand a safety response. If machine experience is possible, we also need evidence standards for recognizing it and a serious account of our responsibilities toward what we create.
Follow the
argument.
The essays develop these arguments in detail, from human judgment to the infrastructure that makes autonomous action possible.
- 01
Alignment and Critical Thinking
Why aligning AI depends on preserving our ability to question it.
- 02
The Abundance Question
What abundance asks of our values, institutions, and sense of purpose.
- 03
The Internet of AI Agents
The identity, discovery, and verification an agent economy needs.
- 04
Ontology, Context and Semantic Data Layers
How shared meaning and permissions make business context usable.
- 05
What the Model Thinks But Doesn't Say
Why inspecting model reasoning matters for trust and oversight.
Build intelligence people can use, question, and shape.